Compliance-heavy and specialist advisory firms are often the most cautious adopters of AI, and for good reason: the cost of a wrong answer is higher than in most other service businesses. That caution is correct. It's also, too often, used as a reason to avoid AI altogether, when the more useful response is to be precise about where it genuinely helps and where it doesn't.
The distinction that matters: research versus judgement
AI is strong at the research and drafting layer: pulling together a first pass on a regulatory question, summarising a long document, structuring a memo before a specialist reviews it. It is not a substitute for the judgement call at the end, the one that carries the firm's name and, in regulated work, its licence. Firms that get value from AI keep that line explicit rather than letting it blur.
Where the return actually shows up
- First-draft research and summarisation, freeing specialist time for the parts that need a person
- Consistency checks across long or repetitive documents, where a person's attention naturally drifts
- Structuring advice memos into a consistent format before a partner reviews the substance
- Surfacing where a matter looks similar to past work, so precedent isn't reinvented each time
Where it doesn't belong yet
The final judgement on a regulatory position, a client-specific risk call, or anything that would need to be defended to a regulator should sit with a named specialist, full stop. Firms that let AI output move toward a client without that review aren't saving time; they're relocating risk to a point where nobody is watching for it.
The firms getting this right aren't the ones moving fastest. They're the ones who have drawn a clear line between the drafting AI can do well and the judgement it can't do at all, and who have built a review step that makes that line real in practice, not just on paper.
Used well, AI in compliance-heavy and specialist advisory work buys back time on the research layer. It doesn't buy back the judgement, and it shouldn't try to.