The Core Responsibilities of the AI Product Manager

Product Managers are responsible for the successful development, testing, handout, and adoption of a product, and for passing the team that implements those milestones. Product administrators for AI must satisfy these same responsibilities, carolled for the AI lifecycle. In the first two commodities in this series, we suggest that AI Product Managers( AI PMs) are responsible for 😛 TAGEND

Deciding on the core run, public, and wanted apply of the AI productEvaluating the input data pipelines and ensuring they are maintained throughout the entire AI product lifecycleOrchestrating the cross functional unit( Data Engineering, Research Science, Data Science, Machine Learning Engineering, and Software Engineering) Deciding on key interfaces and schemes: user interface and experience( UI/ UX) and feature engineeringIntegrating the mannequin and server infrastructure with existing software productsWorking with ML operators and data scientists on tech stack blueprint and decision makingShipping the AI product and finagling it after releaseCoordinating with the engineering, infrastructure, and locate reliability crews to ensure all sent features can be supported at proportion

If you’re an AI product director( or about to become one ), that’s what you’re signing up for. In this article, we turn our attention to the process itself: how do you delivering a commodity to marketplace?

Identifying the problem

The first step in building an AI solution is identifying the problem you want to solve, which includes defining the metrics that will demonstrate whether you’ve attained. It resounds simplistic to state that AI product administrators should develop and ship makes that improve metrics the business cares about. Though these concepts may be simple to understand, they aren’t as easy in practice.

Agreeing on metrics

It’s often difficult for businesses without a matured data or machine learning practice to define and agree on metrics. Politics, personalities, and the tradeoff between short-term and long-term outcomes can all contribute to a lack of alignment. Many firms face a problem that’s even worse: no one knows which bars contribute to the metrics that impact business outcomes, or which metrics are important to the company( such as those reported to Wall street by publicly-traded corporations ). Rachel Thomas writes about these challenges in “The problem with metrics is a big problem for AI.” There isn’t a simple fix for these problems, but for brand-new companionships, investing early in understanding the company’s metrics ecosystem will pay dividends in the future.

The worst case scenario is when a business doesn’t have any metrics. In this case, the business probably got caught up in the hype about AI, but hasn’t done any of the readying.( Fair warning: if the business needs metrics, it probably too needs study about data infrastructure, accumulation, governance, and much more .) Work with senior management to design and align on appropriate metrics, and make sure that executive leadership concurs and consents to using them before starting your experiments and developing your AI products in earnest. Getting this kind of agreement is much easier said than done, peculiarly because a company that doesn’t have metrics may never have pictured gravely about what clears their business successful. It may require intense negotiation between different discords, each of which has its own procedures and its own political interests. As Jez Humble said in a Velocity Conference civilize period, “Metrics should be distressing: metrics should be able to obligate you modify what you’re doing.” Don’t expect agreement to come simply.

Lack of precision about metrics is technical pay merit compensating down. Without lucidity in metrics, it’s absurd to do meaningful experimentation.


A product manager needs to think about ethics-and urge the produce team to think about ethics-throughout the whole product development process, but it’s particularly important when you’re defining the problem. Is it a problem that must be resolved? How can the mixture be abused? Those are issues that every make crew needs to think about.

There’s a substantial literature about ethics, data, and AI, so rather than repeat that discussion, we’ll leave you with a few cases reserves. Ethics and Data Science is a short journal that helps developers think through data problems, and includes a checklist that crew members should revisit throughout the process. The Markkula Institute at the University of Santa Clara has an excellent list of resources, including an app to aid ethical decision-making. The Ethical OS also provides excellent an instrument for contemplation through the impact of technologies. And finally-build a team that includes parties of different backgrounds, and who will be affected by your concoctions in different ways. It’s surprising( and upsetting) how many ethical troubles could have been avoided if more beings “ve thought about” how the products would be used. AI is a powerful implement: give it for good.

Addressing the problem

Once you know which metrics are most important, and which levers affect them, you need to run experiments to be sure that the AI products you want to develop actually map to those business metrics.

Experiments stand AI PMs not only to exam presuppositions about the relevant and functionality of AI Concoction, but too to understand the effect( if any) of AI products on the business. AI PMs must ensure that experimentation pass during three phases of the product lifecycle 😛 TAGEND

Phase 1: ConceptDuring the concept phase, it’s important is required to determine whether it’s even possible for an AI product “intervention” to move an upstream business metric. Qualitative experiments, including experiment surveys and sociological studies, can be very useful here.For example, many companies use recommendation engines to boost marketings. But if your commodity is highly specialized, patrons may come to you knowing what they want, and a recommendation engine merely gets in the way. Experimentation should show you how your customers use your place, and whether a recommendation engine would help the business.Phase 2: Pre-deploymentIn the pre-deployment phase, it’s essential to ensure that certain metrics doorsteps are not flouted by the core functionality of the AI product. These measures are commonly referred to as guardrail metrics, and they rest assured that the product analytics aren’t leaving decision-makers the wrong signal about what’s actually important to the business.For example, a business metric for a rideshare firm might be to reduce pickup time per consumer; the guardrail metric might be to maximize trips per consumer. An AI product is likely to be reduce median getaway hour by sink solicits from customers in hard-to-reach orientations. Nonetheless, that activity should be reflected in negative business outcomes for the company overall, and eventually gradual adoption of the service. If this sounds fanciful, it’s not hard to find AI systems that made inappropriate actions since they are optimized a poorly thought-out metric. The guardrail metric is a check to ensure that an AI doesn’t make a “mistake.”When a measure becomes a target, it ceases to be a good meter( Goodhart’s Law ). Any metric can and will be abused. It is useful( and recreation) for the improvement team to brainstorm creative ways to game the metrics, and think about the unintended side-effects this has been possible to. The PM time needs to gather the team and request “Let’s think about how to abuse the pickup time metric.” Someone will inevitably come up with “To minimize pickup time, we could just drop all the goes to or from distant locations.” Then you can think about what guardrail metrics( or other means) you can use to keep the system working appropriately.Phase 3: Post-deploymentAfter deployment, the product needs to be instrumented to ensure that it continues to behave as expected, without harming other systems. Ongoing monitoring of critical metrics is yet another form of experimentation. AI performance tends to degrade over time as the environment reforms. You can’t stop watching metrics only because the product has been deployed.For example, an AI product that helps a garb producer understand which fabrics to buy will become stale as manners change. If the AI product is successful, it may even effect those varies. You must see when the model has become stale, and retrain it as necessary.

Fault Tolerant Versus Fault Intolerant AI Problems

AI product directors need to understand how sensitive their projection is to error. This isn’t always simple, since it doesn’t just take into account technical peril; it also has to account for social risk and reputational injure. As we mentioned in the first article of this succession, an AI application for product recommendations can make a lot of mistakes before anyone notices( dismissing very concerned about bias ); this has business impact, of course, but doesn’t cause life-threatening harm. On the other hand, an autonomous vehicle genuinely can’t afford to make any mistakes; even if the autonomous vehicle is safer than a human move, you( and your fellowship) will take the blame for any accidents.

Planning and managing the project

AI PMs have to determine tough picks when deciding where to apply limited resources. It’s the age-old “choose two” rule, where the parameters are Speed, Quality, and Features. For example, for a mobile phone app that uses objective detection to identify pets, fasted is crucial. A make manager may relinquish either a more diverse set of animals, or the accuracy of detection algorithms. These decisions have stunning implications on project length, resources, and goals.

Figure 1: The” select two” govern

Similarly, AI product managers often need to choose whether to prioritize the scale and blow of a product over the difficulty of product growth. Times ago a health and fitness engineering corporation realized that its content moderators, used to manually see and remediate offensive material on its stage, were experiencing extreme fatigue and very poor mental health outcomes. Even beyond the humane considerations, moderator burnout was a serious product issue, in that the company’s platform was rapidly growing, thus disclosing the average user to more potentially offensive or illegal content. The predicament of content moderation work was exacerbated by its repetition quality, uttering it potential candidates for automation via AI. However, the difficulty of developing a robust material moderation arrangement at the time was significant, and would have involved years of development time and research. Ultimately, the company decided to simply drop the most social components of the platform, a decision which restraint overall swelling. This tradeoff between repercussion and evolution difficulty is particularly pertinent for makes based on deep read: breakthroughs often lead to unique, defensible, and highly lucrative commodities, but investing in produces with a high chance of flop is an obvious threat. Makes based on deep memorize is very difficult( or even absurd) to develop; it’s a classic “high return versus high risk” situation, in which it is inherently difficult to calculate return on investment.

The final major tradeoff that AI product overseers must evaluate is how much time to expend during the course of its R& D and intend phases. With no restrictrictions on release years, PMs and engineers alike would choose to spend as much time as necessary to nail the concoction purposes. But in the real world, concoctions need to ship, and there’s rarely sufficient time to do the research necessary to ship the best possible product. Therefore, produce managers must make a judgment call about when to carry, and that call is usually based on imperfect experimental outcomes. It’s a balancing play, and admittedly, one that can be very tricky: achieving the product’s goals versus getting the product out there. As with traditional software, the best way to achieve your goals is to made something out there and iterate. This is particularly true for AI products. Microsoft, LinkedIn, and Airbnb has been particularly candid about their journeys towards building an experiment-driven culture and the technology required to support it. Some of the best lessons are captured in Ron Kohavi, Diane Tang, and Ya Xu’s book: Trustworthy Online Controlled Experiments: A Practical Guide to A/ B Testing.

The AI Product Development Process

The development chapters for an AI project map nearly 1:1 to the AI Product Pipeline we was reflected in the second article of this succession.

Figure 2: CRISP-DM were compatible with AI Pipeline

AI projects require a “feedback loop” in both the produce development process and the AI products themselves. Because AI products are inherently research-based, experimentation and iterative growth are necessary. Unlike traditional software exploitation, in which the inputs and results are often deterministic, the AI development cycle is probabilistic. This asks various important modifications to how projects are set up and executed, irrespective of the project management framework.

Understand the Customer and Objectives

Product managers must ensure that AI projects collect qualitative information about patron action. Because it might not be intuitive, it’s important has pointed out that traditional data measurement tools are most effective at calibrating magnitude than feeling. For most AI products, the make overseer will be less interested in the click-through rate( CTR) and other quantitative metrics than they are in the continued relevance of the AI product to the user. Therefore, traditional make research teams must engage with the AI team to ensure that the chasten insight is applied to AI product blooming, as AI practitioners are likely to lack the appropriate skills and experience. CTRs are easy to measure, but if you build a system designed to optimize these various kinds of metrics, you might find that the system sacrifices actual usefulness and user satisfaction. In this case , no matter how well the AI product contributes to such metrics, it’s output won’t ultimately provide the goals of the company.

It’s easy to focus on the wrong metric if you haven’t done the proper research. One mid-sized digital media company we interviewed reported that their Marketing, Advertising, Strategy, and Product teams formerly wanted to build an AI-driven user traffic forecast tool. The Marketing team improved the first framework, but because it was from marketing, the model optimized for CTR and guide alteration. The Advertising team was more interested in cost per make( CPL) and lifetime value( LTV ), while the Strategy team was aligned to corporate metrics( income jolt and total active useds ). As a solution, many of the tool’s consumers were dissatisfied, even if they are the AI served perfectly. The ultimate reaction was the development of multiple representations that optimize for different metrics, and the redesign of the tool so that it could present those yields clearly and intuitively to different kinds of users.

Internally, AI PMs must engage stakeholders to ensure alignment with the most important decision-makers and top-line business metrics. Put simply , no AI product will be successful if it never launches, and no AI product will propel unless the project is patronized, funded, and connected to important business objectives.

Data Exploration and Experimentation

This phase of an AI project is laborious and time consuming, but accomplishing it is one of the strongest shows of future success. A produce needs to balance the speculation of available resources against the risks of moving forward without a full understanding of the data landscape. Acquiring data is often difficult, especially in regulated industries. Once relevant data has been obtained, understanding what is valuable and what is simply noise necessitates statistical and technical rigour. AI product overseers probably won’t do studies and research themselves; their role is to guide data scientists, psychoanalysts, and domain experts towards a product-centric evaluation of the data, and to inform meaningful experiment design. The destination is to have a measurable signal for what data exists, solid penetrations into that data’s relevance, and a clear vision of where to concentrate efforts in designing features.

Data Wrangling and Feature Engineering

Data wrangling and piece engineering is the most difficult and important phase of every AI project. It’s generally accepted that, during a normal concoction growing round, 80% of a data scientist’s time is spent in feature engineering. Trend and tools in AutoML and Deep Learning have certainly abbreviated the time, knowledge, and effort required to build a prototype, if not an actual product. Nonetheless, improving a superior aspect pipe or model architecture will always be worthwhile. AI product overseers should make sure projection projects account for the time, struggle, and people needed.

Modeling and Evaluation

The modeling phase of an AI project is exasperating and difficult to predict. The process is inherently iterative, and some AI projects fail( for good reason) at this quality. It’s easy to understand what clears this stair difficult: there is rarely a feeling of steady progress towards a destination. You experiment until something runs; that might happen on the first day, or the hundredth daytime. An AI product director must cause the team members and stakeholders when there is no tangible “product” to show for everyone’s labor and investment. One approach for maintaining incitement is to push for short-term erupts to beat a carry-on baseline. Another would be to start multiple yarns( perhaps even multiple activities ), so that some will be able to demonstrate progress.


Unlike traditional software engineering projects, AI product overseers must be heavily to participate in the build process. Engineering overseers are usually responsible for offsetting sure all facets of a application concoction are properly compiled into binaries, and for organizing improve scripts meticulously by version to ensure reproducibility. Many full-grown DevOps processes and implements, honed over years of successful application produce releases, determine these processes more feasible, but they were developed for traditional application commodities. The equivalent tools and processes simply are not available in the ML/ AI ecosystem; when they do, they are rarely mature enough to use at flake. As a develop, AI PMs must take a high-touch, customized approach to run AI products through product, deployment, and release.


Like any other production software system, after an AI product is live it must be monitored. However, for the purposes of an AI product, both pattern recital and application operation must be monitored simultaneously. Notifies that are triggered when the AI product accomplishes out of specification may need to be routed differently; the in-place SRE team may not be able to diagnose technical issues with the prototype or data pipelines without support from the AI team.

Though it’s difficult to create the “perfect” project plan for monitoring, it’s important for AI PMs to ensure that project resources( specially engineering ability) aren’t immediately released when the make has been deployed. Unlike a traditional software commodity, it’s hard to define when an AI product has been deployed successfully. The development process is iterative, and it’s not over after the make has been deployed-though, post-deployment, the stakes are higher, and your options for dealing with issues are more limited. Therefore, members of the development team must remain on the upkeep team to ensure that there is proper instrumentation for logging and monitoring the product’s health, and to ensure that there are resources available to deal with the inevitable troubles that is an indication after deployment.( We announce this “debugging” to distinguish it from the evaluation and testing that takes place during make blooming. The final section in this series will be devoted to debugging .)

Among activities technologists, the idea of observability is gradually replacing monitoring. Monitoring requires you to predict the metrics you need to watch in advance. That clevernes is certainly important for AI products-we’ve talked all along about the importance of ensuring that metrics. Observability is critically different. Observability is the ability to get the information you need to understand why the system reacted the direction it does; it’s less about measuring known parts, and more about the ability to diagnose “unknown unknowns.”

Executing on an AI Product Roadmap

We’ve spent a lot of epoch talking about planning. Now let’s shift gears and discuss what’s needed to build a product. After all, that’s the point.

AI Product Interface Design

The AI product overseer must be a member of the design squad from the beginning, ensuring that the make provides the desired outcomes. It’s important to account for the ways a produce will be used. In the best AI products, users can’t tell how the underlying simulates impact their experience. They neither know or care that there is AI in the application. Take Stitch Fix, which helps a multitude of algorithmic approaches to provide customized vogue recommendations. When a Stitch Fix user interacts with its AI products, they is compatible with the projection and recommendation devices. The intelligence they treated with during that suffer is an AI product-but they neither know , nor attend, that AI is behind everything they consider. If the algorithm makes a perfect prediction, but the user can’t imagine wearing the items they’re shown, the commodity is still a failure. In reality, ML patterns are far from perfect, so it is even more imperative to fingernail the user experience.

To do so, produce managers is necessary to ensure that blueprint gets an equal seat at the table with engineering. Decorators are more attuned to qualitative research about customer action. What signals show user satisfaction? How do you construct commodities that satisfaction consumers? Apple’s sense of motif, clearing things that “just succeed, ” pioneered through the iPod, iPhone, and iPad products is the foundation of their business. That’s what you need, and there is a requirement that input from the start. Interface design isn’t an after-the-fact add-on.

Picking the Right Scope

“Creeping featurism” is a problem with any application commodity, but it’s a particularly dangerous problem for AI. Focus your make growing act on troubles that are relevant to the business and purchaser. A successful AI product measurably( and positively) influences metrics that matter to the business. Therefore, restraint the dimensions of an AI product to pieces that can create this impact.

To do so, start with a well-framed hypothesis that, upon validation through experimentation, will grow meaningful upshots. Doing this effectively means that AI PMs must learn to move business suspicions into produce increase implements and processes. For example, if the business seeks to understand more about its customer basi in order to maximize lifetime value for a due concoction, an AI PM would do well to understand the tools available for customer and product-mix segmentation, recommendation locomotives, and time-series forecasting. Then, when it comes to developing the AI product roadmap, the AI PM can focus engineering and AI units on the right ventures, the chasten outcomes, andthe smoothest path to production.

It is alluring to over-value the performance incomes achieved through the use of more complex modeling procedures, leading to the dreaded “black box” problem: patterns for which it’s difficult( if not impossible) to understand the relations between the input and the production. Black box examples are seldom beneficial in business environments for various reasonableness. First, being able to explain how the prototype manipulates is often a prerequisite for manager approbation. Ethical and regulatory considerations often require a detailed understanding of the data, descended pieces, grapevines and composing mechanisms involved in the AI system. Solving problems with the simplest model possible is always preferred, and not just because it leads to poses that are interpretable. In addition, simpler modeling approachings are more likely to be supported by a wide variety of frameworks, data platforms, and communications, increasing interoperability and decreasing technological debt.

Another scoping consideration concerns the processing engine that will power the product. Difficulties that are real-time( or near real-time) in quality can only be addressed by most performant stream processing buildings. Instances of this include product recommendations in e-commerce systems or AI-enabled messaging. Stream processing requires significant engineering try, and it’s important to account for that struggle at the beginning of development. Some machine learning approaches( and numerous software engineering patterns) are simply not appropriate for near-real time applications. If the problem at hand is more flexible and less interactive( such as offline churn likelihood projection ), quantity processing is probably a good approach, and is typically easier to integrate with the average data stack.

Prototypes and Data Product MVPs

Entrepreneurial product managers are often associated with the term “Move Fast and Break Things.” AI product mangers live and die by “Experiment Fast So You Don’t Break Things Later.” Take any social media company that sells advertisings. The timing, capacity, and type of ads exposed to segments of a company’s user population are overwhelmingly determined by algorithms. Patrons contract with the social media company for any particular chose budget, expecting to achieve specific gathering exposure thresholds that can be measured by relevant business metrics. The budget that is actually spent successfully is referred to as fulfillment, and is directly related to the revenue that each purchaser generates. Any change to the underlying simulates or data ecosystem, such as how sure-fire demographic peculiarities are weighted, can have a stunning impact on the social media company’s revenue. Experimenting with brand-new simulations is essential-but so is yanking an underperforming framework out of production. This is only one example of why rapid prototyping is important for crews improving AI products. AI PMs must create an environment in which continuous experimentation and failure are stood( even celebrated ), along with supporting the processes and implements that enable experimentation and learning through failure.

In a previous segment, we introduced the importance of user research and interface design. Qualitative data collection tools( such as SurveyMonkey, Qualtrics, and Google Forms) should be joined with boundary prototyping tools( such as Invision and Balsamiq ), and with data prototyping implements( such as Jupyter Notebooks) to formation an ecosystem for produce exploitation and testing.

Once such an environment exists, it’s important for the product administrator to codify what constituting an “minimum viable” AI product( MVP ). This product ought to be robust enough to be used for user the studies and quantitative( sit evaluation) experimentation, but simple enough that it can be quickly jettisoned or adjusted in favor of brand-new iterations. And, while the word “minimum” is important, don’t forget “viable.” An MVP needs to be a product that can stand on its own, something that customers will crave and use. If the produce isn’t “viable”( i.e ., if a used wouldn’t want it) you won’t be able to conduct good user study. Again, it’s important to listen to data scientists, data designers, software developers, and pattern team members when deciding on the MVP.

Data Quality and Standardization

In most makings, Data Quality is either an engineering or IT problem; it is rarely addressed by the product team until it blocks a downstream process or project. This relationship is impossible for crews developing AI products. “Garbage in, debris out” holds true for AI, so good AI PMs must concern themselves with data health.

There are many excellent reserves on data quality and data governance. The specifics are outside the scope of this article, but here are some core principles that should be included in any produce manager’s toolkit 😛 TAGEND

Beware of “data cleaning” approaches that mar your data. It’s not data cleaning if it deepens the core properties of the underlying data. Look for quirks in your data( for example, data from legacy organizations that truncate text studies to save opening ). Understand the risks of bad downstream standardization when scheduling and implementing data collection( e.g. arbitrary staunch, stop statement removal .). Ensure data stores, key grapevines, and inquiries are properly documented, with structured metadata and a well-understood data flow.Consider how duration wallops your data assets, as well as seasonal gists and other biases.Understand that data bias and artifacts can be introduced by UX picks and examination designing.

Augmenting AI Product Management with Technical Leadership

There is no intuitive space to foresee what will work best in AI product progress. AI PMs can build amazing things, but this often comes predominantly from the right frameworks rather than the chastise tactical activities. Many new tech capabilities have the potential to enable software engineering abusing ML/ AI techniques more quickly and accurately. AI PMs will need to leveraging newly emerging AI techniques( idol upscaling, synthetic text generation using adversarial systems, buttres memorize, and more ), and partner with expert technologists to set these implements to use.

It’s unlikely that every AI PM will have world-class technical thought in addition to excellent product sense, UI/ X knowledge, customer learning, lead abilities, and so on. But don’t gave that procreate despair. Since one person can’t be an expert at everything, AI PMs it is necessary structure a partnership with a engineering chairwoman( e.g ., a Technical Leador Lead Scientist) who are familiar with the state of the arts and is familiar with current investigate, and trust that tech leader’s educated intuition.

Finding this critical technical collaborator can be difficult, especially in today’s competitive knack busines. However, all is not lost: there are many excellent technical make presidents out there posing as skillful engineering managers.

Product manager Matt Brandwein advocates seeing what potential tech induces do in their idle period, and taken due note of which domains they find handsome. Someone’s current role often doesn’t reveal where their interests and expertise lie. Most importantly, the AI PM should look for a tech result who is able mitigate their own weaknesses. For example, if the AI PM is a utopian, picking a technological lead with operational ordeal is a good idea.

Testing ML/ AI Product

When a commodity is ready to ship, the PM will work with user research and engineering teams to develop a handout propose that obtains both qualitative and quantitative consumer feedback. The volume of this data will be concentrated on user interaction with the user interface and front end of the product. AI PMs is required to plan to collect data about the “hidden” functionality of the AI product, the percentage no used ever determines instantly: pattern recital. We’ve discussed the need for proper instrumentation at both the pose and business status to gauge the product’s effectiveness; this is where all of that projecting and hard work pays off!

On the modeling slope, operation metrics that were validated during improvement( predictive ability, simulation fit, precision) must be constantly re-evaluated as the pattern is exposed to more and more unseen data. A/ B testing, which is frequently used in web-based software development, is useful for evaluating model performance in product. Most firms already have a framework for A/ B testing in their release process, but some may need to invest in testing infrastructure. Such assets are well worth it.

It’s inescapable that the pose will require adjustments over meter, so AI PMs is necessary to ensure that whoever is responsible for the commodity post-launch has access to the development team in order to investigate and resolve issues. Now, A/ B testing has another benefit: the ability to run champion/ challenger pose evaluations. This framework allows for a deployed prototype to run uninterrupted, while a second model is evaluated against a test of the entire population. If the second model outshines the original, it can simply be swapped out-often without any downtime!

Overall, AI PMs should remain closely involved in the early release lifecycle for AI products, taking responsibility for coordinating and organizing A/ B tests and user data collection, and resolving issues with the product’s functionality.


In this article, we’ve focused primarily on the AI product development process, and planning the AI product manager’s responsibilities to each stage of that process. As with many other digital concoction progress cycles/seconds, AI PMs must first ensure that the problem be resolved is both a number of problems that ML/ AI can solve and a number of problems that is vital to the business. Once this criteria has been met, the AI PM has to determine whether the concoction should be developed, considers the myriad of technical and ethical considerations at movement when developing and releasing a yield AI system.

We propose the AI Product Development Process as a blueprint for AI PMs of all industries, who may develop myriad different AI products. Though this process is by no means exhaustive, it emphasizes the kind of critical thinking and cross-departmental collaboration necessary to success at each stage of the AI product lifecycle. However, regardless of the process you use, experimentation is the key to success. We’ve said that repeatedly, and we aren’t tired: the more experiments you can do, the more likely you are to build a product that works( i.e ., positively impacts metrics the company cares about ). And don’t forget qualitative metrics that help you understand user behavior!

Once an AI system is exhausted and in use, nonetheless, the AI PM has a somewhat unique persona in commodity upkeep. Unlike PMs for many other software concoctions, AI PMs is necessary to ensure that robust testing frames are constructed and utilized not only during the development process, but also in post-production. Our next article concentrates on perhaps the most important phase of the AI product lifecycle: maintenance and debugging.

Read more: