This one is from a fintech infrastructure company built the fast way: by acquisition. Four acquired product stacks plus the parent platform — five engineering teams, each with its own codebase, its own sprint cadence, its own roadmap. Shared services underneath, yes. A profitability clock ticking loudly in the background.
Every team took to AI early and seriously. Coding copilots. Test generation. Ticket deflection. Each team's velocity genuinely improved — more shipped per sprint, fewer bugs per release. Real gains, honestly earned.
And the P&L didn't notice. Costs kept growing faster than revenue. The burn stayed the burn.
Nobody was doing anything wrong inside any team. That's exactly the problem.
The fault wasn't headcount, and it wasn't effort. It was the shape of the problem-solving.
Each team developed its own use cases, for its own pain, on its own roadmap. Each defined success inside its own boundary — our velocity, our tickets, our stack. Five good teams, solving in parallel. Never together.
So every gain was a single-team gain. And single-team gains have a ceiling: they make each part cheaper without making the whole more valuable.
The stacks that should converge. The client onboarding that crosses every seam. The cross-sell that needs two product lines to behave as one. Every one of those is worth more than any team's velocity — and every one needs several teams to move together. Nothing in the design made them.
Here's the simple question that reorders everything: what does this company look like from the customer's seat?
A client — a bank, a lender — doesn't buy from five teams. They experience one company: one onboarding, one integration, one support relationship that happens to cross every stack. From their chair, the seams are invisible and the friction is total.
From inside, no team can even see that journey end to end. Five roadmaps, each rational alone, that never meet. When two of them conflict, it resolves in an escalation — months later, at the founders' table — or it doesn't resolve at all.
Read from the customer lens, the real work-cases surface instantly — and almost none of them fit inside one team: make onboarding one motion, make the modules snap together, make support see the whole client. The AI use-case lists, built team-by-team, had walked right past all of them.
Everyone nods at "align the teams." The machine org has tools for it: steering committees, sync meetings, quarterly planning. Coordination by calendar. It's slow, it burns leadership time, and it decays the week attention moves elsewhere.
Getting teams to actually move together takes three designed things, not more meetings:
Context available across. One shared picture — the customer, the journey, the live state of every thread that touches them — that every team reads, so "what the customer needs" stops being one team's opinion.
Decisions placed at the seams. The calls that sit between teams — converge or keep, whose roadmap moves first, one onboarding motion or five — each given a clear owner, clear authority, a clear tie to the outcome. Today those decisions belong to everyone, which is to say no one.
Roadmaps read from one outcome. Not five lists merged in a workshop — one customer outcome, decomposed backwards into what each team builds and when. Merging roadmaps is the output of a shared outcome, never the input.
That's the multi-stakeholder work — the unglamorous middle where the compounding gains live. AI makes it genuinely possible for the first time: the shared context can now be built and kept live. But the context doesn't align the teams by itself. The decisions and the org design do.
Three questions, one afternoon: