The Product Manager's Guide to AI Product Strategy

A common misconception about AI product strategy holds teams back: the belief that you need an AI-first approach to compete in today’s market.

Product teams contort themselves trying to “add AI” to their roadmap because investors asked about it, competitors announced it. The result is usually features that feel bolted on, deliver marginal value, and consume disproportionate engineering resources.

Here’s the my take: AI product strategy isn’t about AI at all. As always, it’s about solving customer problems effectively. Sometimes AI is the best tool for that. Often, it isn’t.

Let me show you how to think about this.

Product Applications

Use Cases Worth Pursuing

The first question isn’t “where can we use AI?” but “where are our customers struggling with tasks that are repetitive, data-intensive, or require pattern recognition at scale?”

Those are the sweet spots. AI excels when:

Volume overwhelms humans: Reviewing thousands of support tickets, processing millions of data points, or monitoring streams of real-time information
Patterns hide in complexity: Fraud detection, recommendation systems, or predictive maintenance where signals are buried in noise
Personalisation matters at scale: Tailoring experiences for millions of users in ways that would be impossible manually
Speed creates competitive advantage: Real-time translation, instant content moderation, or dynamic pricing where milliseconds matter

Where AI typically disappoints:

Tasks requiring genuine creativity or novel problem-solving
Decisions with high stakes and low tolerance for error
Situations where explainability matters more than accuracy
Problems where you don’t have enough quality data to train or fine-tune models

Integration Approaches

When you do identify legitimate AI opportunities, you’ve got three integration patterns to consider:

AI as feature: A specific capability within your existing product. Think Grammarly’s writing suggestions or Spotify’s Discover Weekly. Users interact with AI explicitly but within a familiar product context.

AI as enhancement: Invisible improvements to existing experiences. Search that understands intent better, feeds that surface more relevant content, or interfaces that adapt to user behaviour. Users benefit without knowing AI is involved.

AI as product: The entire value proposition centres on AI capabilities. Tools like Gemini or ChatGPT fall here. This approach requires deep AI expertise and carries higher technical risk.

“The best AI products don’t feel like AI products. They feel like magic that just works.”

Most product teams should start with AI as feature or enhancement. These approaches let you learn incrementally, contain risk, and build internal capability before attempting more ambitious bets.

Technology Overview

Current State

The AI technology landscape has shifted (and, of course, shifts all the time) dramatically. What required dedicated ML teams and months of development can now be accomplished in weeks using APIs and pre-trained models.

Key capabilities available today:

Large Language Models (GPT, Claude, Gemini) enable natural language understanding, generation, and reasoning through simple API calls
Embedding models power semantic search, similarity matching, and recommendation systems without custom training
Vision models handle image classification, object detection, and image generation with remarkable accuracy
Speech models deliver accurate transcription and natural-sounding synthesis in dozens of languages
Multimodal models combine text, image, and audio understanding in single systems

The barrier to entry has collapsed. The challenge has shifted from “can we build this?” to “should we build this?” and “can we do it better than competitors who have access to the same tools?”

Key Capabilities to Evaluate

When assessing AI technologies for your product, focus on:

Accuracy vs. cost trade-offs: More capable models cost more per request. Understanding your tolerance for errors helps you choose appropriately.

Latency requirements: Real-time applications need fast responses. Batch processing can tolerate slower, more thorough analysis.

Data privacy implications: Where does data flow? Is it used for model training? What are the compliance implications for your industry?

Customisation options: Can you fine-tune models on your data? How much improvement does that provide?

Vendor lock-in risks: How difficult is it to switch providers if pricing changes or capabilities evolve?

Future Implications

Trends to Watch

The AI landscape continues to evolve rapidly. Here’s what I’m paying attention to:

Commoditisation of basic capabilities: Natural language processing, image recognition, and similar capabilities are becoming table stakes. Competitive advantage will come from how you apply these tools, not from having access to them.

On-device AI: Processing moving from cloud to edge devices enables new use cases around privacy, latency, and offline functionality. Apple’s approach to on-device intelligence is worth studying.

Smaller, specialised models: Not every problem needs thinking mode on. Also, some purpose-built models that are faster and cheaper can be more accurate for specific domains.

AI agents: Systems that can plan and execute multi-step tasks autonomously represent the next frontier. Early examples are rough, but the trajectory is clear.

Regulation and liability: As AI becomes more consequential, legal and regulatory frameworks are developing. Understanding these constraints will become part of product strategy.

Preparing Your Team

Building organisational readiness for AI is as important as technical capability:

Invest in AI literacy broadly. Everyone on the product team should understand what AI can and cannot do. This prevents both unrealistic expectations and missed opportunities.

Develop evaluation frameworks. Create processes for assessing AI opportunities that balance potential value, technical feasibility, and strategic fit.

Build experimentation capacity. AI initiatives benefit from rapid prototyping and testing. Make it easy to try things quickly and learn from results.

Establish ethical guidelines. Determine your principles around bias, privacy, transparency, and safety before you’re under pressure to ship.

Create feedback loops. AI systems improve with data. Design your products to capture signals that help you refine and improve over time.

Key Takeaways

AI product strategy should start with customer problems, not with “how can we use AI?”
Focus AI efforts on tasks involving high volume, complex patterns, scale personalisation, or real-time processing
Start with AI as feature or enhancement rather than building AI-first products until you’ve developed capability
The barrier to AI implementation has collapsed, shifting the challenge from “can we?” to “should we?”
Prepare your team with broad AI literacy, clear evaluation frameworks, and established ethical guidelines

Getting Started This Week

Here’s a practical exercise: Map your product’s core workflows and identify where users face friction involving repetitive tasks, pattern recognition, or data processing at scale.

Pick the single most promising opportunity. Spend thirty minutes exploring whether existing AI APIs could address it. Don’t build anything yet—just research.

Then ask yourself: if this AI capability worked perfectly, how much would it improve the customer experience? Is that improvement worth the complexity, cost, and ongoing maintenance?

That’s the foundation of good AI product strategy. Start with value, not technology.

Have questions or thoughts? Get in touch - I’d love to hear from you!