We built our own CRM instead of buying one: what we learned
AI Metric runs on a CRM we built ourselves: contacts, companies, deals, follow-ups and call notes in one hub, wired into the tools we already use. We did not build it to prove a point. We built it because seat-based pricing is a poor fit for a small consultancy, and because we wanted our communications to land in the record on their own rather than being retyped into someone else's idea of a pipeline.
This post is the honest version of how that went: what was easy, what was harder than expected, and when building your own is the wrong call. One caveat up front, stated plainly: we build systems like this for a living, so our build cost is not typical. Read what follows as evidence about where the real difficulty sits, not as a claim that everyone should do the same.
Why build instead of renting seats?
Three reasons, in decreasing order of respectability.
The structural one: off-the-shelf CRMs model a sales pipeline, and our work does not look like one. We wanted deals shaped like our engagements, follow-ups shaped like our habits, and nothing we would have to fight. It is the same mismatch we describe in why generic CRMs fail builders, applied to a consultancy.
The economic one: per-seat pricing punishes small teams that want everyone in the system, and the features that make mainstream CRMs expensive are mostly features we would never touch.
The honest one: we wanted to eat our own cooking. We tell clients that the useful system is automation, not another software login, and building our own hub was the cheapest way to find out where that argument creaks.
What turned out to be easy?
The part everyone worries about: the database and the screens. A schema for contacts, companies, deals and notes is a solved problem, and the first working version existed inside a week of part-time effort.
The genuinely pleasant surprise was capture. Wiring meeting summaries into the record meant call notes started existing without anyone writing them up, which is the effect we described in meeting notes are free money. Follow-ups generated from those notes turned the CRM from a filing cabinet into a short morning list of who to get back to. That loop, note in, nudge out, is the whole value of the system, and it was days of work rather than months.
What was harder than we expected?
Everything that involves discipline rather than code.
| What we assumed | What actually happened |
|---|---|
| The database would be the hard part | The schema took days; hygiene is forever |
| Notes would flow because the tool is ours | Notes flow only where capture is automatic |
| Duplicates would be rare | The same company appeared under three spellings within weeks |
| Maintenance would be occasional | It is small and constant: fields, fixes, backups |
| The win would be features | The win is that communications land in it with no effort |
Data hygiene is the honest headline. Duplicate companies arrive constantly: Ltd against Limited, trading name against registered name, the same person from two email addresses. Merging is judgement, not code, and if nobody owns it the database quietly rots. Manually typed notes remain the weak point too: anything captured automatically stays current, and anything relying on someone typing degrades exactly when you get busy, which is exactly when the record matters.
Two responsibilities also move onto your own shoulders when you build. Holding contact data lawfully, including deciding what to delete and when, is yours to manage, and the ICO's guidance for organisations is the reference, not folklore. So are backups and access control, for which the NCSC small business guide covers the unglamorous basics. A vendor was doing some of this for you. Now you are the vendor.
When is building your own CRM the wrong call?
Most of the time, and it is worth being blunt about the cases.
Do not build if nobody will maintain it. A homemade CRM is a small product you now own, and a product with no owner becomes the least trusted system in the business within a year. Do not build if your sales process is genuinely standard, because the off-the-shelf tool already matches it and your edge cases are imaginary. Do not build if no one in-house can debug it on a Tuesday when something breaks. And do not build as a way of avoiding selling; a fortnight of schema design is procrastination wearing a hard hat.
Buying, or staying on a disciplined spreadsheet, is the right answer for most small firms. There is no shame in either; there is only shame in paying for seats nobody logs into.
What is the lesson even if you never build one?
The CRM that works is the one your communications already flow into.
That is the entire finding. Ours works not because it is clever but because emails, meeting notes and follow-ups arrive in it with zero clicks, so the record is current without anyone being good. Judge any system you are about to pay for by that one test: how much lands in it if everybody forgets it exists for a fortnight? If the answer is nothing, the tool will be a monument by spring, whoever built it.