Margin leaves in small increments
Professional services firms rarely lose money dramatically. They lose it four minutes at a time: the same data typed into a second system, a report assembled by hand each month, an hour spent finding evidence for something everyone already knows to be true.
None of it is a technology problem exactly, which is why it survives so long.
1. Asking people is the wrong diagnostic
Automation projects usually start by asking the partners what wastes time. The answers are collected and the loudest item gets built.
It reliably picks the wrong target, because what people report is what annoys them, and what costs money is what has stopped annoying them.
A task that takes twenty minutes and fails occasionally generates a complaint every time. A task that takes four minutes, works fine, and is done eleven times a day by six people generates none — and costs several times more. It has been absorbed. It is just how the work is done.
2. Watch the work instead
Sit with people while they do their job for a few days. Not a workshop, not a process map on a whiteboard — the actual work, including the parts nobody would think to mention.
What consistently appears:
- Re-keying. The same information typed into a second system because the two do not talk. Usually the largest single item, and never volunteered, because it is regarded as normal.
- Assembly. Building a document, pack or report by copying from three places.
- Chasing. Finding out the status of something, which exists because status is not visible anywhere.
- Checking. Verifying that a previous step happened, done by people who no longer trust it.
- Reformatting. Data arriving in one shape and needed in another, fixed by hand every month.
3. Rank by total time, then by stability
Multiply duration by frequency by the number of people doing it. The resulting ranking is usually nothing like the one produced by asking, and is almost always dominated by short tasks done constantly.
Then apply the second filter: how stable is this? Automating something that changes every quarter buys a maintenance liability. Automating something done the same way for six years is a purchase.
What to leave alone
- Judgement. If the step requires expertise, automate the preparation around it and leave the decision.
- Genuinely rare work. A quarterly hour is four hours a year and does not repay a build.
- Anything upstream of a broken process. Automating a bad process makes it faster and harder to change.
4. Adoption is where these projects die
The build finishes, the tool works, the demo goes well. Six months later usage is a handful of people and the old spreadsheet still circulates.
Nobody resisted. They had a way of doing the job that worked, and the new way asked for something the old way did not: a password, a tab, a habit, a reason to switch. Each is small; together they exceed the benefit of a task that took four minutes.
The competition is not the old tool. It is the cost of moving.
5. Build into the workflow, not beside it
The version that gets used usually has no interface of its own:
- It happens in the tool they already have — the document system, the practice management system, the inbox. If the work happens in Outlook, the automation happens in Outlook.
- It triggers itself, on an event: a matter opened, a form submitted, a date passing. Not on someone remembering to run it.
- It delivers where the person already looks. A report that arrives is used; one available in a portal is not.
- It leaves the existing route intact at first. Removing the old way before the new one has earned trust makes its failures unrecoverable, and people will not forgive that twice.
Make the automated path less effort at the moment of choice — a form pre-filled with what is already known, a pack assembled by default with a review step, a status that updates itself so nobody has to be chased. If the automated path is more work than the workaround, the workaround wins regardless of how much better the outcome would have been.
6. Capture once, and let evidence be a by-product
Firms assemble audit and assurance evidence by going and finding it: searching inboxes for approvals, establishing which version was current in March, asking people to remember a decision.
The cost recurs at every audit, the quality is worse than contemporaneous records, and it finds problems at the worst possible moment.
The alternative is structural — evidence as a view over records created as the work happened:
- Structured capture at the point of work, not written up later.
- Approvals as records, not emails. Who approved, what version, when.
- Versioning that answers “what applied on this date” — the most common question and the hardest to answer retrospectively.
- Packs generated on demand, in the format the assessor wants, in minutes.
This only holds if capture is genuinely cheap at the moment of work. If recording it properly takes longer than the task, people record it badly and the evidence is worthless — which is the failure mode of most compliance systems.
We built ECS Group Command on this basis across nine service areas: work captured once, offline if needed, with client-ready compliance packs assembling themselves.
7. Where AI fits, and where it does not
Drafting, summarising and extracting are genuinely useful and genuinely reduce effort. Keep a person approving anything that leaves the firm.
Full automation of a professional judgement is a small additional time saving for a large increase in risk, and the first bad outcome ends the programme — including the parts that were working.
8. Measure adoption, not usage
Not usage of the new tool. Whether the old artefact has disappeared: the spreadsheet, the shared inbox, the printed checklist.
If both exist, you have added a system rather than replaced a process, and the total cost is now higher than before you started.
A sequence that works
- Watch the work for a fortnight. Do not ask.
- Rank by total time, filter by stability.
- Pick two or three. Not ten.
- Build them into the tools people already use, triggered by events.
- Run both routes until the new one is obviously easier.
- Retire the old artefact deliberately, and confirm it is gone.
Wondering where your fee-earner time goes? Get in touch, or read about our professional services work.