Structured enough to find the real issue. Practical enough to act on it.

1. 1

   Understand

   The owner wanted growth without adding another layer of managers, and without customers noticing the strain. A few conversations made it clear this wasn't a headcount problem: every safeguard in the business had been added after something went wrong, not before, and years of quick fixes had never been tidied up.

2. 2

   Diagnose

   Following a handful of contracts and orders end to end told the real story. Quotes went out with no visibility of margin until the very end, if at all, so nothing stopped a deal being agreed at a rate the business hadn't intended. Customer and billing details were typed into one system, then retyped into another, because the two had never been connected — and a couple of long-running calculations had quietly drifted apart over time, so the same contract could show a different number depending on which report was pulled. Stock told a similar story: goods moved constantly, but those movements weren't linked to the accounts, so using or selling stock never automatically moved its value from the balance sheet to the P&L, and month-end needed manual adjustments nobody could fully explain. Underneath all of it sat one person who alone held the month-end numbers, the stock adjustments and the payments — if they were off sick, that work simply waited. Meanwhile, a few team members had already started using AI writing tools on their own initiative, without anyone agreeing what was safe to put into them.

3. 3

   Prioritise

   Connecting the systems, putting an approval gate on quoting, and taking the highest-risk manual work out of one person's hands were obvious quick wins: high benefit, low effort, low risk. Untangling the stock and billing history properly was worth doing, but only once the daily numbers could be trusted. The AI use already happening wasn't banned or waved through — it went on the same list, with a plan to keep what worked and govern it properly.

4. 4

   Implement

   A margin check was added so a quote outside the agreed range needed sign-off before it could go any further, rather than relying on someone catching it after the event. The two disconnected systems were brought together so information only had to be entered once, and the drifting calculation was retired in favour of one shared version everyone could see. Stock movements were connected to the finance system, so goods received, moved and used or sold triggered the right accounting entries automatically — giving an accurate, real-time stock picture that matched the accounts and meant month-end no longer needed guesswork. The single point of failure was addressed by documenting the process in plain language so more than one person could pick it up. On AI, one narrow, carefully governed use was kept and rolled out properly, with clear rules on what information could go near it; a separate attempt at using AI to automatically flag data-quality issues was mostly undone after testing showed it missed far more real problems than it caught, sounding confident while being wrong — those checks were rebuilt as plain, explainable rules instead.

5. 5

   Embed

   Ownership of the new setup, including the parts that used to live in one person's head, passed to named people internally, backed by a simple internal reference written in plain language rather than technical jargon. Check-ins at two, six and twelve weeks confirmed it was sticking. The business chose to keep CoveOps on a monthly retainer as a fractional Business Operations Director through the next stage of growth — support that isn't always necessary or advised, but suited this business at this stage.

Details have been anonymised to protect the business involved.