This is an illustrative example, not a client. Real reports stay confidential, so I invented this practice and every figure in it. The structure, the scoring and the reasoning layout are exactly what you get.

Day 7 report

A 28-person accountancy practice

Three partners, nine qualified, eight part-qualified and juniors, four in bookkeeping, payroll, a practice manager and two admin. They asked whether AI could take work off the desk in a year when they could not hire into it.

Summary

You spend about £343,620 a year on recurring work (166 hours a week across 28 people), and nobody here had added it up before this week.

Roughly a third of that is assembly: taking figures from one place and lining them up in another. That part is worth moving. The rest is either judgement your clients are paying for, or already handled well enough that touching it would cost more than it saves.

The two largest lines in this report are not in my top three. One costs you nothing and is in section 5. The other is the one I am telling you not to start with.

VAT return preparation

21 hours a week, £43,470 a year

New client onboarding pack

12 hours a week, £24,840 a year

Shared inbox triage

8 hours a week, £16,560 a year

The one thing I would not do: start with the year-end accounts file. It is the biggest number in this report and it is what a vendor will pitch you. It runs on whatever state the client’s bookkeeping arrived in, and if the input is messy the draft comes back wrong in a way nobody catches.

Where the time goes

Hours a week across the practice, triangulated instead of guessed once: your own rough numbers on the intake form, checked against what came up in both calls. Where they disagreed I have said so. That gap is often a finding in itself. Costed at a blended £45 an hour, the rate we agreed on day 1, over 46 working weeks.

Recurring work	Who does it	Hours / week	Annual cost
Preparing year-end accounts from client bookkeeping	Qualified and part-qualified staff	44	£91,080
Chasing clients for records and signatures	Practice manager, and every junior in January	26	£53,820
Preparing VAT returns and quarterly filings	The bookkeeping team	21	£43,470
Management accounts and the commentary that goes with them	Qualified staff, reviewed by a partner	18	£37,260
Payroll runs and the exceptions they throw	Payroll	16	£33,120
Drafting client advice letters	Three partners, and two of the qualified staff	14	£28,980
Onboarding a new client: AML, engagement letter, records request	Admin, with a partner on the AML decision	12	£24,840
Sorting and acknowledging the shared inbox	Practice manager and admin	8	£16,560
Billing and credit control	Admin	7	£14,490
Total		166	£343,620

This total is usually the most valuable line in the report, because nobody has ever added it up. It is not a target, and most of it is work you are right to be doing.

Opportunity register

Every opportunity I found, scored against the same schema, ordered by verdict and then by value. The rejections are in here too. A register that is all green reads like a sales document.

Preparing VAT returns and quarterly filings

Do now

Hybrid: it drafts, a person files

HMRC set the rules, they are on paper, and the answer is arithmetic. The 21 hours do not go on deciding anything. They go on pulling figures out of nine different bookkeeping systems and lining them up.

Effort: ~6 days to build, plus a day of your time agreeing the review rules
Risk: Low. A return either reconciles or it does not, and the draft shows its workings.
Reversibility: Complete. The workings are the same spreadsheet you use now.

Onboarding a new client

Do now

Hybrid. The pack assembles, the AML decision does not

The engagement letter, the checklist and the records request are assembly from facts you have already collected. The part that needs a partner is the AML judgement, and that stays exactly where it is.

Effort: ~4 days, plus your standard engagement letter as a starting point
Risk: Low. Nothing goes to a client unsigned, and the AML step stays where it is.
Reversibility: Complete.

Sorting and acknowledging the shared inbox

Do now

Hybrid: it routes and drafts, a person sends

Routing across a short, stable list of categories, checkable at a glance. Whoever opens the queue sees the suggested route and a drafted acknowledgement and can change either.

Effort: ~3 days, plus a month of past email to check the routing against
Risk: Low. A wrong route is visible at once and costs a click.
Reversibility: Complete. Switch it off and the queue is the queue again.

Billing and credit control

Do now

Machine: scheduled, with an escalation rule

Your practice management system already does staged reminders. This is a settings change and a set of templates, not a build.

Effort: An afternoon. No cost.
Risk: Low, though set the escalation to stop at the second reminder for clients on a payment plan.
Reversibility: Complete.

Chasing clients for records and signatures

Do now

Machine. The first two chases only

The largest line in the report, and it needs nothing built. I have put it in quick wins (section 5) because you can do it yourself this month.

Effort: An afternoon. No cost.
Risk: Low.
Reversibility: Complete.

Preparing year-end accounts from client bookkeeping

Do later

Hybrid, eventually

This is the biggest number in the report, and I would leave it alone for now. The work does not need much judgement, but it runs on whatever state the bookkeeping arrives in. If the input is messy the draft comes back wrong in a way nobody catches. Fix the input first: the quick win in section 5 improves the records you get, so measure again after a quarter.

Effort: Not costed yet, deliberately. Re-scope after the intake changes have run a quarter.
Risk: Moderate, and it is the wrong kind. An error here reaches a filed set of accounts.
Reversibility: Complete, but a wrong number that got filed is not.

Management accounts: the figures

Do later

Hybrid: figures yes, commentary no

The pack assembles. The commentary does not, and the commentary is what the client is paying to read. Worth doing once the VAT work has proved the same plumbing, because it draws on the same ledgers.

Effort: ~5 days once the VAT build exists; most of it is reuse
Risk: Low for the figures. The commentary never leaves the partner.
Reversibility: Complete.

Payroll runs and the exceptions they throw

Not worth it

Leave alone

Your payroll software already automates the runs. The 16 hours left are the exceptions: a starter mid-month, a wrong tax code, a director’s dividend. Those exceptions are the reason a person is there, and automating them means automating the judgement, for the smallest saving on this list.

Effort: High, for the least return of anything here.
Risk: Payroll errors are personal, visible and remembered.
Reversibility: Poor. You would have unpicked a working process.

Drafting client advice letters

Not suitable

Human

See section 4. The wording carries professional liability, and the failure mode is a letter that reads correctly and is subtly wrong.

Effort: None
Risk: The exposure is the point. It is not a time question.
Reversibility: You cannot recall a letter once it has gone out.

The three to do first

Written so that someone else could act on them. If a recommendation only makes sense with me involved, it is not specific enough to be in this section.

VAT return preparation

The rules do not move, the inputs arrive in a fixed shape, and you can check every output against a reconciliation. It is the largest thing on the list where you catch a wrong answer the same day.

Time back: 21 hours a week, £43,470 a year
Cost to set up: ~6 days to build
Cost to run: About £90 a month in model and hosting costs
What you provide: Read access to the bookkeeping systems, and one afternoon agreeing what the reviewer must check before filing
Roughly how long: Live for one client’s quarter within three weeks; all nine by the following quarter

What “working” looks like

Nine returns in a quarter where the reviewer changed nothing material. If the reviewer is still rebuilding the workings by hand after a month, it is not working and you should stop.

New client onboarding pack

Assembly from facts you already hold. The one judgement in it, the AML decision, stays with a person.

Time back: 12 hours a week, £24,840 a year
Cost to set up: ~4 days
Cost to run: Negligible. It runs a handful of times a week
What you provide: Your engagement letter, your records checklist, and the list of what differs by client type
Roughly how long: Two to three weeks

What “working” looks like

A partner opens the pack, changes nothing but the AML note, and signs. Measure it at ten clients.

Shared inbox triage

This is the smallest of the three, and the one your team notices first, because it removes an interruption instead of a task.

Time back: 8 hours a week, £16,560 a year
Cost to set up: ~3 days
Cost to run: About £40 a month
What you provide: A month of past email so you can check the routing before it goes live
Roughly how long: Two weeks

What “working” looks like

Nothing sits unacknowledged overnight, and the practice manager stops being the routing table. If people start bypassing the queue, it is not working.

What’s not worth automating

This section earns the fee, and you were probably hoping it would be shorter. Each of these is something a vendor would sell you.

Drafting client advice letters

14 hrs/week · £28,980/yr

Every letter turns on a judgement about that client, and the wording carries professional liability. The problem is that AI writes something that reads well and is quietly wrong, and that gets through review. A partner skimming a tidy letter about a client they know is checking the tone, not the facts. I do not think the time saved is worth it.

What I looked at

Eleven letters from the last quarter, and no two shared a structure. The reasoning in each turned on something your systems do not hold. A conversation, a plan the client mentioned, a view on how HMRC has treated them before.

What would change this

Nothing on the current evidence. If a fixed subset follows a standard structure, the routine ones confirming a filing position you have already agreed, it would be worth another look. On what I saw that is about one letter in nine, so it is narrower than it looks and I would not build for it yet.

The AML decision on a new client

Collecting the documents is clerical, and it is already in the onboarding recommendation above. Deciding what they mean is a judgement you are personally accountable for. A tool that gives a confident answer here makes it easier to stop thinking, not easier to get right.

What I looked at

Your last six onboardings, two of which needed a second conversation before a partner was comfortable.

What would change this

Nothing. This is the one item on the list I would argue against even if it worked perfectly.

Payroll exceptions

16 hrs/week · £33,120/yr

Your software already automates the runs. The hours that remain are the exceptions, and the exceptions are the part that needs a person. Automating what is left means automating the judgement, for the smallest saving on the list.

What I looked at

A month of payroll queries: 31 of them, no two the same shape.

What would change this

If the exception volume ever concentrates into two or three repeating causes, fix the cause instead of automating the handling.

Quick wins you can do yourself

This month, without me. Giving away the work I could have charged for is what makes the rest of this document believable. The first one here is the largest actionable line in the report.

Switch on staged reminders in your practice management system

This is the largest line in the report, 26 hours a week and £53,820 a year, and it needs nothing built. The reminders and the escalation rule are already there, switched off. Writing the four chase templates takes an afternoon and removes about nine of the 26 hours. The rest is the twelve clients who need a phone call, and a reminder was never going to move them.

Add the three fields you chase most to the records request

Most of the chasing exists because the form does not ask. Fix the form before you build anything that chases better.

Turn on duplicate detection in the ledger import

You are already paying for it. It is off on six of the nine client files I looked at.

Stop producing the Friday exception report

Nobody has opened it since March. I checked before suggesting it, and the practice manager spends about 40 minutes a week on it.

Data and governance

Half a page, specific to you, and not a compliance lecture. What your people already do with AI, where client information is going, and what would need to be in place before section 3.

Four of your staff already paste client figures into a free AI tool. Nobody approved that disclosure, and the free tiers of most of them train on whatever you hand over. I'd fix this first, because it costs nothing and needs nothing built.
Your engagement letters do not mention AI at all. Two of your larger clients will ask within a year, and the answer is much easier to give before you have started than after.
Client financial records are personal data where the client is a sole trader or a partnership. Any tool that reads them is a processor and needs a contract that says so.
The three recommendations in section 3 keep client data inside systems you already hold contracts for. That was deliberate, not a coincidence. It is why the VAT build sticks to your own bookkeeping systems instead of a general-purpose assistant.

I invented every figure above. This practice never existed, and real client reports stay confidential. Yours would be about this long, built from two conversations and a handful of your own documents. I take no system access and no data extracts, so the numbers come from what you tell me and what we work out on the call.

How the audit works

Get started

Want one of these for your firm?

£2,500 fixed for one week and three hours of your time.
The report lands on day 7 and you own it.

Book a call