|
Introduction
"Data democratization" has become a buzzword for a reason. Modern organizations rely extensively on data to make informed decisions about their customers, products, strategy, and to assess the health of the business. But even with an abundance of data, if your business can’t access or leverage this data to make decisions, it’s not useful. To that end, data democratization, or the process of making data accessible to everyone, is quintessential to data-driven organizations. Providing data access to everyone also implies that there are few if any roadblocks or gatekeepers who control this access. When stakeholders from different departments—like sales, marketing, operations, and finance—are permitted and incentivized to use this data to better understand and improve their business function, the whole organization benefits. Successful data democratization requires constant effort and discipline. It’s founded on an organization-wide cultural shift that embraces a data-first approach and empowers every stakeholder to comfortably use data and make better data-driven decisions. As Transform co-founder James Mayfield put it, organizations should think about "democratizing insights, not data." In this article, I will provide a detailed overview of data democratization, why organizations should invest in it, and how to actually implement it in practice. Why democratize access to data? Historically, data used to be kept in silos, usually under the purview of the IT or Analytics departments. When any stakeholder from outside these departments required data for their work, they had to go through these data gatekeepers to access the necessary assets. This philosophy has been the norm for decades but is no longer relevant for modern data-driven organizations. Removing these types of bottlenecks is a necessary first step toward data democratization. Guidelines for data democratization can be noted in a data governance framework to improve access and provide high-quality data for downstream analytics. Improving access is just the first step of an ongoing process where every individual employee is encouraged and trained to make use of data. The more people who can make decisions based on data, the more the organization stands to benefit from a variety of perspectives and ideas. Companies have been dedicating huge investments in data infrastructure and tooling in order to build an analytics advantage over their competitors. The dream is to “democratize data” and get employees to change their ways of working and start making decisions informed by data, not gut feelings. By investing in data education and helping analysts influence, then building modern tools to support metrics, we will continue making progress toward that goal of truly democratized data" —James Mayfield, co-founder, Transform While data analytics and business intelligence efforts are traditionally the domain of data experts, organizations can empower non-technical stakeholders to perform basic data operations via in-house training programs, workshops, and self-service tools that can simplify their onboarding and learning process. They can also use software that surfaces data in an easy-to-consume format for business stakeholders. Data democratization has multiple downstream benefits. It leads to greater data literacy, which can facilitate not only greater data-driven decision-making but also potentially lead to creation of new products or services based on insights mined from the data. Therefore, greater democratization, usage, and adoption of a data-driven approach can unlock massive commercial value and new growth levers for businesses. How do you actually democratize data? Implementing data democratization is a hard challenge and an ongoing process. To be successful, it needs support, buy-in, and a lot of patience from the leadership. Apart from conceptualizing and implementing curated data governance frameworks and policies, organizations can leverage tools to enable data democratization at scale. Tools to enable data democratization The Data Catalog A data catalog is a collection of metadata that, combined with data management and search tools, helps data stakeholders find and acquire data for downstream analytics. A data catalog provides a managed and scalable data discovery and metadata management capabilities which are fundamental requirements of attaining higher levels of data democratization in an organization. The Data Mart A data mart is a subset of a data warehouse focused on a specific business vertical or data domain. Data marts enable specific users to access specific data that empowers them to quickly access these datasets without wasting time searching for the same in the data warehouse. For instance, individual departments like sales, marketing, operations, and finance can have their respective data marts for accelerating their domain-specific data-driven decision making. The Metrics Catalog A metrics catalog is a new layer in the modern data stack. It is a centralized store for all of your organizations’ most important metrics (or key performance indicators) and it's uniquely positioned between the data warehouse and downstream tools. As a self-service place for business KPIs, every stakeholder in the organization has access to track their own metrics and share context with others. By capturing core business metrics in this fashion and this location in the modern data stack, a metrics catalog provides immense visibility and transparency into an organization's most critical metrics and metric lineage for all stakeholders in an organization. This new concept of a metrics catalog can have a significant role to play in democratizing data to everyone. As a single source of ground truth for business data, a metrics catalog enables diverse stakeholders to base all key decisions on the same foundation. It also allows for disparate teams to use the same metrics, ask questions, and keep everyone aligned and on track. This greatly enhances the level of data democratization within an organization. Challenges for data democratization Although the benefits of data democratization are pretty evident, there are also numerous challenges. Some challenges are common, like data being kept in silos and unclear data ownership. The informational silos problem is antithetical to data democratization, and can adversely impact an organization's ability to leverage data for improvising its business performance and decision making. Different teams have ownership of different types of data, which contributes to the problem of information silos. When a particular team has exclusive access to specific data assets, they not only hinder other teams from accessing the data but also guard their analysis and insights derived from the same data. This often leads to duplication of efforts across teams, causing a massive waste of organizational time and resources. As each individual team or department hoards its own data and analyses, it contributes to the adoption of the same undemocratic processes across other teams further compounding the challenges in promoting data democratization. With greater access to the organizational data assets, there is also a challenge of data security, privacy, and potential misuse of the data. It increases the number of gaps in the organization which might become vulnerable to adversarial attacks and data breachers. This is why it’s important to have a balance between data security and data access—including having stronger safeguards around who can access and analyze personally-identifiable information and customer data. Looking ahead If implemented well, data democratization can provide an immense competitive edge that will only compound over time as organizations mature in their digital transformation journey. Several tools and data artifacts can aid in better implementation and adoption of best practices and policies that help in democratizing data. A metrics catalog is one relatively new tool that provides a centralized store of business critical information accessible to multiple stakeholders. It captures essential business metrics and provides a simplified interface that is agnostic of the separate analytics, CRM, and BI platforms used by various teams in the organization. Learn more about how a metrics store can promote data democratization and governance at Transform.co.
0 Comments
Introduction
Data governance is a fundamental pillar of modern digital businesses. It refers to a framework of processes and guidelines that companies use to ensure all enterprise data assets are managed and utilized appropriately. Even if an organization has large investments in data infrastructure and teams, without a structured data governance framework, organizations will struggle to harness the full value of their data. A strong framework provides a clear set of guidelines for all employees who access and consume data in downstream applications. It also contributes to greater trust in the authenticity and quality of data and allows data stakeholders to focus on core data tasks instead of worrying about whether the data was created, processed, stored accurately, and in compliance with national or domain-related legislations like GDPR, HIPAA, CCPA, and data localization laws. Given recent data breaches, the importance of a structured data governance framework cannot be emphasized enough. In this article, you’ll learn how to ensure data quality through better data governance mechanisms, leading to an increase in data informed decision-making. You’ll also learn how a clear data governance framework contributes to improved data quality and value creation across the entire organization. Why do you need data governance? The digital revolution is founded on data and the idea that data can generate insights that are critical for decision-making and long-term planning. With the emergence of cloud technologies, it’s easier for businesses to see the importance of data and store it in a more accessible, scalable, and secure way. A data governance framework is a set of rules and processes for collecting, storing, and using data. This diagram shows a simplified outline for how to think about building a data governance framework for your organization.However, collecting and storing data is just the tip of the iceberg. Without a clear and robust governance framework, you can’t fully understand the value of your data. High-quality data will help you make the best possible decision for your company. A data governance framework consists of several layers, stakeholders, business goals, and structured processes with a focus on information and project management. This accountability means organizations can build high-quality data products with confidence. This is evident in the case of top technology companies like Google and Amazon that have invested early and massively in data and data-driven technologies. They benefited from investing and enforcing a data governance framework that lowers the organizational threshold, velocity, and efficiency with which businesses can adapt to change. So, why is data governance important? Investing in data governance leads to many benefits including:
Ensure data quality through governance A major outcome of a solid data governance framework, if carried out properly, is improved data quality. When organizations follow these guidelines, it leads to a clearer understanding of their data assets and increases accountability. First, think about your data lineage. Record the source of each data set and the date/time that it is accessed. It’s also critical to understand the teams that are accessing the data including the applications they’re using.This ensures compliance and prevents data breaches. You can test data quality by asking different stakeholder teams to provide the value for a common business metric. More often than not, different teams will have conflicting answers for the same metric. This can be the result of a flaw in your data governance strategy, fuzzy guidelines, or scattered metrics logic across downstream tools. Create policies that ensure data accuracy Maintaining accurate data across the organization is difficult but rewarding. Once a new data asset is created, either internal or external, it needs to be systematically logged and entered into the appropriate databases. Consistently using data governance best practices for completeness, relevance, reliability, and lifecycle can lead to better data quality and accuracy. Develop practices to test data completeness Data completeness refers to the wholeness of the data. Data is complete when there are no missing values, records, or duplicates. Basic automated checks to validate the number of rows and columns, dimensionality, missing and null values, and data format mismatch can help identify missing elements. Adopt technologies to check data relevance Data relevance refers to the utility of data in providing critical insights. It’s important to remember that not all data is useful or relevant to particular business problems, and identifying the right set of input data can help focus subsequent analytics and modeling efforts. Track relevance with data reliability Data reliability is an indicator of how useful and relevant it is over time. It builds upon the concepts of completeness and relevance, and is more likely to be used and reused by teams for their work. This lays the foundation for multiple use cases and business insights. Stay compliant with data depreciation and lifecycleData timeliness and lifecycle management provides clear timelines for the validity and deprecation of data, ensuring that it’s used only when relevant and compliant with privacy laws. This regulates the lifecycle before it is depreciated or deleted permanently. Standardizing metrics as part of your data governance strategy Let’s take a look at how you can standardize your metrics through metrics catalogs and policies and build into a data governance strategy that ensures data quality. Catalog metrics in a metrics storeStandard metrics like annual recurring revenue (ARR), gross merchandise value (GMV), customer acquisition cost (CAC), customer lifetime value (LTV), and net promoter score (NPS) are common. Once you've defined your metrics, these metrics can be stored in a metrics catalog for greater ease of access, use, and re-use across the organization. A metrics catalog has several advantages. It reduces valuable organizational time and effort to reproduce the underlying analysis, and it creates a centralized metrics store that facilitates better understanding and decision-making. As depicted in the figure below, a metrics store is a centralized and governed place for organizations to store key metrics, creating a repository for stakeholders to access key metrics in a repeatable way, regardless of where people access their data. Policies and practices for sign-off Before creating a metric, there needs to be a clear policy on the steps that people use to analyze and validate their business metrics. Data quality policies should not be treated as an administrative exercise but regarded as an important milestone in this stage of data transformation. In addition to assigning an owner for each of your critical metrics, you should also think about executive sponsorship for the organization’s most important, “north-star” metrics. A stamp of approval from the C-suite or an executive sponsor conveys the importance of the data policy framework to the entire organization but can also be used to negotiate and expedite resolutions when conflicts arise. Conclusion In this article, you’ve learned about data quality as an index that can be used for many attributes of data in an organization. A data governance framework creates a set of best practices that improve data accuracy and relevance. A data governance framework also makes it possible to distribute high-quality data to your teams in the most efficient way possible. Building a metrics store is a critical part of this process because metrics are the language that you use to express whether you achieved your organizational goals. A metrics store, like the Transform Metrics Store, centralizes all of this knowledge in one place for easy access and collaboration. To learn more about the metrics catalog and other solutions, visit Transform.co. Introduction
Data drift is a common problem for production machine learning systems. It occurs when the statistical characteristics of the training (source) and test (target) data begin to differ significantly. As illustrated in the image below, the orange curve depicting the original data distribution shifts to the purple curve, representing a change in statistical properties like the mean and variance. Understanding data drift is fundamental to maintaining the predictive power of your production machine learning systems. For instance, a data science team may have started working on a machine learning use case in 2019, using training data from 2018, but by the time the model is ready to go into production, it’s 2020. There could be a huge change in the distribution between the source data from 2018 and the live data coming from 2020. Any time a machine learning model is ready to be shipped, it needs to be rigorously tested on live data. It’s critical that you detect data drift before deploying a model to production. In this article, I’ll illustrate the various types of data drift and how data drift impacts model performance along with several examples. I’ll also address data labeling, one of the popular ways to tackle data drift, and how to perform data labeling efficiently. Why Data Drift Happens? In real-world situations, data drift can occur due to a variety of reasons:
Continuing with the COVID-19 example, a model trained on data prior to the onset of global lockdowns, say from January to February 2020 will yield poor predictions on data in March and April 2020 after the lockdowns started. Thus, the original trained model is no longer relevant or practically useful and needs to be retrained. Even small changes in the data structure or format of the source data can have significant consequences for machine learning models. For instance, a change in the format of a data field, like an IP address or hostname or ID, can often go undetected for a long time without effective root cause analysis. Types of Data Drift There are different types of data drift, but the two principal ones are: Covariate drift refers to data drift associated with a shift in the independent variables. It happens when a few features change while still maintaining the same relationship between the feature and the target variable. Covariate drift primarily occurs due to sample selection bias, which is a systematic bias in the selection of training data that results in a nonuniform and nonrepresentative training dataset. Nonstationary environments, where the training environment differs from the test environment, also cause covariate drift. Concept drift, on the other hand, occurs when the relationship between the independent variables and the target variable changes. Consider a product recommendation machine learning model in the context of e-commerce, where the original model is trained on user activity and transactions from users located in the US. Now imagine that the e-commerce company is going to launch in a new locale or market with the same product catalog as in the US. The original recommendation model will perform poorly when applied to users from the new market with significantly different online shopping behavior, financial literacy, or internet access for e-commerce. In this example, the online shopping behavior of the users is markedly distinct. Even if the same features are used to train the machine learning model, it might underperform significantly. In such cases, concept drift is the root cause of data drift, and the personalization model needs to be reworked and include new features that better capture the new user behavior. Overcoming Drift with Data Labeling To overcome data drift, you need to retrain the model using all available data, including data from before and after drift occurred. New data needs to be labeled accurately before including it in the new training dataset. Data labeling refers to the process of providing meaningful labels to target variables in the context of supervised machine learning where the target could be an image or text or an audio snippet. In the context of data drift, data labeling is crucial to countering data drift, and thereby directly affects the performance of machine learning models in production. Data labeling is integral to supervised machine learning where a model is fed input data along with relevant labels depending on the use case. For example, for a model learning to detect product placement in videos, the model is fed a video with products highlighted in the video. Typically, data labeling is a manual exercise that’s both costly and time-consuming. It’s often outsourced to vendors in developing countries associated with low cost of labor. Annotators need to be trained to use labeling software, understand the machine learning use case and the annotation framework, and deliver highly accurate labels at a high velocity and throughput. In such a scenario, labeling errors can occur, which exacerbates the problem of data drift if data from the new test or target distribution isn’t labeled accurately. In practice, several controversial labeling errors have occurred that cause reputational damage to the company, for instance, when Google Photos labeled two Black people as “gorillas.” Big technology companies like Google and Facebook are grappling with such issues in their automated data labeling algorithms. Labeling errors can be made by human annotators, and also by machine learning models. Once trained, the predictions made by machine learning models on new data are often reused to augment the original training data to further improve the models. In such scenarios, data labeling errors can compound resulting in imperfect models that often yield such bizarre and controversial results. Data labeling helps alleviate data drift by incorporating data from the changed distribution into the original training dataset. If enough new data is labeled, then it is possible to drastically reduce data drift by simply dropping the older data and only using the newly labeled data. Therefore, proper and efficient data labeling is a crucial exercise with significant commercial impact, depending on the nature of the machine learning application. For example, incorrect data labels in a fraud detection use case can result in monetary loss every time the fraud detection machine learning model makes an incorrect prediction. Inaccurate data labels not only impact the performance of the machine learning model but also indirectly contribute to data drift. Any systematic data labeling errors may compound the problem as the model’s predictions on new data are typically leveraged to augment the training dataset. Data labeling can be improvised and performed effectively through the use of intuitive software that enables human annotators to label data with high speed and low cognitive load. For additional improvement in data labeling, you can implement inter-annotator agreement; a particular training example is assigned a label that’s selected by a majority of the annotators. For example, if four out of seven annotators assign “Label1” to a particular data sample and the other three annotators assign it “Label2,” then the data sample would be tagged with “Label1.” Strong operational practices including auditing of randomly selected labels for accuracy can improve the process and provide feedback about systematic labeling errors. You can also use machine learning to aid data labeling with a model trained on a sample of data that’s labeled by humans to generate predictions on new or unlabeled data. These noisy labels can then be leveraged to build better machine learning models by incorporating the data samples associated with high probability and sending the data samples with low probability back to human annotators for more accurate labels. This process can be repeated iteratively to improve the overall performance of the model with minimal human data labeling efforts. Conclusion Data drift can have a negative impact on the performance of machine learning models as data distribution changes. This can cause a machine learning model’s predictive accuracy to go down over time if not countered effectively. Data labeling is one technique to reduce data drift by applying labels to data from the new or changed distribution that the model does not predict well. This helps the machine learning model to incorporate this new knowledge during the training process to improve its performance. There are several tools available today that enable annotators to label data efficiently. For example, Label Studio is an open-source data labeling tool that provides a platform for labeling different data types, including images, text, audio as well as multi-domain data. It’s already used by leading technology companies including Facebook, NVIDIA, Intel, so check it out if you’re looking for a robust, open-source solution for reducing data drift. Published by Neptune.ai Introduction
Only 10% of AI/ML projects have created positive financial impact according to a recent survey of 3,000 executives. Given these odds, it seems that building a profit generating ML project requires a lot of work across the entire organization, from planning to production. In this article, I’ll share best practices for businesses to ensure that their investments in Machine Learning and Artificial Intelligence are actually profitable, and create significant value for the entire organization. Best practices for identifying AI use cases Most AI projects fail at the very first hurdle – poor understanding of the business problems that can be solved with AI. This is the main bottleneck in successful deployment of AI. This problem is compounded by the early stages of organizational intuition for AI, and for how it can be leveraged to solve critical business problems [2]. What does this mean? Well, not every problem can be feasibly solved with AI. To understand if your particular problem can, you need tried and tested practices and approaches. AI use cases AI has transformed industries. It automates routine and manual processes, and provides crucial predictive insights to almost all business functions. Table 1 shows a list of some of the business use cases that have been successfully addressed using AI. Brainstorming appropriate business problems should ideally be done together with business leaders, product managers, and any available subject matter experts. The list of business problems sourced across the organization should then be vetted, and analyzed for potential solutions using AI. Not every business problem should be solved with AI. Oftentimes, a rule-based or engineered solution is good enough. Additionally, a lot of business problems can be mined from customer reviews or feedback, which typically points to broken business processes that need to be fixed. In table 2, you can see a checklist of questions, both technical and commercial, to determine whether a business problem is relevant for AI. KPIs and Metrics As part of the planning process, the appropriate model and business metric for each potential use case should be discussed. Work backwards from the expected outcome, and it’ll be easier to crystallize which particular metric to optimize. To illustrate this, in table 3 I prepared a list of AI use cases and corresponding model and business metrics. For the success of an AI project, it’s ultimately important to ensure the business metric and goals are achieved. Prioritization We have a set of business problems. They’ve been reviewed and documented after careful consideration of the criteria listed in Table 2, and analysis of appropriate business metrics as in Table 3. The candidate list of use cases needs to be prioritized, or ranked, in terms of impact and relevance to the overarching business strategy and goals. From a detailed written document describing comprehensive facets of the business use case and potential AI-based solutions, it’s useful to have objective criteria to quantify all the proposed use cases on the same scale. Here, it’s crucial for product managers and business leaders to have their own intuition about how AI works in practice, or rely on the judgment of a product-focused technical or domain expert. Whilst it’s easy to rank projects on certain success criteria, it’s not so straightforward to rate the risk associated with AI projects. A balanced metric ought to consider and weigh the likelihood and impact of a successful outcome of the AI projects versus the risk of it failing or not generating enough impact. Risks to the project might be related to organizational aspects, domain-specific aspects of the AI problem, or related to external factors beyond the remit of the business. Once a suitable balanced metric is defined, it aligns all stakeholders and leadership, who are then able to form their own subjective views based on the objective scores. A lot of factors need to be considered before a ‘yes’ or ‘no’ decision is made for a particular AI project, as well as the number of AI-relevant projects selected for a defined period. Securing buy-in from the leadership is difficult. Certain final executive decisions might appear subjective or not data-driven, but it’s still absolutely critical to go through the aforementioned planning process to present each AI project in the best light possible, and maximize the likelihood of the AI project being selected for execution. Best practices for planning AI use cases As part of the planning process with cross-functional teams, it’s important for organizations to have a streamlined mechanism for defining the AI product vision or roadmap, the bandwidth, specific roles and responsibilities of individual contributors and managers in each team, as well as the technical aspects (data pipelines, modeling stack, infrastructure for production and maintenance). In this section, I’ll describe the details of specific planning steps essential to build a successful AI product. AI product requirements For each identified use case, it’s necessary to draw the roadmap for how the product will evolve from its baseline version to a more mature product over time. In Table 4, I outline a set of essential questions and criteria to fulfil for creating a comprehensive AI roadmap for each use case. PR-FAQ (Press Release – Frequently Asked Questions) and PRD (Product Requirements Document) are two critical documents that are generally prepared during the initial stages of product ideation and conception. Pioneered by Amazon, these two documents serve as the north star for all concerned teams to align themselves with and build and scale the product accordingly. It’s absolutely essential that all stakeholder teams contribute meaningfully to these documents and share their specific domain expertise to craft a meticulous document for executive review. It’s necessary for all stakeholder team managers to review and contribute to the document, so that any team- or domain-specific intrinsic biases of product development are laid bare and addressed accordingly. Typically, teams should rely on data-driven intuition for product development. In the absence of in-house data, intuition for the AI product can be borrowed from work done by other companies or research in the same field [2, 4]. Data requirementsAs the roadmap is defined and finalized after stakeholder meetings, it’s always beneficial to have an MVP or a basic prototype of the AI product ready to validate initial assumptions and present to the leadership. This exercise also helps to streamline the data and engineering pipelines necessary to acquire, clean and process the data and train the model to obtain the MVP. The MVP should not be a highly sophisticated model. It should be basic enough to successfully transform the input data to a model prediction, and trained on a minimal set of training data. If the MVP is hosted as an API, each of the cross-functional stakeholder teams can explore the product and build intuition for how the AI product might be better developed for the end customer. From a data perspective, the machine learning team can dive deeper into the minimal training data, and do a careful analysis of the data as listed in Table 5. Model requirements After systematic assessment of the data quality, features, statistics, labels and other checks as listed in Table 5, the Machine Learning team can start building the prototype / MVP model. The best approach at the early stages of product development is to act with speed rather than accuracy. The initial (baseline) model should be simple enough to demonstrate that the model works, the data and modeling pipelines are bug-free, and the model metrics indicate that the model performs significantly better than chance. Machine learning use cases and products have become increasingly complex over the years. Whilst linear regression and binary or multi-class classification models were once too common, there are newer classes of models that are faster to train, and generalize better on real-world test data. For the ML scientist or engineer, no two use cases may be built using an identical tech stack of tools and libraries. Depending on the characteristics of the data relevant for the AI use case (see Table 2), the data science team must define the modeling stack specific to each use case (see Table 6 below). Best practices for executing AI use cases After identifying and planning for promising AI use cases, the next step is to actually execute the projects. It might seem that execution is a straightforward process, where the machine learning team gets to weave their magic. But, simply ‘building models’ is not enough for successful deployment. Model building has to be done in a collaborative and iterative fashion:
In the next section, I will discuss the best practices for the operational aspects of executing and deploying AI models successfully and realizing the proposed commercial value. Reviews and feedback Once the AI project has kickstarted, it’s essential for the machine learning team to have both periodic as well as ad-hoc review meetings with stakeholders, including product teams and business leadership. The documents prepared during the planning phase (PR-FAQ and PRD) serve as the context in which any updates or changes should be addressed. The goal of regular meetings is to assess the state of progress vis-a-vis the product roadmap, and address any changes in:
While planning is important, most corporate projects don’t go as initially planned. It’s important to be nimble and agile, respond to any new information (regarding technical, product or business aspects), and re-align towards a common path forward. For example, the 2020 lockdowns severely impacted the economy. In light of such high-impact unexpected events, it’s critical to adapt and change strategy for AI use cases as well. In addition to regular internal feedback, it’s good to keep in touch with the end users of the product throughout the AI lifecycle. In the initial stages (user research, definition of target user personas and their demographics), and especially in product design and interaction with the model predictions. A core group of users from the target segment should be maintained to obtain regular feedback across all stages of product development. Once an MVP is ready, users can be very helpful in providing early feedback that can often bring to light several insights and uncover any biases or shortcomings. When the AI model is ready to be shipped and different model versions are to be evaluated, user feedback can again be very insightful. User insights about the design, ease of use, perceived speed and overall user flow can help the product team to refine the product strategy as needed. Building iterativelyFrom the technical perspective, the model building process is usually an iterative one. After establishing a robust baseline, the team gets insight into how far the model performance is from the established acceptance criteria. In the early stages of model building, the focus should primarily be on accuracy rather than latency. At each stage of model development, a comprehensive analysis of model errors on the validation set can reveal important insights into the model shortcomings, and how to address them. The errors should also be reviewed in conjunction with subject matter experts, to evaluate any errors in data annotation as well as any specific patterns in the errors. If the model is prone to a particular kind of error, it might need additional features. Or it might need to be changed to a model based on a different objective function, or underlying principle, to overcome these errors. This repetitive process helps the machine learning team to consolidate their intuition about the use case, think outside the box, and propose new creative ideas or algorithms to achieve the desired metrics. During the course of model building, machine learning practitioners should systematically document every experiment and the corresponding results. A structured approach is helpful not only for the particular use case, but also helps build organizational knowledge that can be helpful to onboard new hires, or serve as shining examples of successful AI deployment. Deployment and maintenance Once the candidate machine learning model is ready and benchmarked thoroughly on the validation and test sets, errors analyzed, and the acceptance criteria met, the model may be taken to production. There’s a huge difference between the model training and deployment environments. The format in which the model is trained may not be compatible with taking the model to production, and need to be appropriately serialized and converted to the right format. In an environment that simulates the production settings, model accuracy and latency should be validated again on the hold-out dataset. Deployment should be done incrementally by surfacing the model to a small portion of real-world traffic or input to the model, ideally to be tested first by internal or core user groups. Once the deployment pipeline has been rigorously tested and vetted by the MLOps team, more traffic can be directed to the model. In scenarios where one or more candidate models are available, A/B testing of these models should be done systematically, and evaluated for statistically significant differences to determine the winning model. Post-deployment, it’s important to ensure that all the input-output pairs are collected and archived appropriately within the data ecosystem. The launched model should be periodically assessed and the distribution of the real-world data compared with the distribution of the training data to assess for data and model drifts. In such cases, an active learning pipeline that feeds some of the real-world test samples back into the original training dataset helps to alleviate the shortcomings of the deployed model. Finally, once the model production environment and all pipelines are stable, the machine learning and product teams should evaluate the business metrics and KPIs to assess whether the metrics meet the predefined success criteria or not. In case it does, then only can the use case be deemed to be a success and a summary of the overall use case and results should be documented and shared internally with every stakeholder and the business leadership. Wrapping up If machine learning, product and business teams in startups and enterprises adopt a systematic approach and follow the best practices as laid out in this article, then the likelihood of successful AI outcomes can only increase. Adequate upfront preparation is crucial. Without it, teams won’t be able to rectify any errors or respond to changes, nor realize the massive commercial potential that AI can deliver. References
Published by Neptune.ai Introduction
In this article, I have documented the best practices and approaches to build a productive Machine Learning team that creates positive business impact and generates economic value within corporate entities, be it startup or enterprise. If you do Machine Learning, either as an individual contributor or team manager, I’ll help you understand your current team structure and how to improve internal processes, systems and culture. We’ll explore how to build truly disruptive ML teams that drive successful outcomes. Why build an ML team? Artificial Intelligence (AI) is predicted to create global economic value of nearly USD 13 Trillion by 2030 [1]. Most companies across diverse industries and sectors have realized the potential value of AI, and are well on the way to becoming an AI-first entity. From tech companies building cutting-edge AI products like self-driving cars or smart speakers, to traditional enterprises leveraging AI for non-glamorous use cases like fraud detection or customer service automation, the potential of AI to deliver commercial impact is beyond doubt. The adoption of AI in industry is accelerated by a number of trends:
In the following section, I will describe the challenges in building Machine Learning teams for startups and enterprises respectively. Challenges for startups Startups, in the early stages of operations, are typically bootstrapped and have limited budgets to deploy for building machine learning teams. If your startup has a core product or service founded on AI, then it’s imperative to hire machine learning talent early on to build the MVP, and raise funding to hire more talent and scale the product. On the other hand, for startups whose core product or service is focused on other domains like finance, healthcare or education, AI will either be incidental to the core operations, or not essential until product-market fit is achieved. The main challenges of building ML teams in startups are:
In the face of such daunting challenges of machine learning work combined with general organizational challenges at startups [2], it becomes even more important for startups to hire and build the right machine learning team from the very beginning. Challenges for enterprise Unlike startups, big organizations and enterprises don’t suffer from lack of funding or budget to seed a machine learning team. The challenges in an enterprise are unique from one entity to another, but generally arise due to the size of the organization, internal bureaucracy and slower decision making processes – things that tend to benefit startups and help them ship products faster. Although today, it might appear that technology companies are ubiquitous, they’re still a minority compared to the vast number of traditional enterprises focused on diverse sectors like finance, FMCG, retail, healthcare, education and so on. Technology companies have a headstart when it comes to machine learning and AI, and their strong early focus and investment in AI R&D will ensure their dominance compared to their traditional counterparts. However, there are numerous challenges that traditional enterprises face in adopting and onboarding AI across the organization [3], which more often than not result in failed AI projects and reduced trust in the capacity and potential of AI [4]:
Profiles in a Machine Learning team Modern machine learning teams are truly diverse. Yet, at the core, they involve candidates who have strong analytical skills and the ability to understand data from different domains, train and deploy predictive models, and derive business or product insights from the same. SCOPING The first stage of scoping out an AI use case requires AI experts along with business or domain experts. Plenty of successful AI projects start with a deep understanding of the potential business problems that can be solved with AI, and require the combined intuition and understanding of seasoned technical and business experts. In this stage, the usual collaborators involve business leaders, product managers, AI team managers and perhaps one or more senior data scientists with deep, hands-on experience with the underlying data. DATA The second stage is focused on acquiring data, cleaning, processing from the raw form to structured format and storing it in specific on-premise databases or cloud repositories. In this stage, the role of the data engineer is prominent, alongside data scientists. The business and product managers serve a helpful role in providing access to the data, metadata and any preliminary business insights based on rudimentary analytics. MODELING The third stage involves core data science and machine learning modeling using the datasets prepared in the previous stage. In this stage data scientists, applied or research scientists are predominant in training initial models, refining them based on test set performance and feedback from cross-functional stakeholders, developing new algorithms if needed, and finally producing one or more candidate models that meet the required accuracy and latency benchmarks to take the models to production. DEPLOYMENT The final stage of the machine learning lifecycle is focused on deploying trained models to production, where they serve predictions from the inputs received from end users. In this stage, machine learning engineers take the models developed by the data/applied/research scientists and prepare them for production. If the models meet the predefined accuracy and latency benchmarks, the models are good to go live. Otherwise, ML engineers work on optimizing the model size, performance, latency and throughput. Models go through systematic A/B testing procedures before deciding which version(s) of the models are best suited for deployment. Next, I prepared detailed profiles for the different types of experts you may need for your ML team. Data Engineer Skills
Responsibilities
Tech stack
Data Scientist Skills
Responsibilities
Tech stack
Machine Learning Engineer Skills
Responsibilities
Tech stack
Research Scientist Skills
Responsibilities
Tech stack
Product Manager + Business Leader Skills
Responsibilities
Tech stack
Data Science / Machine Learning Manager Skills
Responsibilities
Tech stack
Building productive and impactful Machine Learning teams We explored the typical composition of a Machine Learning team, which includes a variety of different profiles specialized in specific aspects of building machine learning projects. However, the reality on the ground is that having a solid machine learning team is not a guarantee that the team will create and deliver massive business impact. The reality on the ground is that the vast majority of corporate AI projects fail, and a lot of these projects fail despite having a great machine learning team. In this section, I will dive deeper into the cultural, procedural and collaborative aspects of building impactful machine learning teams from first-principles. The success of a machine learning team is founded on several factors related to systems, processes, and culture. When built the wrong way, this will inevitably lead to failed projects and erosion of trust and confidence in the team, as well as machine learning as a business capability and competitive edge. 1. Working on the right AI use cases For a brand new machine learning team to deliver impact in an organization, it’s paramount that the team starts off on the right foot. Early traction is critical to build trust in the organization, evangelize the potential of AI across business verticals, and leverage early successes to deliver riskier or moonshot projects with greater impact. 2. Planning for success – measuring impact As part of the process of selecting and defining the right AI use cases, it’s fundamental to critically assess and evaluate the business impact and return on the investment in the particular machine learning project. The best approach for evaluation is by defining a set of metrics that address several aspects of the project and its potential impact. Technical metrics For classification models:
For regression models:
For deep learning models (depends on the particular application):
Business metrics Business metrics are defined by first-principles, and are often downstream metrics that are impacted by the machine learning models. For measuring outcomes, it’s crucial to a priori identify the relevant business metrics and track the effect of the machine learning models on the same during A/B testing, deployment, and continuously monitor live models. Standard business metrics aim to capture levels of trust, satisfaction, faults, and SLAs, among others. Once a candidate set of machine learning projects is scoped, defined and formulated from conception to production with associated set of metrics, each project needs to be evaluated by leadership teams from the perspective of high-level organizational goals to be achieved in a defined time period. Leaders need to balance the business impact (on the opline or bottomline), budget, team bandwidth, time savings, efficiency savings, and the urgency for delivering projects in the short-term vs. the long term. Executives need to incorporate multiple factors to arrive at a carefully considered decision to give the green signal for one or more machine learning projects. 3. Structured processes – Agile, Sprints Once a project is defined and has the go ahead from the leadership team, it is important to ensure that systems and structured processes are in place to ensure that the machine learning team can work unhindered and execute the project in a timely fashion as per the agreed plan. Key operational infrastructure like data warehouse, database management systems, data ETL pipelines, metadata storage and management platforms, data annotation frameworks and availability of labeled data, access to compute on-prem or in the cloud, licensed as well as open source tools and softwares that streamline the model training process, machine learning experiment, results and metadata management tools, A/B testing platforms, model deployment infrastructure and solutions, continuous model monitoring and dashboards are integral for a smooth data processing, model building, and deployment workflow. However, the existence of such key skeletal infrastructure for machine learning varies from one organization to another depending on how mature the machine learning organization or the company is. Apart from the infrastructure, processes related to planning tasks of the individual contributors of the project using sprints and agile frameworks need to be hardwired and accessible to all stakeholders of the project. While Agile processes have worked well for software projects, machine learning projects are different and may not be that well suited to the same frameworks. Although similarities like iterative model building and refining based on feedback exist, machine learning projects are more sophisticated, as the fundamental blocks include data and models in addition to code. While software engineering best practices like code review and versioning are very well established, the same rigor and structure is not always applied to data and machine learning models. Documentation is another aspect that is even more critical to keep track of multiple hypotheses, experiments, results and all the moving parts associated with machine learning projects. In the absence of well entrenched tools and best practices, most data science work tends to be highly inefficient where data scientists end up spending a lot of time on routine chores that can be automated. It’s imperative that managers try to reduce such barriers to more efficient and productive work, so that the machine learning teams can focus exclusively on their work. 4. Clear communication within and across teams Communication is an essential skill for data scientists. Machine learning is a more intricate discipline and the end results might often be too obscure for generalist and non-technical managers of data science, product or business teams to comprehend easily. However, communication is just the tip of the iceberg, and many more interpersonal skills like persuasion, empathy, collaboration are exercised on a regular basis whilst working in cross-functional teams. Writing emails of results or updates or slide presentations to stakeholders and leadership, live demos, expounding the project for product review documents, writing up the entire project for a blog meant for lay audience or for a journal or conference meant for a technical audience, requires strong writing skills. Typical data scientists may be more proficient in writing code than words, so the organization should invest in corporate training programs for data scientists that include training in written and spoken communication skills. Oral communication skills can’t be underestimated either, and are increasingly important in remote-first organizations. Effective stakeholder management involves building rapport and trust and establishing clear channels of communication, which is much harder to do if a data scientist is not able to speak and communicate clearly in an engaging and delightful manner. Although a lot of workplace productivity apps have created digital channels of reduced in-person communication, the power of live in-person communication with peers, stakeholders and leaders often gets the job done faster. Clear communication destroys information silos, so that each stakeholder is aware, updated and aligned with the progress of various machine learning projects. Regular meetings are important to have checks and balances, in addition to documented progress in tools to ensure that projects are moving in the right direction. 5. Effective collaboration with business Machine learning teams are typically part of the engineering or technology organizations in a company. While this makes natural sense for effective collaboration across colleagues from data, analytics, engineering functions, regular interaction with business teams is a must. Given the fact that most machine learning models are built on historical ‘business’ data that can change in a predictable manner due to new product or feature launches or seasonality patterns, as well as in an unpredictable manner, for instance, during Covid-19 lockdowns, machine learning teams must have a real-time awareness of how the business data is changing on the ground. Not only is it important to adjust the underlying hypotheses in the face of massive changes in customer behavior or new product launches, but also to correct the planned course of action if initial assumptions are violated or the data changes too dramatically for the machine learning models to be relevant or have the same impact as before. Business teams are in the best position to give feedback on early prototypes based on their domain expertise, validate new assumptions or ideas by doing customer research and surveys, and evaluating the impact of deployed machine learning models. For these reasons, the partnership between machine learning and business teams needs to be mutually beneficial and symbiotic. Leaders of machine learning teams need to build close ties with business teams and encourage team members to do the same. 6. Creating a culture of innovation For long-term success of machine learning teams, apart from working on the right use cases and facilitating collaborative work across the organization, it’s imperative to build a culture that embraces and rewards innovation. Here, leadership should lead by example and encourage innovation and R&D across different business verticals. For a machine learning team, it’s critical to make a mark in the ecosystem through patent applications, journal or conference publications, outreach and dissemination via meetups, workshops, seminars by leading experts, collaboration with startups and academic organizations as needed, and so on. Most organizations don’t focus on building such a thriving culture that promotes exchange and cross-fertilization of new ideas and technologies, which can often impact current organizational processes and thinking in a substantial way. Leaders also need to build strong diverse teams and hire new talent, from entry level graduates to experienced engineers and scientists. The inflow of new talent brings in novel ideas that can positively impact the work culture. Otherwise stasis sets in, teams can become narrow-minded, and decline in their capacity to innovate and launch impactful products. Meritocratic executive decisions strongly impact culture, both in terms of promoting talent that demonstrates a consistent track record of exceptional bar-raising work, as well as letting go of non-performing individuals or managers. The appropriate balance and culture in a team is an ongoing process, but it’s important for leaders to ensure that at no point in time, the members of a machine learning team are unmotivated and uninspired by the systems, processes, and culture within the organization. 7. Celebrating and sharing AI success stories Finally, given the low odds of success for AI projects at present, it’s important to make sure that any AI success stories are widely shared within the organization to attract the attention of other business teams who could potentially partner with the machine learning team. Furthermore, given the immense popularity of AI as a discipline, success stories might also attract potential new team members from within the company who feel motivated to upskill in machine learning and become a data scientist. It’s important to recognize the effort of the core contributors to the success of AI projects in a public manner within the company and not behind closed doors. It helps to build morale and confidence and foster a meritocratic culture within the team that will help them in their career development. Additionally, wherever possible, the leadership should take steps to share such AI success stories widely within the broader ecosystem in which the company operates, for instance, via company blogs, social media posts, podcasts or talks at meetups, workshops or conferences. For a machine learning team to continue to deliver strong performance and results, it’s critical to build a portfolio of successful projects starting from simpler ones to gradually more sophisticated ones with an ever increasing scope and commercial impact. The success of a machine learning team acts as a trigger and accelerates the digital and AI transformation of a company. In the highly competitive digital economy, companies that have invested early and invested a lot in AI have emerged as the early winners, for instance, the big tech companies. Thus, impactful machine learning teams act as a lever in the journey towards embracing and onboarding AI and transforming the company into a forward-looking, data-driven, AI-first company. References
Published in BecomingHuman.ai tldr: Poor processes and culture can derail the success of many an exceptional AI team In part 1, I introduced a four-pronged framework for analysing the principal factors underlying the failure of corporate AI projects:
In the second part of the blog series, I will focus on core aspects of organizational processes and culture that companies should inculcate to ensure that their AI teams are successful and deliver significant business impact.  Figure 1. The Facade Culture (source: Dilbert.com) Culture Organizational culture is the foundation on which a company is built and shapes its future outcomes related to commercial impact and success, hiring and retention, as well as the spirit of innovation and creativity. Whilst organizational behaviour and culture have been studied for decades, it needs to be relooked in the context of new-age tech startups and enterprises. The success of such cutting-edge AI-first companies is highly correlated with the scale of innovation through new products and technology, which necessitates an open and progressive work culture. Typically, new startups on the block, especially those building a core AI product or service, are quick to adopt and foster a culture that promotes creativity, rapid experimentation and calculated risk-taking. Being lean and not burdened by any legacy, most tech startups are quick to shape the company culture in the image of the founders’ vision and philosophy (for better or worse). However, the number of tech companies that have become infamous for the lack of an inclusive and meritocratic culture are far too many. There are innumerable examples, from prominent tech startups like Theranos, Uber to big tech companies like Google and Facebook, where an open and progressive culture has at times taken a back seat. However, with the increasing focus on sustainability, diversity and inclusion, and ESG including better corporate governance, it is imperative for tech companies to improve organizational culture and not erode employee, consumer or shareholder trust or face real risks to the business from financial as well as regulatory authorities as recently experienced by BlackRock and Deliveroo. Here is a ready reckoner of some of the ways AI companies tend to lose sight of culture:
 Figure 2. How to scope AI projects (source: DeepLearning.ai) Process There are several processes that are integral for ensuring a successful AI outcome across the entire lifecycle from conception to production. However, from first-principles, the primary process that needs to be streamlined and managed well is identifying the right use cases for AI that have the potential to create significant commercial impact. In this blog, I will focus only on this particular aspect and expound on the other processes in separate blogs. What can go wrong in identifying the right set of AI use cases?
So, having listed a variety of issues that can go wrong in identifying an AI use case, how should one ideally go about scoping AI projects systematically? As per Figure 2, the strategy to scope an AI use case involves 5 steps: from identifying a business problem to brainstorming AI solutions to assessing feasibility and value to determining milestones and finally budgeting for resources. The scoping process starts with a careful dissection of business, not AI problems, that need to be solved for creating commercial value. As discussed above, if not done right, the rest of the AI journey in an organization is bound to fail. Secondly, it is important to brainstorm potential AI solutions across AI, engineering and product teams to shortlist a set of approaches and techniques that are practically feasible instead of going with the latest or most sophisticated AI model or algorithm. Thirdly, AI teams should assess the feasibility of shortlisted methods by creating a quick prototype, validating the approach based on literature survey or discussions with domain experts within the company or partner with external collaborators accordingly. If a particular method does not appear to be feasible, then teams should consider the alternative approaches until they are ruled out. Once the initial efforts have validated the use case, its feasibility and potential approaches, it is critical to define key business metrics, KPIs, acceptance or success criteria. These are not composed of the typical AI model metrics like precision, accuracy of F-1 score, but KPIs need to be defined that are directly correlated with the impact of the AI models on business goals e.g. retention, NPS, customer satisfaction amongst others. The final step involves program management of the entire project from allocating time, bandwidth of individual contributors in the AI as well as partner teams, budget for collecting or labeling data, hiring data scientists or buying software or infrastructure to setup and streamline the entire AI lifecycle. Tldr part 2:
Before you head out to build AI, first ask what are the business problems that are big enough and suitable for an AI-based solution? What business metrics and objectives ought to be targeted? Scope out the problem systematically to ensure the best chance of success. Build on the initial successes of AI and foster a meritocratic and open culture of innovation and cross-functional collaboration to build AI that solves a variety of business use cases. Published in BecomingHuman.ai Tldr: Corporate AI failures can be ascribed to poor Intuition, Process, Systems, People The promise of AI is real. We are at the crossroads of the next industrial revolution where AI is automating industrial processes and technologies that were hitherto considered state-of-the-art. AI is expected to create global commercial value of nearly USD 13 Trillion by 2030 (McKinsey Global Institute). Given the immense commercial value that AI can unlock, it is no surprise that businesses of all kinds and sizes have jumped on the AI bandwagon and are repositioning themselves as ‘AI-first’ or ‘AI-enabled.  Source: Dilbert (https://dilbert.com) However, the groundbreaking progress and transformation that AI has brought across industry belies the stark reality of an increasing number of failed AI projects, products and companies (e.g. IBM Watson, and many more).
How can startups and large enterprises battle these tough odds to drive innovation and digital transformation across the organization? In this blog, I will examine from first principles common themes that typically underlie failed AI projects in corporations, and questions business leaders and teams should address when embarking on AI projects. I have classified these under four broad areas and will tackle each of these themes individually in future blog posts:
Part 1: Intuition (Why) Commercial AI projects often fail due to a lack of organizational understanding of the utility of AI vis-a-vis the business problem(s) to be solved. More often than not, throwing a complex AI-based solution at a problem is not the right approach, where a simpler analytical or rule-based solution is sufficient to have things up and running. It is therefore paramount to decode the business problem first and ask whether an AI approach is the only and best way forward. Unlike software engineering projects, the fundamental unit of AI is not lines of code, but code and data. In an enterprise, data typically belongs to a particular business domain, and is generated by the interaction of customers with specific business products or services. Here, a customer-centric approach is critical to understand the context in which this data is generated so that AI models may be developed to predict or influence user behavior to meet well-defined business objectives with clear success criteria. Wherever possible, the data scientists should themselves use and experiment with their company’s products/services by donning a ‘customer’s hat’ to decode the customer mindset. It’s hard to understand the nuances of training data if you don’t intimately understand the customer ‘persona’ to begin with. Data reflects more than just mere numbers. Making sense of data requires a holistic cross-functional understanding from a business, product, customer as well as technical perspective. Typically, these functional roles are played by different teams within a company, necessitating a strong collaborative effort to demystify the business problem, question the existing solutions and come up with new hypotheses, test and prove or disprove these hypotheses quickly via iterative experiments to hone in on a feasible solution and strategy. Here, the importance of domain knowledge or subject matter expertise cannot be stressed enough. It takes years to gain deep domain expertise which enables practitioners to develop better intuition for the business problem and the underlying data to propose feasible solutions or strategies. As data scientists typically lack expertise in business domains, it is imperative they complement their algorithmic data science skills with expert knowledge from those who work closely with the customer and understand the business problem intimately. Tldr (Part 1/4): Ask why is AI needed for your business problem? Is it the only way to solve the problem? And if yes, build and test hypotheses by leveraging the collective organizational knowledge and intuition across cross-functional teams that specialize in data science, business, product, operations. |
Archives
February 2026
Categories
All
Copyright © 2025, Sundeep Teki
All rights reserved. No part of these articles may be reproduced, distributed, or transmitted in any form or by any means, including electronic or mechanical methods, without the prior written permission of the author. Disclaimer
This is a personal blog. Any views or opinions represented in this blog are personal and belong solely to the blog owner and do not represent those of people, institutions or organizations that the owner may or may not be associated with in professional or personal capacity, unless explicitly stated. |
RSS Feed