AI Use Cases in Professional Services: Where the Non-Billable Hours Go

A professional services firm sells time. That single fact decides which AI use cases are worth building and which are decoration.

Your profit is what remains after the hours nobody can bill. Conflict checks. Engagement letters that sit in a partner's inbox for a fortnight. Timesheets reconstructed on a Friday afternoon from calendar entries and memory. Bills that take three drafting rounds because the narratives read like a filing cabinet. Proposals assembled by four people who each rebuild a credentials section that already exists somewhere on the shared drive. None of it reaches a client invoice, and all of it consumes the people whose time you sell.

So the useful question is not whether AI can write a report. It is which specific pieces of unbillable and low-value billable work an agent can carry end to end, what it reads, what it writes back, and where a qualified person signs. Below are the places we see that work in accountancy, audit, consulting, architecture and engineering practices and agencies.

Engagement setup and conflict checks

New work starts with a gate that is pure overhead. Someone searches the practice management system for the prospective client, its parents and subsidiaries, its directors, and the counterparties named in the matter, then decides whether the firm can act.

An agent does the searching properly, which humans rarely have time to do. It reads the enquiry, extracts every named entity, resolves them against company registry data and your client and matter history, and checks connected parties rather than just the exact string someone typed. It writes back a conflict memo listing each entity searched, each hit, the matter it relates to, the responsible partner, and whether the relationship is live or closed.

Cleared searches with no hits can be recorded automatically under a rule you set. Anything with a hit goes to the conflicts partner with the evidence already assembled. The decision stays a partner decision. What disappears is the two days of waiting and the risk of a missed subsidiary.

Client onboarding and KYC

Onboarding is where a signed engagement stalls. Identity documents, beneficial ownership, source of funds, sanctions and PEP screening, and an anti-money-laundering risk rating, all chased by email from a client who has already agreed to pay you.

An agent runs the collection and the assembly. It requests documents against a checklist derived from the client type and risk band, reads what arrives, extracts and validates the fields, compares stated ownership against registry filings, runs screening, and flags what is missing or contradictory in specific terms rather than as a generic reminder. It drafts the client due diligence file and the risk rating with the reasoning attached.

The money laundering reporting officer or engagement partner approves the rating. The agent never clears a hit and never sets a risk band on its own authority. The trail shows every source it consulted and the person who signed, which is what a regulator or a quality reviewer asks for.

Scopes of work and engagement letters

Most engagement letters are variations on a handful of firm templates, degraded by copy and paste from the last similar job.

An agent that has read your template library, your standard terms and your recent engagement letters can draft from the proposal and the fee basis: scope in and scope out, deliverables, assumptions, dependencies on the client, change control, and the fee and billing schedule. It marks each clause as standard, amended or bespoke, so the reviewing partner reads the exceptions instead of the whole document.

It is equally useful in reverse. On client paper, the agent compares the incoming contract against your standard positions and produces a marked deviation list: liability cap, indemnities, intellectual property, payment terms, termination. Legal or the engagement partner negotiates. The agent has removed the reading, not the judgement.

Timesheet and WIP hygiene

Unrecorded time is a direct loss and never appears in any report, because by definition it was never entered.

An agent with access to calendar, email, document activity and messaging can draft each fee earner's timesheet: probable matter, duration, and a narrative in your house style, presented for confirmation rather than posted silently. People correct a draft far more readily than they build one from nothing, and they do it daily rather than weekly, so the reconstruction guesswork falls away.

The same agent watches work in progress. It reads open matters against budget and fee basis, and raises specific exceptions: this fixed-fee job has passed eighty per cent of budget at forty per cent of deliverables, this matter has unbilled time older than ninety days, this engagement has recorded time with no live letter behind it. Each exception arrives with the underlying entries attached, addressed to the responsible partner.

Billing and narrative preparation

Bill preparation is often the largest single block of partner time that no client will ever pay for. The draft comes out of the system, the narratives are unusable, and someone rewrites them under time pressure.

An agent prepares the bill pack: time entries grouped into phases a client will recognise, narratives rewritten to describe the work and its purpose in the client's own vocabulary, disbursements checked against receipts, and time flagged that should probably be written off with the reason. Where a client has billing guidelines, it checks the draft against them before a human sees it, rather than after the client rejects the invoice.

The billing partner decides the write-offs and releases the bill. That authority never moves. What moves is the ninety minutes per bill spent rewriting narratives.

Assembling audit and diligence evidence

Audit fieldwork and transaction diligence share a shape. A request list, a shared folder that arrives disorganised, and juniors spending days matching documents to requests before any analysis begins.

An agent reads the request list and the client's uploaded population, matches documents to items, checks that dates and periods are what was asked for, and chases the gaps by name. For sampled items it pulls the supporting evidence together, ties amounts back to the ledger or the data room index, and drafts the working paper with each figure linked to its source. Exceptions are written up as exceptions.

In due diligence the same pattern produces a first-pass findings log: contracts with change of control provisions, customer concentration, unusual revenue recognition, employment liabilities, each with the document reference. The engagement team then does what it is paid for, which is deciding what the findings mean.

Proposals, pitches and knowledge reuse

Bid work is the purest form of non-billable cost. It is also the work most degraded by not knowing what the firm already has.

An agent that has read your past proposals, credentials, CVs, methodologies and closed engagements can produce a structured first draft against an incoming brief or tender: a compliance matrix mapping every requirement to a response, the relevant credentials, the right team CVs tailored to the brief, and the methodology sections your firm actually uses. It cites which past document each section came from, so the pursuit lead can verify rather than trust.

The same knowledge layer pays off during delivery. A consultant or engineer asking whether the firm has handled a particular problem gets an answer drawn from real prior work, with the engagement, the partner and the document named, subject to your confidentiality rules on who may see which client's material. That constraint is a build requirement, not an afterthought.

Resourcing

Staffing decisions are made in weekly meetings from stale spreadsheets, which is why utilisation and development both suffer.

An agent reads live availability, skills, grades, chargeability targets, holiday, and the pipeline with its probability weightings, then proposes a staffing plan per engagement with the reasoning stated: this person for the sector experience, this one because their current job ends next Friday, this gap needs a contractor or a delayed start. It surfaces the conflicts early, when they are still solvable. The resourcing partner assigns people. Machines should not decide who gets career-making work.

Where AI fails here

Some of this firm's most useful advice is what to leave alone.

Signed opinions and certifications stay human

An audit opinion, a valuation, a structural certification, a tax position taken on a client's behalf. These carry professional liability and, in most cases, a personal signature backed by a licence. An agent can assemble the evidence and draft the supporting papers. It cannot form the opinion, and no firm should let it appear to.

Judgement under genuine ambiguity

The material judgements in professional work are the ones where the standards do not settle the question and reasonable professionals disagree. A model will produce a confident answer anyway. That is precisely the failure mode you cannot afford, so we build these as escalations with the evidence attached rather than as automated calls.

The client relationship

Reading a boardroom, knowing which finding needs a phone call before it appears in a report, deciding when to tell a client something they will not want to hear. None of that is a document problem.

Bad inputs and thin history

Knowledge reuse works in proportion to what your firm has actually kept. If past engagements live in individual mailboxes and the document management system is a graveyard, the retrieval will disappoint. That is a fixable problem, but it is the first project, not a side effect of another one.

We also refuse to automate anything where accuracy cannot be measured on your own contested cases. If we cannot show you a number on the hard examples from your own files, we will not put the agent in front of clients.

What to build first

Start where the work is high volume, low judgement, and already governed by a checklist your firm can articulate. Conflict checks, onboarding, timesheet drafting and bill preparation usually qualify. Measure the agent against your own recent files, including the awkward ones, before it touches live work, and keep the sign-off exactly where professional responsibility already sits.

If you want to talk through which of these fits your practice, write to hello@adoptaipartners.com.