Eric Ries wrote down ten pivots. I like lists like that because they force you to name the move you're making. "We're iterating" is what you say when you don't want to admit you're guessing.
AI changes the game in a dumb way: you can execute a wrong strategy faster and make it look nicer. You can generate a new landing page, a new pitch deck, and a new "positioning doc" in an afternoon. You can also be confidently wrong in three new fonts.
A pivot is still a structured change in strategy to test a new fundamental hypothesis. It's not "we rewrote the homepage." It's "we learned X, so now we're betting on Y." The "betting" part is still on you, and it still hurts a little.
The ten pivots, with AI-era failure modes
1) Zoom-in pivot
You take one feature and make it the product.
AI failure mode: you pick the feature that demos best, not the one people pay for. LLMs are great at writing a demo script. They can't tell you which feature customers complain about in week three, because they weren't on the call and they don't get yelled at in your support inbox.
AI use: summarize support tickets into themes and counts.
Human check: "Would I bet payroll on this being the whole product?"
2) Zoom-out pivot
The product becomes one feature of a bigger suite.
AI failure mode: you expand because you're bored. Or because the model gave you a "platform vision" that reads like it was scraped from a hundred startup blogs.
I worked at a physical product company that had a killer crowdfunding campaign. The genius programmer who built the core software got scooped up by a big tech company. Respectable move. It also meant we suddenly had a missing chunk of technical capability we couldn't replace fast.
The temptation was to "zoom out" into SaaS to smooth revenue. We tried the mental gymnastics. Physical product margins and SaaS margins don't share a brain. The SaaS story wasn't viable, and AI would've helped us write a prettier deck about it.
3) Customer segment pivot
Same product, new buyer.
AI failure mode: you let synthetic personas replace real buyers. You can generate "CTO Carla" all day. Carla never has procurement. Carla never has a security review. Carla never disappears after the second call because her boss changed priorities.
Traditional software: CRM exports, pipeline stages, loss reasons.
AI use: cluster sales objections and pull verbatim quotes.
4) Customer need pivot
Same customer, different problem.
AI failure mode: you confuse "they talked about it" with "they'll pay for it." LLMs can summarize interviews into clean bullet points that feel authoritative. If you didn't ask about budget and current alternatives, you collected vibes.
5) Platform pivot
App to platform, or platform to app.
AI failure mode: "platform" becomes a synonym for "we don't know distribution." If you can't name who builds on your platform in the next 90 days, you're building infrastructure nobody asked for.
I fought one of these once, on the other side of the ledger. At Atlas Wearables I argued we should stick with hardware. I lost that argument. Building hardware is hard — the firmware, the supply chain, the CAD tolerances that look fine on screen and don't fit in your hand. Building a community of people who actually strap your product to their wrist every day is hard. Building a team that can do both at once is hard. None of that difficulty is a reason to leave, and none of it goes away because you retitle the company "a software platform for wearable data." We pivoted off hardware and into software, and the company didn't get far after that. I don't think AI would've saved that call — a model can write you a beautiful platform roadmap slide. It can't tell you whether the thing you're walking away from was actually the hard part you'd already solved.
6) Business architecture pivot
High margin/low volume vs low margin/high volume.
AI failure mode: you pick the architecture that looks good in a pitch deck.
I've been the scapegoat when go-to-market fell apart. It's a special kind of job where the plan was never going to work, but somebody still needs to be blamed for the miss. As a consultant, I stay away from that pattern on purpose. If your architecture depends on a sales motion you don't actually have, the pivot is cosplay.
I've also watched a company try to run both architectures at once instead of picking one. There was a product line — higher margin, lower volume — and a services line sitting right next to it bringing in more total revenue at a fraction of the profit. Working there felt like a three-legged race, one leg tied to product, one tied to services, both trying to run the same sprint. Eventually the services side won internally, the way the bigger revenue number always tends to win internally, and most of the executive team that had been carrying the product vision got let go along with it. A company can be a product company or a services company. Trying to be both at the same time just means neither leg gets to run at full speed.
Traditional software: runway model, CAC, payback period. Spreadsheets aren't glamorous, but they're consistent.
7) Value capture pivot
Change how you make money.
AI failure mode: pricing pages get optimized before the product works. You can A/B test yourself into bankruptcy if you're testing on the wrong cohort.
8) Engine of growth pivot
Sticky, viral, paid.
AI failure mode: you overfit to vanity metrics because AI can generate dashboards that look scientific. I've watched teams chase numbers that never touched customers. It's motion, not growth.
9) Channel pivot
Change distribution.
AI failure mode: you mistake "AI can write content" for "content is a channel." Content is a channel if it reliably turns into conversations with buyers. Otherwise it's a publishing habit with charts.
10) Technology pivot
Same solution, new tech.
AI failure mode: you swap in an LLM because it's fashionable, not because it's better.
At that physical product company, I learned smaller, specialized models can beat larger general-purpose models. Sometimes you want the boring model that's predictable and cheap. Sometimes you want rules. Sometimes you don't want to ship a probabilistic system into customer support and act surprised when it hallucinates.
The pivot ritual AI can't do for you
I worked for a startup with a CEO who had a full-on cult of personality. People would do things they knew were incorrect because they worshipped the guy. I ask too many questions for that environment.
For pivots, the lesson is simple and uncomfortable: if the room can't tolerate dissent, you're not pivoting. You're acting out agreement.
Here's what I like instead:
- Evidence packet (real artifacts): usage logs, ticket counts, churn reasons, CRM losses, cash position.
- Scenario planning: "What does week one look like after this pivot?" AI is good here. Make it write the first sprint plan, the first sales email, the first onboarding flow. Then critique it like a human who actually reviews AI output.
- Decision record: write down the hypothesis that changed and why. ADRs, memos, board docs. Pick your format, but leave a trail.
The fake pivot to watch for
The fake pivot is when you change messaging and keep the same broken assumptions. AI makes this dangerously easy because it can repaint everything in minutes.
New copy. New deck. Same problem: customers don't care enough to pay, you can't reach them, or the unit economics don't work.
If you want help, I run pivot workshops and strategic teardowns. We show up with evidence, constraints, and consequences. We leave with a decision record and a plan that fits your runway.
-Sethers
