What does an AI Growth PM do?
🌱 Field notes: How AI Growth PMs disproportionately transform a product's growth
You’ve heard of Product Managers, Growth Product Managers, and maybe even Product Marketing Managers. But what the heck is an AI Growth Product Manager?
In my humble, not-at-all biased opinion, it’s one of the most interesting roles in tech. And one of the most in-demand too.
In this post, I’ll break down:
Part 1: What’s an AI Growth PM? And how are they different from the more commonly known Growth PM?
Part 2: What levers does the AI PM pull to grow products at a massive scale?
Part 3: Why it’s one of the most highly demanded roles in tech and why it’s here to stay
I’ve been lucky enough to be an AI Growth PM for about a year now, and it’s been a fascinating ride. I’m learning new things every day from some of the best mentors/managers in the field. These are the notes and observations I’ve gathered along the way.
But first, 🗣️ DISCLAIMER 🗣️ Like any role in fast-moving, dynamic world of tech, the exact title, scope, and day-to-day of an AIGPM can look different from one company to another. Especially with AI, the lines are getting blurred. What I’m about to break down in Part 2 is a framework of strategic levers this role often pulls. Think of it as a playbook that gets adapted to each company. The core idea is that someone (anyone! EVERYONE!) needs to be thinking about these levers strategically to drive AI-powered growth.
Part 1: What exactly is an AI Growth PM?
We’ve got:
The Classic Product Manager (PM): This is often the role people think of first when they hear of PM. They’re the strategic minds behind the product, responsible for defining the product vision (where are we going?), crafting the strategy (how will we get there?), and building the roadmap (what steps will we take?). They immerse themselves in user and market research, leading the team from initial idea through discovery, development, and ultimately, getting the product out the door. They own the ‘why’ and ‘what’ of the product.
The Product Marketing Manager (PMM): Once a product is ready (or nearing readiness), the PMM is point for taking the product to market. Based on their understanding of the target audience and competitive landscape, they craft the messaging and positioning, develop the go-to-market strategy, and create campaigns and materials that drive awareness and adoption.
The Growth Product Manager (Growth PM): While a Classic PM might focus on building new features or entire products, the Growth PM is focused on driving specific business metrics or optimizing product funnels. They constantly look for opportunities to improve things like user acquisition, activation, retention, and revenue. Their toolkit consists of experimentation, and they’re all about making iterative improvements to move the needle on KPIs.
And now…
The AI Growth Product Manager (AI Growth PM or AIGPM as I’ll refer to from now on): Just like the Growth PM role emerged from a specific business need (faster, focused growth), the AI Growth PM is born from the scale and efficiency of AI.
So, what makes them different?
AI Growth PMs share the same core goal as Growth PMs: improve crucial business metrics. But they do it by leveraging AI. They’re driving the requirements and strategy for optimizing the AI models to drive growth in a way that's not only effective in the short-term but also sustainable and incredibly efficient1 in the long run.
Part 2: The AI Growth PM’s Playbook
An AI Growth PM's job isn't just about knowing AI exists; it's about strategically using different aspects of an AI system to hit growth targets. Each "lever" they pull offers a distinct way to improve things:
Problem Framing & Objective Definition: What are we solving, and how do we measure success?
Data & Signals: What raw information powers the model?
Models & Algorithms: How does the system learn to make decisions?
Targeting & Personalization: Who sees what, and when, and where?
UX & Delivery: How are predictions presented and experienced?
Infrastructure & ML Ops: How does it scale, serve, and stay reliable?
1. Problem Framing & Objective Definition
No model matters unless it solves something meaningful. This lever is about defining clear, solvable AI problems that align with both user needs and business goals. It's where data science meets strategy – and where the AI Growth PM shines as the translator between technical potential and commercial value.
Key Components
Business goal alignment: Everything starts with the outcome. AIGPMs will initiate a project by clarifying what success looks like in business terms: higher conversion, increased engagement, improved retention, increased LTV, etc.
AI problem framing: Next, AIGPMs (shortening for my sake) will define the AI problem clearly. They’ll ask: Are you building a model to recommend, rank, classify, or generate? Is this a prediction problem? A retrieval problem? A content generation problem? Framing it incorrectly will result in vastly different solution that might miss the mark on KPIs.
Model metric mapping: Once they’ve framed the problem, AIGPMs in partnership with their technical engineering/data science counterpart set the metric. Based on the business KPI they care about, they’ll select an evaluation metrics (e.g. accuracy, precision, loss, recall, NDCG, AUC).
Proxy design & validation: In cases where business outcomes are hard to measure directly, they’ll define proxies with strong downstream correlation. As an example, "likelihood to watch full video" would be a proxy for content relevance.
Experimentation readiness: While the team goes into model development mode, AIGPMs will think ahead about how they’ll measure success in the real world through A/B or multivariate testing. What are the primary vs. secondary metrics? What’s the baseline? They’ll define this upfront.
Stakeholder & partner alignment: Throughout this entire process, AIGPMs will collaborate early with data science, product, engineering, and analytics to scope feasibility, timeline, and risks and partner on defining some of the requirements related to modeling and experimentation.
AI Growth PM Role & Responsibilities
✅ Identify high-impact, model-worthy problems tied to KPIs (growth, retention, monetization)
✅ Translate business needs into clear AI problem types and data requirements
✅ Define success metrics – not just for the model, but for the product and customer
✅ Bridge conversations between strategic teams and technical teams to align on value and approach
✅ Ensure downstream evaluation and experimentation plans are in place before launch
Common Pitfalls to Avoid
❌ Assuming a high model accuracy (e.g., 90% AUC) automatically means business success. If that accuracy doesn't translate to improved KPIs, the model isn't truly working.
❌ Building a complex AI solution without first understanding how a simpler heuristic or the current system performs. You need a baseline to prove AI's incremental value.
❌ Launching an AI feature without clearly defining upfront what “good” looks like in measurable terms.
❌ Underestimating the ‘last mile’ integration and not considering how the model will integrate into the existing product.
2. Data & Signals
Data is the fuel of models. The quality, quantity, diversity, relevance, and freshness of data directly determine how well models perform. Improving your dataset often yields greater returns than tweaking algorithms. For AI Growth PMs, this means becoming obsessed with the completeness, structure, and utility of the signals feeding the model.
Key Components
Signal expansion: AIGPMs are always on the hunt for data (1P or 3P). When it comes to 1P, they’ll prioritize capturing a broader range of user behaviors and contextual triggers, such as clickstreams, hover time, scroll depth, or feature usage (more signals = richer understanding).
Signal quality: On top of that, they’ll ensure the signals are clean, unbiased, and complete. That includes dealing with sparsity, noise, or skew in datasets that can lead to misleading model outcomes.
Signal representation: Once they have the data, AIGPMs will work with data engineers (sometimes data scientists or machine learning engineers) to transform raw behaviors into useful, model-ready features (e.g. creating sequence modeling of click history, generating user/item embeddings from behavioral data).
Feedback as signals: Thinking holistically, AIGPMs will design systems to capture and utilize user interactions to continuously improve data quality and model relevance. This includes explicit feedback (e.g. direct user inputs like likes, ratings, thumbs up/down) and implicit feedback (e.g. clicks, hovers, time spent, scrolls).
AI Growth PM Role & Responsibilities
✅ Champion richer instrumentation and signal coverage across the product
✅ Build a data roadmap that includes initiatives like instrumenting new user events, integrating streaming clickstream data, or third-party enrichments
✅ Partner with product, engineering, data science, and analytics on feedback ingestion as reliable, retrainable signal sources for model improvement
Common Pitfalls to Avoid
❌ Assuming the necessary data for your AI idea will just "be there”. Data strategy and instrumentation often require significant lead time.
❌ Ignoring data freshness and using stale data. Model performance can degrade quickly if data inputs aren't reliably maintained.
❌ Overlooking potential biases in data sources, which can lead to skewed model outcomes.
3. Models & Algorithms
If data is the fuel, then the algorithms are the engine that processes it. This lever about selecting, building, and refining the right model to maximize predictive power for the job at hand.
Key Components
Model selection & architecture: Typically, data scientists select the appropriate algorithm based on the problem. Where AIGPMs help weigh in are on the product requirements such as the degree of explainability needed per use case as well as trade-offs between performance, explainability, development time, and other factors.
Models in an AIGPM’s toolkit:
Foundational models:
Affinity/Propensity models: These predict the likelihood of a user to take a specific action (e.g. engage with content, buy a product, click an ad.)
Embedding models: Like ‘fingerprints’, these are compact representations that capture complex relationships and similarities for users or items. They serve as inputs for downstream ML models and are pretty powerful.
Advanced models:
LLM-based content generation: Large language models can be used to auto-generate metadata or product descriptions. This can save a ton of manual labeling efforts and enable personalization at scale.
AI Growth PM Role & Responsibilities
✅ Collaborate with data scientists on model selection, iteration, and evaluation, ensuring alignment with business objectives
✅ Build a roadmap for developing and improving generalizable, foundational models (e.g. affinity, embedding models) that can serve multiple use cases
✅ Identify opportunities to leverage advanced techniques like LLMs to solve content gaps and power personalized experiences at scale
Common Pitfalls to Avoid
❌ Pushing for the most complex, cutting-edge model if a simpler, more interpretable one can achieve 80-90% of the desired outcome with less effort and risk.
❌ Blindly accepting the algorithm selection because your data scientist chose it. Instead, ask the data scientists to explain (at a high level) how the model works, its assumptions, and limitations, making sure the right trade-offs are being made.
❌ Forgetting that models degrade over time and failing to plan for ongoing monitoring and retraining
4. Targeting & Personalization
In a world saturated with content and information, relevance is king. When experiences for tailor-made, users are dramatically more likely to engage, convert, remain loyal, and ultimately drive the growth metrics AIGPMs care about. This lever is about moving beyond the one-size-fits-all experience and delivering the most relevant content to the right user segments at the right moment in their journey and through the most effective channel.
Key Components
Audience segmentation (who): AIGPMs will work with marketing and analytics to identify well-defined groups personalize for. Some examples of how audiences can be segmented:
Behavior-based: Grouping users by real-time actions or patterns (e.g. ‘cart abandoners’, ‘repeat browsers’, ‘inactive trialers’)
Predictive: Forecasting likely outcomes to identify users ‘at risk’ or ‘ready to convert’
Content matching (what): While AIGPMs aren’t building the matching algorithms (that’s what the data scientists will do!), they are responsible for defining what needs to be matched. Here, content can include things like articles, messages, offers, images, and much more. These will often be tailored to the user’s preferences and histories (via affinity models, embeddings, etc.)
Journey orchestration (when & where): Given what's happening right now (location, device, current activity, time of day), what is the most relevant thing to show? AIGPMs are striving to achieve a specific growth goal. Journey orchestration is one way to do that. AIGPMs will consider how to optimize the sequence and timing of interactions across the user’s journey to achieve some goal like an onboarding sequence to activate new users or an in-app step-flow to encourage adoption of a new feature.
Contextual targeting (when & where): Closely related to journey orchestration, contextual targeting involves more short-term "right time, right place" delivery. Based on the user’s immediate/real-time situation or environment like device, location, session context, or time of day, AIGPMs will focus on optimizing a specific interaction or piece of content.
Explore/Exploit strategy: One of the most interesting (and imo life-transferrable) components is setting the explore/exploit strategy. AIGPMs consider the audience segment, product goals, and surface to balance safe bets with discovery by introducing diversity or novelty to avoid filter bubbles (where users see more of the same).
Feedback loops: This was touched on earlier as “Feedback as signals”. Here, AIGPMs ensure real-time implicit and explicit feedback (likes, skips, scrolls, “not interested”) are leveraged to enrich personalization and guide adjustments in what content gets shown.
Multi-objective optimization: For some more mature, complex products, AIGPMs will be involved with setting multiple objectives since real-world product decisions often involve multiple goals that need to be balanced. One example is deciding whether to optimize for revenue or conversion. Optimizing solely for revenue might lead to pushing expensive items that don't convert well, hurting conversion rate. On the flip side, optimizing solely for conversion might lead to promoting cheap items, hurting revenue and margin. It’s a balance. AIGPMs need to get stakeholders to align on the priorities and trade-offs, define measurable metrics for each objective, and ensure that business optimization doesn’t degrade the user experience to a point where it harms long-term growth (like retention or trust).
AI Growth PM Role & Responsibilities
✅ Identify and prioritize high-leverage personalization opportunities across the funnel to drive growth
✅ Define segmentation logic and targeting strategies in partnership with data science and analytics
✅ Collaborate closely across product, design, and engineering to activate adaptive experiences with feedback loops
✅ Advocate for personalization that earns user confidence through relevance, control, and transparency
Common Pitfalls to Avoid
❌ Ignoring the ‘Cold Start’ problem and forgetting to plan how personalization will work for new users (no history) or new content (no interaction data).
❌ Assuming a single 'one-size-fits-all’ personalization strategy works for all user segments or all parts of the product. Nuance and context is critical.
❌ Implementing personalization that feels overly invasive or reveals data they didn't realize was being used.
❌ Building personalization systems without considering ways for users to provide feedback, adjust preferences, or understand why they are seeing certain things.
5. UX & Delivery
Even the smartest prediction falls flat if users don’t understand or trust it. This lever covers the experience layer of AI—presentation, performance, and explainability—to make intelligent experiences feel intuitive, credible, and seamless.
I’m actually quite passionate about this space being a visual person that I am and having done research in Human-AI computer interaction. But I’ll bite my tongue here and keep it brief.
Key Components
Presentation layer: Sometimes referred to as the ‘surface’, this is where users see the AI-driven content.
Interaction design: AIGPMs make sure their AI outputs are actionable and navigable with minimal friction on the surface they’re presented on.
Explainability & trust elements: Not captured in the targeting section is the ‘Why’ or the ‘Why am I seeing this?” messaging. To increase user trust and adoption, AIGPMs need to determine the degree of explainability needed for the use case and work with content or experience designers to bake that in via messaging or UI cues.
Latency management: One trade-off that AIGPMs might need to weigh is speed of delivery vs. model complexity. They’ll work with the product development or product engineering team to implement techniques like caching, async loading, and fallback logic to limit latency.
Resilience & fallback logic: This ones pretty straightforward. What happens when AI fails? AIGPMs need to author backup rules or fallback content, of course not in isolation but working with their team.
Orchestration layer: This is a meaty one. I think of this layer as the final set of decisions and logic that determine what content actually makes it to that presentation layer and in what order, for a specific user at a specific moment. Interesting things get set here including diversity rules, impression capping, boosters/penalties (positive or negative multipliers added to content, you might see these added around the holiday season), model switching logic, and A/B test delivery infrastructure.
AI Growth PM Role & Responsibilities
✅ Define end-to-end experience of personalized features in partnership with designers
✅ Collaborate with engineers on real-time performance tuning and resilient system behavior
✅ Own orchestration strategy: prioritization logic, boosters/penalties, model switching
✅ Ensure feature rollout is testable, measurable, and aligned with experience goals
Common Pitfalls to Avoid
❌ Assuming users will understand or trust AI outputs just because they are "smart." The UI/UX needs to make them intuitive and valuable.
❌ Ignoring latency and underestimating how slow AI responses can kill the user experience and lead to drop-off.
❌ Launching AI features without robust fallback mechanisms for when the AI is wrong, data is missing, or the system fails.
❌ Underinvesting in explainability and assuming a complex model can't have some form of user-facing explanation
6. Infrastructure & ML Ops
All the data and models in the world don’t matter if they can’t reliably get into production. This lever covers the systems that support model development, deployment, experimentation, and delivery at scale.
AIGPMs will collaborate frequently with platform teams to enhance foundational infrastructure to serve their use cases:
Key Components
Feature store: These are centralized systems for storing, sharing, and querying reusable ML features across teams and workflows
Model serving & deployment: Infrastructure for delivering model predictions into real products, reliably and at scale. This should support batch, real-time, and streaming inference along with safe rollout practices like versioning and canary releases.
Monitoring & observability: These are tools that track model health in production, from performance drift and prediction anomalies to latency and throughput issues.
Experimentation infrastructure: This includes end-to-end tooling for running A/B tests on models or algorithm variants. Some features that AIGPMs are particularly interested in are traffic splitting, gating, and metric logging to evaluate both model performance (e.g. accuracy, loss) and product impact (e.g. conversions, retention).
Training-serving consistency: One thing my architect heavily pushed for was architecture that ensures features and preprocessing logic used during training are preserved during inference. The benefit of this is ensuring the model behavior in production matches what was validated offline » Important for serving relevant recommendations!
AI Growth PM Role & Responsibilities
✅ Advocate for infrastructure investments that accelerates model-to-production cycles without sacrificing quality or control
✅ Balance long-term infra roadmaps (e.g. retraining pipelines) against short-term product impact via customer-facing use cases
✅ Translate cross-functional needs into infrastructure priorities that unlock developer velocity and business value
Common Pitfalls to Avoid
❌ Assuming moving a model from a data scientist's notebook to production is trivial. It involves significant engineering. I learned this lesson the hard way.
❌ Simply ‘making do’ with current infrastructure. AIGPMs, being closest to the business use case, have a responsibility to articulate specific platform requirements needed to support and scale their AI initiatives.
Each of these levers is a world unto itself. Great AI Growth PMs blend technical fluency, business savvy, and product intuition to judiciously prioritize short- and long-term investments across each pillar. They know which levers to pull, and when, to create a kind of self-improving "personalization flywheel" that keeps getting better. Naturally, I’d expect the specific way these levers are owned and operated will look different depending on the company, but the core challenges and opportunities remain the same.
Part 3: Why it’s one of the most highly demanded roles in tech (and why it’s here to stay!)
Since Youtube popularized the term ‘creator’ in 2011, the creator economy has grown into a $250 billion industry. Platforms like Substack, Spotify, and TikTok have built their own ecosystems where writers, podcasters, hobbyists, and quite literally anyone can produce and distribute content directly to their audiences.
The pace of growth has only accelerated. Generative AI tools like ChatGPT, Midjourney, and Suno.AI have reduced the time required to create content by orders of magnitude. Content that once took hours or days to produce can now be generated in seconds. With these additional tailwinds to an organically growing industry post-COVID, it’s estimated the creator economy could reach +$480 billion by 2027.
When content creation becomes nearly frictionless, the bottleneck shifts to discovering and recommending the right content to the right person.
Enter: Recommendation systems 🎭
During the 1800s Gold Rush, the people who built lasting wealth were the ones selling shovels rather than the prospectors themselves. In the AI surge, a similar pattern emerged: companies like Nvidia, which sells GPUs to train large-scale ML models, became foundational to the entire industry building large language models.
In today’s creator-led digital economy, recommendation systems have taken on that same infrastructural role. They power the discovery, engagement, and retention loops across creator platforms, influencing not just what consumers see but also which creators take off.
Who’s responsible for growing and scaling this? 🤔
Enter: AI Growth PM 🧍🏻♀️
As content volume grows, someone needs to strategically connect the supply with the demand at scale. AI Growth PMs fill this role, defining experiment strategy, optimizing the model, and ensuring that the content users see aligns with both relevance and business impact. They guide AI systems to understand what users value, enabling businesses to translate those insights into sustainable growth.
Humans will always create and always seek connection. Amidst the digital deluge of content, the AI Growth PM ensures technology helps us find the people and ideas we resonate with most.