A platform group's next four quarters are 70% cost-reduction work. The company has 300 engineers, revenue of about 120M dollars, cloud spend of about 5M dollars a year, and a board target of growing revenue 40% with no new headcount. Which observation most changes the roadmap?

The diagnosis

Do the two divisions before touching the roadmap. Revenue per engineer today is 120M over 300, about 400k. The board's target is 168M with the same headcount, about 560k per engineer — an extra 160k of revenue per engineer per year, which is the ratio the company is actually being measured on. Cloud spend of 5M is about 4% of revenue, so the finance team's 20% saving is 1M, or 0.6% of revenue.

Now compare ceilings. A cost-reduction programme can never be worth more than the cost line it attacks, and that line is 5M. A throughput programme multiplies payroll instead: 300 engineers at a loaded cost on the order of 200k is roughly 60M a year, twelve times the cloud bill. A 5% improvement in how much product change each engineer can land is worth about 3M a year — three times the entire 20% cloud saving, every year, and it compounds with headcount while the cost saving does not.

So the roadmap is pointed at the smaller lever. The work that matches the binding ratio is the work that removes engineer-time per unit of change: build and test duration, environment provisioning, the integration work between services, the time to get a change to production safely.

The misleading signal

The 5M is a big absolute number and it appears on a bill with the platform group's name on it, which is why it attracts the roadmap. Ownership of a cost line is not evidence that the line is the constraint. The test is always the ratio the business is measured on, not the size of the invoice you happen to receive.

Why the other options fail

• The competitor's infrastructure per user. A ratio computed without visibility into their product, pricing or margin. Even if true it argues about the same 5M line, so its ceiling is identical.
• Finance asking for 20%. A real stakeholder request, and the correct response is to show what it is worth — 1M, 0.6% of revenue — and ask which alternative it displaces. Accepting a request as a roadmap is how a platform group ends up optimising a number nobody can connect to the board target.
• Spot capacity at 30% of compute. True, useful, and worth perhaps 0.5M to 1M, minus the engineering cost of interruption handling. That is one team for one quarter, not four quarters for a group.

When this is the wrong answer

Flip it where infrastructure is the binding ratio rather than a rounding error. In a thin-margin high-volume consumer business, infrastructure can run 25–40% of revenue, and then cost per transaction is the roadmap. Flip it also on a runway constraint: for a company with nine months of cash, 1M is a month of life and the discounted value of next year's throughput is close to zero. The rule is to find the ratio the business is being measured on this year and size each roadmap item against it before ranking anything.