Should we automate one process or launch a firm-wide programme?

What does a firm-wide programme look like six months in?

It usually starts with enthusiasm and a structure. A steering group is formed. Each department nominates someone. A readiness assessment is commissioned, a long list of use cases is scored for effort and impact, and licences are bought for everyone. Somebody volunteers to write the policy.

Six months later, the honest status report reads like this. There is a policy and a ranked list of twenty ideas. Two or three keen people use an assistant for their own drafting, in their own way, with nobody checking the output. Every other job is done exactly as it was a year ago, except that the firm now pays a monthly bill for software most people opened twice.

Nothing went wrong in a way anyone can point to. That is the problem. The programme was busy from the first week, and busy is hard to argue with. But attention was spread across every department at once, and attention is the one input that does not divide.

Why does breadth fail where depth works?

Rebuilding how a job gets done is detailed work. You have to sit with the people who do it and follow every input, every template, every handoff and every check. You have to find out that the senior associate keeps her own version of the engagement letter because the one on the system has a clause she dislikes. You have to learn that the report cannot go out until a particular spreadsheet has been reconciled, and that the reconciliation is the actual bottleneck, not the writing.

That level of detail takes a focused week for one process. Try to do it for eight processes at once and you do not get eight maps. You get eight surveys: a page of bullet points per process, accurate at the level of a flow diagram and useless at the level of a rebuild. Build from the diagram and the output will not match how anyone works, so the team quietly reverts. Why AI rollouts fail sets out that pattern in more detail.

Breadth struggles for a second reason too. A programme must keep every department included, so it moves at the pace of the most sceptical part of the firm. One process needs only the people who touch it.

Isn't one process too small to matter?

Only if you pick a small one. The hours in a professional services firm are not spread evenly across a hundred tasks. They sit in a few places: writing up and reporting, producing bids and proposals, keeping clients updated, and assembling evidence for compliance and audit. Those are large, repeated jobs that senior people cannot delegate cleanly.

Work out the size before you argue about it. Take the hours one of those jobs costs a fee earner in a normal week, multiply by the number of fee earners doing it, multiply by your own charge-out rate, and multiply by forty-six working weeks. The unbillable hours calculator does the arithmetic, and the guide explains what to count. The figure for a single process is usually large enough to stop the conversation about whether it is worth doing.

How is one process different from a pilot?

This matters, because the two get confused and they are opposites. A pilot is usually designed to be reversible. It runs alongside the real process, often on sample work, with a small group who agreed to try it. Nobody depends on it. At the end there is a review, and because the pilot never carried real pressure, the review concludes that it was promising and more work is needed.

Rebuilding one process is narrow, not shallow. The old way stops. The new way is how the job is now done, on live work, with the firm's own documents and templates, and a named person signing off anything that reaches a client. Narrow scope, full depth. That distinction is also why this is not the same as chasing a quick win: see the trouble with quick wins for why easy and visible are poor selection criteria.

What does the first process buy you beyond the hours?

Three things, and two of them are worth more than the time saved.

• A method you have tested on yourselves. After one rebuild you know how your templates behave, how your review step should work, what your people will tolerate and what they will not. Every later process starts from those answers instead of from a blank page.
• An argument that is hard to dismiss. A partner who distrusts all of this will not be moved by a maturity model. A colleague saying the monthly report now takes twenty minutes and reads better might move them. Getting the partners to agree and the partner who says no both come down to evidence from inside the firm.
• A real decision about what comes next. Before the first rebuild, the list of candidates is guesswork. Afterwards you know which of your processes share the same bottleneck, and the sequence picks itself.

Which process should go first?

Look for a job that happens often, that several fee earners do, that produces a defined output somebody already recognises as right or wrong, and that has an owner who can make decisions without convening a committee. Avoid anything where the firm has not agreed what good looks like, because no system can produce a document nobody has defined. How to choose a process gives the full set of tests, and the process priority scorer ranks your candidates against them.

In practice most firms land on something close to client reporting, proposal and bid production or compliance evidence. Which one depends on where your work concentrates. An accountancy firm in busy season has a different answer from a practice that lives on tenders.

When is a broader programme actually right?

Later, and smaller than the word suggests. Once two or three processes have been rebuilt, a firm has a house style for this work: how output is reviewed, where records are kept, who owns what. At that point sequencing more processes is sensible, because each one is a variation on a method that already works.

One thing should not wait, though. People are already using these tools on firm work whether or not anyone approved it, which is the subject of shadow AI, and a short policy costs little and prevents a lot. Governance can be firm-wide from day one. Delivery should not be.

What would you do this month?

Pick the job your senior people complain about most. Spend a week finding out what it really costs, in hours and in delay. Agree what the finished output should look like, and who signs it. Then set a date by which the job is done the new way, and treat that date as progress rather than meetings held. Six months on, check it against the standard you agreed, which is what good looks like after go-live describes.

If you are not sure which job to start with, the audit takes about three minutes and names one.