The people the rollout forgot.
Firmwide AI programmes are, almost without exception, attorney programmes. The other half of the building does work these tools handle beautifully, and receives nothing. This is the cheapest available return in a legal AI deployment.
01 The arithmetic that created the gap
Nobody at any firm ever decided that business services staff should not learn to use AI. What happened is subtler and entirely structural.
AI spend gets justified in billable hours. That arithmetic works cleanly for fee earners — hours recovered, realisation, matter margin — and does not work at all for a records manager. So the business case is written for attorneys, the training budget follows the business case, and everyone else receives a licence, a launch email, and silence.
Depending on the firm, somewhere between a third and half of headcount is not a fee earner. In a 2,000-person firm that is several hundred people who do document-heavy, writing-heavy, deadline-heavy work — the exact profile these tools serve best — with no structured training at all.
02 Where the work actually is
| Group | What compresses |
|---|---|
| Paralegals | Chronologies, indexing, transcript and deposition summaries, first-pass review, exhibit preparation, cite-check support |
| Marketing / BD | Pitch and RFP drafting from precedent, matter descriptions, directory submissions, event follow-up, tailoring one credential twenty ways |
| Finance / billing | Narrative review against client guidelines, variance explanation, reporting commentary, chasing correspondence |
| HR / recruiting | Job descriptions, interview guides, policy drafting, first-pass CV screening with human decisions |
| Records / IG | Classification, retention analysis, index and metadata cleanup, policy drafting |
| Secretarial / admin | Correspondence, meeting and call notes, formatting, calendar and travel logistics, document comparison |
Look at that list and notice something: a lot of it is writing under time pressure to a house standard. That is the single task current AI does best.
A firm has budget to train exactly one group properly. Partners bill the most; associates are the largest fee-earning population; paralegals are cheaper but do the most repetitive document work. Which group returns the most on the training?
Pick one, then open.
There is a real case for each, and the honest answer is that it depends on what the firm is short of. But the case for paralegals is stronger than most firms assume, for a reason that has nothing to do with rates.
Partners have the least time to learn and the most discretion to opt out. Associates are the standard answer and a reasonable one. Paralegals, though, do the highest volume of exactly the tasks these tools handle well, have the most consistent workflows — which means training transfers rather than evaporating — and are the most likely to still be doing the same work in two years. Consistency of workflow is what makes training stick, and it is systematically undervalued when the business case is written in billable hours.
03 What to teach this group (it is not the same curriculum)
Attorney AI training tends to lead with research and drafting sophistication. For business services staff the priorities are different and, frankly, more basic — which is why vendor training rarely fits.
Permission, first and unambiguously. This group is the most uncertain about what they are allowed to do and the least likely to ask. Uncertainty produces one of two bad outcomes: they use nothing, or they use a consumer chatbot on their phone because nobody told them not to. The second is an actual confidentiality problem and it is created by silence.
Their own work, not legal work. A pitch document, a billing narrative, a records policy. The demo matter that lands with associates is irrelevant here and signals that the training was not built for them.
Verification at their level. Not case-law checking — checking that a figure in a client report matches the source, that a summarised email chain did not drop the one caveat that mattered.
Where to stop. The most valuable thing you can give a non-lawyer is a clear line: this is drafting support, and anything that looks like a legal conclusion goes to a lawyer. People are far more willing to use a tool confidently when they know exactly where its authority ends.
04 The argument to make internally
If you need to sell this, the framing that works is not productivity. It is risk and consistency.
Untrained staff with tool access is the actual exposure. They will use something — the sanctioned platform if they understand it, a free consumer tool if they do not. Training is the cheaper half of that choice, and it is the only version where the firm knows what is happening.
The productivity case is real too, and it is easier to demonstrate than to project: pick one team, train them properly on their own work, and measure something they already track. A pilot beats a business case built on assumed percentages, particularly with an audience trained to distrust assumed percentages.
This week's challenge
Ask one person in business services — marketing, finance, records, admin — two questions: do you have access to the firm's AI tool? and has anyone shown you how to use it for your work? The gap between those two answers is, in most firms, the largest untouched improvement available. It is also free to find.