handbooksInnovation accounting
How to show progress on a product with no revenue yet. Learning milestones, the baseline, and the three steps that turn guesses into a number a board can read.
- Normal financial reporting says nothing about a product with no customers yet, so a startup needs its own way to show progress.
- Three steps: get a baseline, tune the engine toward the target, then decide whether to pivot or keep going.
- Set learning milestones the way you would set any goal: specific, measurable, and dated.
- This is the boring principle, and it keeps a team honest when the numbers are flat.
The problem it solves
"We are making great progress" is not a claim anyone can check. Without revenue or a plan to compare against, a team can work hard for a year and nobody can say whether it went well.
Innovation accounting fixes the reporting gap. It gives you something real to hold a team to before there is a profit and loss statement worth reading.
Three steps
| Step | What you do |
|---|---|
| 1. Establish the baseline | Build an MVP and measure how the product performs today, however bad. That is your starting line |
| 2. Tune the engine | Change one thing at a time and move the baseline toward the target. Every experiment should shift a number you named in advance |
| 3. Pivot or persevere | When the numbers stop moving toward the target, change direction rather than keep tuning |
The order matters. Teams often skip step one because the early numbers are embarrassing. Without a baseline, later improvements cannot be shown.
Learning milestones
Set goals about what you will know rather than what you will ship. Treat them like any other goal: specific, measurable, achievable, relevant, and dated.
| Weak | Better |
|---|---|
| "Validate the onboarding idea" | "By 30 April, get activation from 42% to 55% for one cohort" |
| "Improve retention" | "By Q3, 30-day retention above 15% for two cohorts running" |
| "Ship the referral feature" | "By 15 May, know whether referrals bring one in ten new signups" |
The right-hand column can fail. That is the point of it.
Reporting it
Cohorts are the reporting tool. Metrics that are not lies covers how to read them.
That is a signal, not a reason to try harder. A run of tuning experiments that does not move the numbers is the clearest evidence you have that the direction is wrong.
Ask what your baseline was for the metric you are trying to improve. If nobody can say, you cannot demonstrate progress on it either.