AI product strategy: how Intercom went all in and what they learned

Executive overview

Most product teams treat AI as an add-on feature. The real question is whether AI can do what your product does — and if so, your strategy needs to change from first principles. Customer support is the clearest early case, but the same logic applies to any product with workflows or multimedia.

Map what AI can do against what your product does, then decide: replacement, augmentation, or irrelevant.

Failure as a product practice

• Paul froze and walked off stage mid-keynote at Cannes — the lesson: you survive, adapt, move on.
• Google Buzz and Google+ failed because they came from competitive fear, not genuine user insight.
• Research consistently showed people needed better small-group communication tools; the products built around that insight still failed because the motivation was wrong.
• Ship to learn (now: ship fast, ship early, ship often) is Intercom's core principle — it encodes the expectation that things will go wrong.
• Tension exists between high craft standards and fast shipping; navigating it is an ongoing practice, not a solved problem.

How to think about AI and your product

• Start with the core premise of your product: what problem does it solve, why do people use it?
• Ask: can AI do that? Fully, partially, or not at all?
• Map your product's functions against AI's current capabilities: writing, summarising, answering queries, finding facts, scanning images, taking actions, reasoning, writing code.
• Almost all B2B SaaS products with workflows or multimedia are in the path.
• For some functions, AI replaces; for others, it augments — both are valid outcomes.
• The reporting product of the future may be a chat box querying your database directly.

Intercom's AI pivot: Finn

• ChatGPT's launch was Intercom's "before/after" moment — they ripped up their strategy and rebuilt from first principles.
• Sam Altman named customer service as one of the first industries AI would disrupt; Intercom's core product was exactly that.
• Finn is an AI-first chatbot that acts as the first line of defence for customer support, before a human agent.
• Some customers resolve 50–70% of inbound queries with Finn.
• The harder challenge isn't the technology — it's helping customer support teams rethink their roles and org structures.
• Finn also operates inside the Intercom inbox, suggesting answers to human reps and rephrasing responses.

What changes when you go all in on AI

• Invest in ML engineers first — building on top of foundation models (OpenAI, Anthropic) requires real ML capability.
• Don't bolt AI on as a side team; integrate it across every product team.
• Prefer generalists who can learn new things over deep specialists — specialists thrive in large stable companies but can struggle in fast-moving, ambiguous environments.
• The ML team handles foundational work; product teams build on top of it once the layer is stable.
• Expect the PM-to-engineer ratio and the nature of engineering work to change as AI writes more code.

Avoiding the pitfalls

• Stay current: read constantly, follow people on X, subscribe to relevant newsletters (e.g. Matt Rickard, OpenAI blog).
• Block dedicated time — treat learning about AI like any other priority, not a nice-to-have.
• Deliberately read the skeptical view; balancing optimism with dissenting opinions leads to better judgment.
• Don't be afraid: job loss scenarios are often overstated; high-attrition roles like customer support are more likely to see headcount reductions through natural attrition than mass layoffs.
• The hardest part is navigating internal ambiguity — conviction needs to be earned through evidence and examples, not mandated.

Differentiation vs table stakes

• Differentiation: what is different and better in ways customers care about — the attraction of the new solution.
• Table stakes: baseline features required to compete; boring, easy to ignore, expensive to catch up on later.
• Intercom spent early years over-indexing on differentiation; customers loved it but couldn't switch because basic reporting and permissions were missing.
• In an established category, newcomers need significant differentiation to offset years of accumulated table stakes from incumbents.
• As a company matures, the balance shifts — Intercom now runs roughly 50/50.
• Use the framework to audit your roadmap: which of these two things matters more to our customers right now?

Swinging the pendulum

• When something is in an undesirable state, fixing it creates the temptation to overcorrect.
• Intercom swung from too much differentiation to too much table stakes focus — multiple times.
• Hiring "people who've done it before" at scale brought the culture of other companies, not Intercom's; the correction was to hire adaptable generalists with some relevant experience.
• You often have to cross the boundary to know where it is — being overly cautious means never finding the real limit.
• The question isn't whether to swing the pendulum, but how far and for how long.

Product market story fit

• Product market fit = right product for the right market (people with the same important problem).
• The missing piece is often story: a great product in a good market can fail if the explanation is convoluted or absent.
• Irdio vs Spotify: arguably a better product, same market, wrong story.
• Product people need to think about how to explain the product as rigorously as how to build it.
• Work closely with product marketing; learn positioning — why are you better, and can you say it clearly?

Jobs to be done and the four forces

• Jobs to be done is most valuable as a centering device: what is the thing people are trying to do, and do they care a lot about it?
• Energy matters — people with high energy around a problem are the right segment to study.
• Avoid getting lost in the academic debate between frameworks; use the version that helps you build better.
• The four forces (attraction of new solution, habits, anxieties, push away from status quo) is a practical tool for understanding why people switch — or don't.
• Intercom used it internally, including to recruit Paul himself.

Pricing

• Pricing is as difficult as onboarding — deceptively hard, rarely done well.
• Intercom's mistake: too many pricing models, add-ons on top of add-ons, tiering within add-ons; customers couldn't understand their bill.
• The principle "align price to value" is correct but nearly impossible to execute without simplicity constraints.
• Fight hard against adding new pricing dimensions; the complexity compounds over time.
• Keep it simple — even when a new feature seems to justify a new add-on.

Before/after moments

• A before/after moment is any event that changes the rules: a product launch, a rebrand, a pricing change, a technology shift like ChatGPT.
• Once you cross into the "after", go back out and talk to customers — were you right? What do they think now?
• The framework is a reminder to reset assumptions rather than coast on prior research.