Every implementation has a plan, a budget, a named owner and an end date. The month after it goes live has none of those, and that is the month the estate starts ageing.
This is not a complaint about implementation partners. A project ends because projects end, and a good one hands over cleanly. The question is what was supposed to happen next, and whether anybody was asked to price it.
1 · The warranty is not what people think it is
A migration or an ERP rollout closes with a sign-off, a document set and a warranty period, commonly thirty days. That warranty covers defects in what was built. It does not cover the estate continuing to work.
Those are different obligations. A patch that has to be applied in month four is not a defect. A certificate that expires in month nine is not a defect. A nightly job that starts failing because an upstream system added a field is not a defect either. All three are Tuesday problems, and on a thirty-day warranty every one of them lands after the cover has ended.
2 · What breaks first, roughly in order
The useful part is that the sequence is predictable. If you know what goes first, you know what to ask for before the project closes.
- Certificates. Ninety days for the automated ones, a year or two for the purchased ones. The renewal usually lives in one person's calendar, or in nobody's.
- Backups that have never been restored. A green backup job is evidence that a job ran. It is not evidence that you can get the data back, and the gap between those two facts is normally discovered on the worst possible day.
- The scheduled job nobody watches. Most batch failures are silent. The report is simply empty, or yesterday's, and somebody notices in week three.
- Access that was never revoked. Implementation accounts, vendor VPN profiles and a shared admin login that four people still know. Every one of them was legitimate during the project.
- The person who knew. Not always a resignation. One person taking two weeks of leave is enough, if the knowledge was never written down.
3 · The three answers, and what each one actually costs
Hire somebody. The honest cost is not the salary. It is the salary, the recruitment time, the ramp-up, and the fact that one person is not a rota: they take leave, they sit in one timezone, and when they resign the environment leaves with them. Above a certain size this is still the right answer, and we will say so when it is.
Keep the implementer on. Often sensible for the first quarter, because they know what they built. It gets expensive as a permanent arrangement, and it quietly removes your ability to change partner, because nobody else has ever had access.
A retainer. Buys coverage, breadth and continuity instead of one person's availability. The thing to check is what it actually includes: whether the SLA promises a reply or a fix, and whether monitoring means somebody is looking or means a dashboard exists.
4 · What to ask for before the project closes
None of this costs anything while the project is still open, and all of it is awkward to obtain afterwards.
- An as-built document that matches what was actually deployed, not the design that was approved in month one.
- A runbook for the five things most likely to need doing: a restore, a certificate renewal, a failed batch, a user lockout, a rollback.
- One restore, actually performed, in front of somebody from your side.
- A list of every credential and account created during the project, and who still holds them.
- A renewal calendar: certificates, licences, support contracts and domains, with dates.
- The infrastructure code and configuration in a repository you own.
If a partner cannot produce that list, the handover has not happened yet, whatever the sign-off says.
5 · If it has already gone live and none of this exists
This is the common case, and it is recoverable. It does not require rebuilding anything.
What it requires is somebody going through the estate as it stands, writing down what is there, finding the renewals before they find you, and testing one restore. That is a couple of weeks of work, and it is the same two weeks whether you then run it yourselves, hire somebody, or ask us to.
The thing not to do is wait for the first Tuesday that goes badly, because that is the most expensive possible moment to start writing documentation.
Need somebody to run it?
Monitoring, patching, backups and the phone answered when something breaks, on a retainer sized to the estate rather than to a rate card.