What building our own tools taught us about selling automation
AI Metric runs on the systems it sells. Site records are captured from WhatsApp through Construction Metric, the same product we put in front of contractors. The CRM that tracks our enquiries and nudges the follow-ups is one we built ourselves. Proposals, follow-up emails and document packs come out of our own pipelines. None of this is a marketing posture; it started because we needed the tools and were the cheapest available client.
Running your own automation every day teaches you things a sales deck never would. Most of them are mildly humbling. The four below are the ones that changed how we sell, because they changed what we believe the product actually is. The short version: the first version is wrong, adoption is a habit problem, and maintenance is the product.
Why is the first version always wrong?
Because you design for the process you believe you run, and then the tool shows you the process you actually run.
Our first pass at follow-up automation assumed enquiries arrived, were qualified, and moved forward in tidy stages. Real enquiries arrive half-formed, go quiet for three weeks, then reappear as a different question from a different person at the same firm. The workflow we drew on day one did not survive contact with our own inbox.
That experience now shapes how we scope client work. We no longer try to specify the perfect system up front, because we know the specification is a hypothesis. We build the smallest version that touches real traffic, watch it be wrong in specific ways, and correct it. This is exactly the shape of a properly run pilot, and we trust it because we inflicted it on ourselves first.
Is adoption a software problem or a habit problem?
A habit problem, and it does not spare the people who built the tool.
We built the capture system; we still had weeks where messages went to the wrong group, or a record everyone assumed was being kept was not. The tool worked. The habit had lapsed. If the builders of a system drift away from using it, expecting a busy site team to adopt one because it is "better" is fantasy.
What actually holds a habit in place is the removal of alternatives and the shrinking of effort. Capturing site records through the WhatsApp groups people already use works because nobody has to open anything new. The same logic drives our WhatsApp site assistant service: meet the habit where it lives. The organisational version of this argument is laid out in atomic habits for organisations, and we would summarise our own experience the same way: design the system so the lazy path is the correct path.
What did running our own CRM teach us?
That the follow-up you automate is worth more than the record you keep, and that data you do not look after goes stale faster than you expect.
A CRM you update manually is a diary of your good intentions. Ours only became useful when it started doing work back: surfacing the enquiry that had gone quiet, drafting the nudge, filing the reply. The fuller account is in what we learned building our own CRM, but the sales lesson is simple: clients do not buy a database, they buy the behaviour the database drives.
It also made us tidier about data handling than any policy document would have. Once contact records, messages and documents flow through automated pipelines, you are processing personal data at a steady hum, and the ICO's guidance for organisations stops being abstract. We also apply the basics in the NCSC small business guide to our own systems, because an automation firm that gets breached has no product left.
Why is maintenance the product?
Because every automation we run has needed attention since the day it went live, and there is no reason a client's would not.
| What we assumed | What running it taught us |
|---|---|
| Build it once, it runs forever | Inputs drift: formats change, people leave, suppliers rename PDFs |
| The demo is the hard part | The demo is the easy part; month four is the hard part |
| Users will report problems | Users route around problems silently; you must watch the system |
| Documentation can wait | The person who built it will forget how it works in six months |
This is why we are wary of selling automation as a one-off project, and why ongoing support is in every proposal. A system nobody maintains decays into a system nobody trusts, then into a system nobody uses, and the client reasonably concludes automation does not work. It worked; it was abandoned.
Does eating your own cooking actually help sell?
It is the only demo that holds up.
A sceptical contractor can smell a canned demo instantly. What lands is showing our own live systems: the real record of a real week, the follow-up log with its unglamorous gaps, the pipeline that failed in March and what we changed. We make no claims about client results in these conversations; we show our own operation and let it argue for itself.
There is a discipline hiding in that. If you sell automation you refuse to run on yourself, you are asking clients to take a risk you declined. Everything AI Metric proposes, from document automation to workflow plumbing, is a pattern we run internally first, at our own expense, on our own traffic.
The lessons are not flattering: our first versions were wrong, our habits slipped, our systems needed constant tending. But those are the true costs of automation, and a supplier who has paid them personally will scope your project more honestly than one who has only ever watched from the invoice side.