What do people mean when they say the data is a mess?
Ask ten partners whether their firm's data is ready for AI and most will say no. Press for detail and the answer splits into separate problems that have nothing to do with each other.
- Documents saved in three places: the document management system, a shared drive and somebody's inbox.
- Naming conventions that were agreed once and quietly abandoned.
- A practice management system with fields nobody fills in, because nobody ever uses them.
- Four versions of the same template, each with a champion.
- Knowledge that exists only in the head of whoever has been there longest.
All of these are real. None of them is one problem, and none of them has to be solved before a single job can be rebuilt. Treated as a single blockage called "our data", they become an immovable object. Treated separately, most of them turn out to be irrelevant to the work in front of you.
Why is "fix the data first" such an effective way to do nothing?
Because it sounds responsible and it has no edge. A firm-wide tidy-up has no definition of done. There is no point at which anyone can stand up and say the files are now clean, because new ones arrive every day and the standard was never written down. So the work gets a sponsor, a working group and a slot on the quarterly agenda, and it stays there.
Meanwhile the hours carry on leaking. The report that takes a senior person a day to assemble still takes a day. The bid that eats a weekend still eats a weekend. The cost of waiting is not zero, it is simply invisible, because nobody records it anywhere. That is the argument we make in doing nothing versus acting now.
There is a second reason the advice persists. Tidying data is safe. It upsets nobody, changes no one's day and cannot fail. Rebuilding a process means deciding how the job will be done, which means someone has to choose between the four templates. The clean-up is often a way of postponing that choice.
How much of your data does one process actually touch?
Take a quarterly client report. Walk through what the person producing it actually opens: the previous report, the current figures from one or two systems, the firm's report template, any notes from the review meeting, and the sign-off step at the end. That is the whole universe. Ten years of archived matters sitting on the same drive are not part of it. Their state is irrelevant, because the process never opens them. Client reporting is a good example precisely because it looks data-heavy and is not.
Bid production is wider but still finite: past bids worth reusing, CVs in a current format, scheme or case descriptions, accreditations, insurance details and the standard answers to the questions that come up every time. That is a list you can write on one page. It is not the firm's entire history. Proposal and bid production depends on those inputs being current, not on everything else being filed correctly.
This is the point the blanket advice misses. Data readiness is not a property of a firm. It is a property of a process. A firm with a chaotic drive can have perfectly clean inputs for one job, and a firm with immaculate folders can still have a bid library full of outdated accreditations.
What does a scoped clean-up look like?
It is a short, finite piece of work, done in the week the process is mapped rather than in a separate programme.
- Name the one job you are rebuilding.
- List every input it uses: documents, templates, fields, systems, the message from the client that starts it.
- For each input, find the version that is actually current and correct.
- Decide where that version lives and who maintains it.
- Fix those inputs, and only those.
Most of the time steps three and four are the whole job, and they are decisions rather than labour. Which template is the real one. Which figure is authoritative when the finance system and the tracker disagree. Who owns the standard text about the firm's quality procedures. The guide to preparing templates sets out how to do this for the documents, and how to document a process covers the mapping around them.
Which parts of the mess genuinely do have to be fixed?
Three, and they are narrower than people expect.
A single source of truth for each input. If two versions of a template are in circulation and both are defended, nothing built on either one will hold. This is the problem described in why templates rot, and it gets worse when the versions live in different places. Pick one. Archive the others.
Fields the process actually reads. If a draft depends on a matter type or a fee basis that is often left blank, you have two options: fill it in for live work going forward, or stop depending on it and take the information from somewhere more reliable. What you do not need to do is backfill every historic record.
Who can see what. Before anything reads client documents, you need to know which information is restricted, which clients have terms limiting how their material is handled, and where the data sits. That is a confidentiality and data protection question, not a tidiness one, and it is covered in data protection and AI on client work.
Everything else, the naming conventions, the duplicate folders, the inbox archaeology, can wait. It is housekeeping. It is not a precondition.
Doesn't the clean-up stick better if you do it properly first?
The opposite, in our experience of how these things go. Tidying with no process to serve has no acceptance criteria, so standards drift back within months. Tidying in service of a live process is enforced by the process itself: if the system draws the firm's standard text from one agreed location, that location stays current, because the output is wrong the moment it does not.
The same logic applies to the spreadsheet that half the firm depends on. Rebuilding the job around it exposes what it really holds and who really maintains it, which a tidy-up never does. That is a theme we pick up in the spreadsheet that runs the firm.
Where should a firm start?
Pick the job that costs the most time, not the one with the cleanest inputs. Work out what that time is worth using hours, fee earners, charge-out rate and forty-six working weeks, which the unbillable hours calculator will do for you. Then list the inputs that one job touches and look at them honestly. In most firms the list is shorter than the fear suggests.
If you are not sure which job to take first, how to choose a process sets out the criteria, and the audit will point at one in about three minutes. Start there, fix what that process needs, and leave the rest of the drive for another day.