Shipping to production is not giving a demo

A demo runs on the machine of the person showing it, at the moment they show it, on the data they picked. Production runs on a Sunday at three in the morning, on data somebody else typed in, with nobody watching the screen. These are two different jobs, and the gap between them is almost always underestimated.

What the demo does not show

On Noqen, the visible platform fits in a handful of screens. Behind them sit 148 API routes, 31 data models, 23 payment webhooks, 26 transactional e-mail templates and 3,274 translation keys the team can edit without a redeployment. None of those numbers show up in a demo. All of them show up on the invoice of whoever takes the product over without them.

The 22 user journeys covered by end to end tests are the clearest example. Nobody ever looks at them. They exist for exactly one day: the day a harmless looking change breaks payments, and you want to hear about it before your customer does.

The authentication case

At Hero, the neobank served 13,000 small businesses. I built the strong authentication there: a custom second factor, a PSD2 compliant SCA journey on mobile, public and private key encryption.

In a demo, that layer impresses nobody. It shows a code, you type it in, it goes through. It gets noticed in exactly two situations: the day it lets through a payment it should have refused, and the day it refuses a legitimate one. Both are expensive, and neither can be tested in a meeting.

The questions I ask before starting

There are three, and none of them are technical.

Who gets called when it falls over? If there is no answer, there is no production, there is a launch. The difference gets paid at the first incident.

Where are the traces? A system you cannot observe is a system you cannot repair. You can restart it, and that is all.

Who is allowed to deploy? If the answer is one person, and that person goes on holiday, the product goes with them.

What that does to the quote

A product announced as six weeks of work may well be six weeks, if you count the time it takes to write what can be seen. You have to add what cannot be seen, and that is not a safety margin, it is half the job.

I would rather say so at the start than at the closing review. It is less pleasant for twenty minutes and a great deal more pleasant for six months.