A model innovation framework that orders every change by reversibility

Changing a business model on a live company is a sequencing problem before it is a strategy problem. Most of the damage comes from moving several things at once and then being unable to say which one worked. Figures are invented placeholders.

What to take away

Two things to settle before any move

Start from a constraint, not an idea. "We want recurring revenue" is an ambition. The constraint is the thing that actually caps the business today.

It is a calendar, a channel you do not control, a cost that does not fall with volume, or a buyer who cannot approve an unknown number. Write it in one sentence with the measure that shows it. If you cannot produce the measure, you are about to change the model on a hunch, and every result afterward will read as confirmation.

Then decide what you are deliberately holding constant: the product, the segment, or the delivery method, one of them, not all three. A firm that changes its pricing, its target customer and its delivery approach in the same quarter has spent the year and bought no information.

The ladder: five changes, easiest to undo first

Severity is the wrong ordering principle. The most damaging problem is often the least reversible one, and starting there means learning nothing before you are committed. Climb in this order instead.

The ladder works because each rung leaves you able to read what happened. A discount rule can be reversed before the next quarter, so a poor result costs little and teaches something. A change of customer cannot, so it waits until the earlier rungs have produced evidence.

If the first rung is a discount shown against a former price, the FTC's Guides Against Deceptive Pricing say a former price is a legitimate basis for a comparison when it was the actual, bona fide price offered on a regular basis for a reasonably substantial period.

Cost the transition as its own line

The new model's arithmetic assumes the new model exists. Getting there costs money the model never sees: two price structures running at once, a sales team selling both, support answering questions about both, and a period where neither is fully working.

Budget it as its own line with its own end date. Transitions funded out of the new model's expected margin stop halfway, which is the worst place to be, because the company then runs two models and gets the benefits of neither. The cost behavior underneath is in unit economics.

Reading the result of each rung

Name the confounders you already know, and treat an observed pattern as a hypothesis to test, not a finding; that is ordinary discipline in MIT's data mining course.

A live company is not a controlled experiment: your mix moved, the season moved, and a competitor did something you did not see. Accept a directional answer, not a manufactured precise one.

If the change touches a platform's fee, the mechanics are in the platform repricing guide.

Common questions

What if the biggest problem is who the customer is?

Then it still goes after the cheaper rungs. Who the customer is cannot really be undone, so it is tested last, once the discount rule, the packaging and the price basis have taught you something.

Can we run the old and new models permanently?

Sometimes, and only if each has its own arithmetic and its own owner. What kills companies is running two models with one blended set of numbers, because neither can then be assessed.

How long should a transition take?

Longer than planned, and the useful discipline is naming the end date in advance so that overrunning it triggers a decision rather than a shrug.

How do we know the constraint is real and not a story?

It has a measure attached and the measure moves when the constraint binds. A constraint you cannot measure is a belief, and beliefs are expensive to restructure around.