About a year ago I sat in a room where the entire roadmap got rewritten in ninety minutes.
The trigger wasn't a customer. It wasn't the market. It was a competitor's press release - an "AI-powered" version of the same product we sold.
By the time the meeting finished, we had AI initiatives on the roadmap.
No named product gap. No customer asking for it. Just a competitor logo we didn't want to lose to.
I've now watched that exact meeting happen at companies at every stage - VC-backed, PE-backed, pre-IPO. The disruption is real. The fear is rational.
And the response is almost always wrong in the same way.
Here's the part nobody says out loud.
The hard AI decision in 2026 isn't whether to build a new AI product. It's what to do with the product you already have.
The new-product conversation gets all the oxygen. Agentic workflows. Greenfield copilots. The demo that raises a round.
But most of the enterprise value in your company isn't in the thing you might build.
It's in the product people already pay for - the one with the customer base, the renewal cycle, the margin, and the trust you spent years earning.
And that product is exactly where AI gets added carelessly. Because adding it feels like progress. And not adding it feels like falling behind.
So teams add it fast. An assistant here. A "summarize with AI" button there. A model dropped into a core workflow that used to be deterministic and trusted.
Then the renewal call gets awkward.
I'm not theorizing about this. I'm living it twice.
For fifteen years I made hard product and capital-allocation calls inside companies - VC-backed, PE-backed, pre-IPO - with real money and real customers on the line. So when AI disruption hit, I wasn't watching it from outside I was inside companies where it was actively reshaping the markets we sold into, deciding in real time what to do about the products we already had.
And right now I'm making the same call on my own product. I'm adding AI to OneRank. In public. Which means I have to run my own framework on myself - and it is genuinely uncomfortable, because the framework doesn't care what I'm excited to build.
So here's the framework. Three of the questions I run before any AI capability goes near a shipping product.
Question 1: What business gap does this close - with the word "AI" deleted?
Strip the AI acronym off the strategic bet and read what's left.
A gap a customer can name. A job they're already trying to do and failing. A cost you're carrying that this removes.
If the only thing holding the initiative up is the word "AI," it was never a strategic bet. It was a reaction to someone else's press release. Real strategy survives the deletion of the buzzword. A reaction doesn't.
Question 2: Does it protect the job the product already does well - or quietly degrade it?
This is the one that most teams skip.
Your existing product earns its renewal because it does a specific job reliably. Search that returns the right answer. A number a customer can trust. A workflow that behaves the same way every time.
Put a probabilistic model into the middle of that, and you can trade away the exact thing customers pay for. A "smarter" search that's now wrong 8% of the time. An assistant that hallucinates into a workflow people used to trust blindly.
A great AI feature that breaks the core job is not a feature. It's churn with a better demo.
Question 3: Can the business afford it at the scale customers will actually use it?
The demo runs on ten queries. The renewal runs on ten million.
Inference has a unit cost. The more customers love the feature, the more it costs you to serve - and unlike a one-time build, that cost recurs every single time it's used.
I've watched teams ship an AI capability that delighted users and quietly inverted the margin on the account. The feature worked. The business case didn't survive contact with adoption.
If you can't draw the cost-to-serve curve at scale, you haven't finished the decision.
These three are the test.
There are more - the questions about what you're choosing not to fund to pay for this, how you'll know within a quarter if you were wrong, and how you protect customer trust while you learn. That's the full sequence I run live.
But if a proposed AI feature can't survive these first three, the rest doesn't matter. You're not adding AI to your product. You're adding risk to your renewals and calling it innovation.
So here's the question worth sitting with this week:
Look at the AI feature highest on your roadmap for an existing product. If a customer never saw the letters A and I, would they still ask you to build it?
If the answer is yes, you have a strategic bet worth pursuing.
If the answer is no, you have a press release you're funding with your roadmap.
I run the full version of this framework live, free, as a Lightning Lesson series - the complete set of questions every AI bet has to survive before it goes near a product people pay for. Bring the shakiest initiative on your roadmap and we'll run it through all of them.
Save your seat for the series →
See you in the room,
Elena Leonova
Executive product & business-strategy leader
I work with senior product leaders, Directors, and VPs to help them master product strategy when decisions are high-stakes, ambiguous, and made at scale - where trade-offs matter and the cost of getting it wrong is real.
This newsletter reflects the thinking behind my work across:
• Product Executive education - From PM to Product Executive: Master Strategy, Finance and Executive Influence
• Advisory & coaching - product strategy and executive decision-making
• Writing & research - including my forthcoming book The Art of Platform Products