Mid-Market Math Is Not Enterprise Math · The Reluctant Guide to Shopify Migrations

Conference room. Wednesday. Three finance people, two screens, one spreadsheet with too many tabs.

The spreadsheet is modeling something the head of finance asked for on Monday in a voice that suggested it would have been wiser to ask three months ago. The question, in plain English: what does a one-point drop in conversion rate cost us, for one day, on the new platform?

A cell turns yellow. A number resolves. The youngest person in the room makes a sound somewhere between a cough and a laugh, then looks embarrassed about it. The number is more than two of them earn in a year.

Nobody says anything for a moment.

Then somebody says, “Right.”

Dear CFO Who Already Knew This,

You did the math in your head three months ago, the first time someone said the word replatform in a steering committee. You did not bring it up. You did not bring it up because you were waiting to see whether anyone else in the room would. They didn’t. They are still talking about deliverables and milestones and velocity. The agency calls it velocity. The agency’s slide deck has a graph of velocity. The graph goes up and to the right. Nobody has labeled the y-axis.

You were going to bring it up next week.

You don’t have to now. The spreadsheet brought it up.

The decimal point changed. The discipline didn’t.

Most Shopify agencies built their methodology working with brands doing five to thirty million a year. The methodology works there. It works there because the math, at that size, forgives. A bad week of checkout is a bad week. A botched data migration is a long weekend with the engineering team and an apology email to the customer base. The brand absorbs it. The brand goes on.

At a hundred million, an hour of degraded checkout is a Lufthansa Business Class flight to Tokyo. (One way. Two passengers.) A day of it is a small house in a not-very-fashionable suburb. A week of it is the kind of number that lands in the audit committee’s notes with a font change.

A small bucket on a bare floor under a vast dark ceiling, nearly full from a slow steady drip falling from the darkness above. — The bucket was built for a much smaller ceiling.

The same agency, running the same playbook, can do good work at ten million and disastrous work at a hundred million without ever noticing they changed anything. They didn’t change anything. That was the problem.

(Anyone who tells you otherwise is selling you something. Ask us how we know.)

The risk discipline at this scale looks different from the outside.

Longer testing cycles, run against real data volumes rather than the sample dataset somebody pasted into a Google Sheet at midnight. Staged rollouts that validate functionality on a slice of traffic before the slice becomes the whole pie. Parallel running of the old and new systems during transition, so the cutover is a moving and not a jumping. Conservative timelines that account for the four people who will, predictably, be on parental leave or at a wedding in Greece during the week the plan calls for hypercare. (One of them will be your DBA. It’s always your DBA.)

From the inside, the discipline looks like cowardice. It is not. It is professional respect for the math.

The agency that makes you feel most confident during the sales process may be the one least equipped to handle the reality of your migration. Confidence is easy when you’re applying a template. The template was built for a smaller animal.

Senior Engineer Reading This Late at Night, you already understand this. You are not the audience for this chapter. The audience is the colleague three doors down who keeps using the word velocity and who has not, as far as you can tell, looked at the spreadsheet.

Slide the spreadsheet under their door.

Here lies the migration that optimized for velocity.

The velocity, it turned out, was downhill.