A five-minute test for any step
Before you add a script, a new integration, or an AI add-on you cannot maintain, run this in a room with someone who will own the result — not just the person who likes new tools.
- How often? A task that happens ten times a day with the same input shape is a different bet from something that happens twice a year in five variants you never documented.
- How wrong is wrong? If a bad run touches money, a patient, a regulator, or a live customer order, the bar for automate is higher than save five minutes.
- What is the exception rate? If most weeks need a human override, you are not buying speed — you are buying a bot to argue with until someone reverts to a spreadsheet. That is a debt line item, not a win.
- Is the process stable? Map the real flow first — if the Tuesday version still changes every month, automation is a moving target with a maintainer you have not named.
- Who signs off at go-live and who is on call when it breaks? If that is nobody or IT when they have time, keep the step manual until ownership is clear.
Tie the business case to simple ROI levers — time, rework, speed to cash — in bands the team can defend, not a vendor calculator.
Book a free systems review to walk a single workflow with us. Read our process page for how we work, and what we offer for process audit, build and integration, and ongoing system care when something is in production. Teesside, the North East, and beyond when the fit is right.
When exceptions matter more than speed
Keep a careful human in the loop when:
- The edge case is the job — creative quotes, complex credits, we have never seen this before — and a wrong default embarrasses the brand or reopens a closed dispute.
- The data going in is not trusted yet. Automation that magnifies a dirty field is a faster way to be wrong at scale. Clean the system of record and ownership first.
- The regulatory or story is we can show a person decided this, not the script said so.
Speed is not the only currency. Sometimes the right answer is a better checklist and a single owner, not a third integration. If the problem is we should not have built that custom thing at all, read is custom software worth it? before you fund the next box.
When automation is worth the maintainer
Worth the ongoing cost when the same repeated failure is measurable: same re-key ten times a week, same missed handoff, same report nobody trusts until someone re-types it. You are buying reliability and time returned, not novelty.
Before you call it production:
- A named person (not the system) is responsible for what happens when the API changes, the vendor deprecates a field, or someone changes a password.
- A test you actually run — e.g. dry run or a copy environment — and a rollback path so go-live is not a leap of faith.
- A throttle — the script cannot silently queue ten thousand bad rows because nobody was watching a dashboard you never open.
If the maintainer is whoever is least busy next month, keep the manual step, document it, and revisit when the process map has stopped moving.
Book a systems review to sanity-check what you are about to script, or see all our offers for audit, build, rescue, and care.