How to Find AI Savings Line by Line in Your Profit and Loss

Print last year's profit and loss, then for each cost line over about 2% of turnover ask three questions: how much of this is people's time on repeatable admin, how much is paid to outsiders for work AI can draft, and how much is waste AI could spot? Put a cautious figure against each, then subtract the AI tools' cost.

The catch is that most AI savings are hours, not cash. An hour saved only reaches the accounts if it stops overtime, shrinks a supplier's bill, removes the need for a hire, or gets spent on work you can charge for. Say AI saves your office manager four hours a week: if those hours were paid overtime, that's cash; if they're quietly absorbed into the rest of the week, your P&L won't move at all.

Follow me on Instagram@sagnikteaches

Sort every saving into cash, capacity or avoided cost

Before touching the P&L, agree three labels, because mixing them is how AI business cases get inflated.

Connect on LinkedInSagnik Bhattacharya

• Cash: a bill that gets smaller. A cancelled subscription, a subcontractor invoiced for fewer hours, overtime that stops. You'll see it in next quarter's accounts.
• Capacity: the same people get time back. Real, but it only turns into money if that time goes on billable work or growth. Otherwise it gets absorbed by longer lunches and slower emails, which is fine, but it isn't a saving. Why AI saves time but not money covers how to capture it.
• Avoided cost: a hire or contract you no longer need. Only count it if you genuinely would have spent the money this year.

Every figure on your worksheet gets one of these labels. When you total up, cash and avoided cost go in the headline; capacity goes underneath, reported separately.

Subscribe on YouTube@codingliquids

Get the P&L in a form you can work with

Ask your bookkeeper or accountant for the last 12 months, month by month if possible, with cost of sales (the direct costs of doing jobs) separated from overheads. For the three or four biggest overhead lines, ask for the ledger detail too: the individual payments that make up the total. "Professional fees: $18,000" tells you nothing; "bookkeeper $13,200, legal $2,300, payroll bureau $2,500" tells you where to look.

Then cross off the lines AI won't move, so you don't waste time on them: rent, rates, insurance, depreciation, loan interest, and anything under about 2% of turnover. What's left is usually eight to twelve lines. If reading the statement itself is the hard part, using AI to understand your profit and loss explains each section first.

The line-by-line worksheet

Go down the remaining lines with this table beside you. The "planning assumption" column is a cautious starting point for your own estimate, not a benchmark; replace it with your measured numbers as soon as you have them.

The software line deserves a special note: it usually goes up when you add AI. Run an audit of your AI subscriptions at the same time, because cancelling a tool your new plan already includes can pay for the new plan outright.

The subcontracted line hides a trap when the supplier charges a flat fee. Suppose a physiotherapy clinic's bookkeeper bills $900 a month, fixed. The clinic starts photographing receipts into a capture app and asks, in a short email, how many of the bookkeeper's hours go on keying them. The reply is about six of the roughly eighteen hours a month. That's a real reduction in the bookkeeper's work, but on a fixed fee none of it reaches the clinic's P&L until the fee is renegotiated. Write it on the worksheet as "cash, subject to agreement", raise it at the next fee review with the hours in hand, and don't count it before then. A fair outcome might be a lower fee, or the same fee with a monthly management summary the clinic never used to get.

The recruitment line needs the same scepticism. An estate agency planning a part-time admin hire at around $24,000 a year might be tempted to write the full figure down as avoided cost once AI starts drafting property descriptions and viewing follow-ups. Before doing that, list what the hire was actually for. If half the reason was answering the phones on Saturday mornings, AI drafting doesn't touch it, and the honest entry is the share of the role that the automation replaces, if any, and only once it has run reliably for a quarter.

Ask the people who do the work, not the ledger

The P&L shows what a line costs, not what the time inside it is spent on. For the wage lines, ask each office or admin person to keep a simple log for one normal week: every 30 minutes, a word or two on what they were doing. It feels tedious, but a week is enough to show patterns nobody had noticed, such as a full morning every Monday spent re-typing the weekend's enquiries into the diary.

Here's what one morning of that log might look like from the office administrator at a kitchen-fitting company (illustrative):

MON 08:30  weekend web enquiries -> retyping into diary and CRM
    09:00  same (11 enquiries, 2 duplicates)
    09:30  same; rang 2 people for missing phone numbers
    10:00  supplier call: worktop delivery moved to Thursday
    10:30  rebooked 2 fitters and texted both customers
    11:00  chasing 3 overdue invoices by phone (1 answered)
    11:30  quote write-up from fitter's site notes (photos + scribbles)
    12:00  same, finished; emailed to customer

Two patterns show up in a single morning. Ninety minutes went on copying enquiries from one system to another, which is a job for a simple form-to-diary automation, possibly with no AI in it at all. And the quote write-up took an hour from rough notes, which is exactly the kind of drafting AI shortens. The supplier call and the rebooking are judgement work; leave them alone.

Then sit down with them and ask five questions about each recurring task:

1. How many times a week do you do this, and how long does each one take?
2. Is it the same steps every time, or does it change?
3. Where does the information come from, and where does it end up?
4. What goes wrong with it, and how long does fixing that take?
5. If this took half the time, what would you do with the hours?

The last answer tells you the capture rate. "Catch up on the backlog of quotes" is capacity that turns into revenue. "Not sure" means the saving will probably evaporate unless you plan where the time goes.

Turn hours into money honestly

For any line where the saving is time, use this calculation:

Annual value = hours saved per week
             x loaded hourly cost
             x 46 working weeks
             x capture rate

Loaded hourly cost = hourly wage + employer costs (pension, payroll
                     taxes, holiday cover). Your bookkeeper can give
                     you this figure per role.

Capture rate = 1.0 if the time replaces overtime or a supplier's hours
               0.5 if roughly half goes on billable work
               0   if nobody can say where the time went

The capture rate is the honest part. Most owners skip it and end up with a business case that looks brilliant and never shows up in the accounts.

To see how much it swings the answer, take a dental practice receptionist whose AI phone summaries and reminder drafts save three hours a week, at a loaded cost of $22 an hour. Without a capture rate, that's 3 × $22 × 46 = $3,036 a year, and it's tempting to put that figure in the case. If the practice decides those hours go on ringing patients who are overdue for a check-up, and about half the time genuinely does, the capture rate is 0.5 and the honest figure is $1,518. If nobody decides, the rate is 0 and the saving is $0, however real the three hours felt. Same tool, same hours, three different answers. The difference is a decision about where the time goes, made before you switch the tool on.

Worked example: a landscaper's P&L, line by line

An illustration, not a real business. Say a 14-person landscaping firm turns over $1,150,000 a year, with four crews and two office staff. Here is the worksheet after an afternoon with the owner and office manager:

Cash and net savings total $18,288, plus $1,288 of freed office time reported separately. Against that:

• ChatGPT Business for three people on annual billing: 3 × $20 × 12 = $720 a year.
• Zapier Professional on annual billing at $19.99 a month for the reminder automations: about $240 a year.
• Setup: roughly 30 hours of the office manager's time at $28 an hour loaded, $840 one-off.

Year one comes to about $16,500, roughly 1.4% of turnover. Not dramatic, and that's the point: an honest line-by-line review of a hands-on business usually finds a handful of solid savings rather than a transformation. Two of the figures (fuel and bad debts) are assumptions, so I'd plan on around half until they're measured. The materials line may turn out to be the biggest prize, but only once profit per project is tracked well enough to show where the over-ordering happens.

A prompt to get a first pass from AI

AI is useful for spotting candidate lines you'd overlook, as long as you treat its output as a list of questions, not answers. Remove names, bank details and payroll by individual first.

I run a [type of business] with [number] staff and turnover of about
[amount]. Below is my profit and loss for the last 12 months, line totals only.

For each cost line over 2% of turnover:
1. Say what kind of work or purchase usually sits inside a line like this
   in a business like mine.
2. List any part that is repeatable admin, outsourced drafting or
   avoidable waste that AI tools could reduce. If none, say "none".
3. Say what I would need to measure to confirm the saving.
Do not estimate percentages or give industry averages. Do not do any
arithmetic; I will calculate the figures myself.

[paste P&L lines]

The last two instructions matter. Chat assistants readily invent "industry average" savings figures, and they make arithmetic slips; keep the numbers in your own spreadsheet.

Even with those instructions, check what comes back. For a small print shop, part of an illustrative reply read:

Line: Professional fees ($16,500, 3.1% of turnover)
1. Usually bookkeeping, accountancy and occasional legal work.
2. Bookkeeping hours spent keying supplier invoices and receipts
   could be reduced with receipt capture and bank rules.
3. Ask your bookkeeper how many hours a month go on data entry.

Line: Wages - studio ($148,000)
1. Designers and print operators.
2. AI design tools could typically cut studio time by 20-30%.
3. Track hours per job before and after.

The first entry is exactly what you want: a specific place to look and a specific question to ask. The second broke the rule. "Typically 20-30%" is an invented average, and applied to a studio wage bill it would put a large, imaginary number at the top of the worksheet. Delete the figure, keep the measurement suggestion, and ask the studio staff which parts of a job are repetitive before you estimate anything.

Where AI can quietly add cost

• Review time. Drafts need checking. If the office manager saves four hours writing and spends two checking, the saving is two hours.
• Usage-based charges. Automations billed per task or credit can creep as volume grows. Look at the usage page monthly for the first quarter.
• Quality slippage. Bringing marketing in-house saves the fee but can cost enquiries if the posts get worse. Track enquiries alongside the saving.
• Setup that never finishes. Half-built automations cost staff time and deliver nothing. Finish one line before starting the next.

Check after 90 days whether the savings are real

Write down each line's starting figure before you change anything; setting a baseline before you introduce AI shows what to capture. After a full quarter, compare against the same quarter last year and adjust for volume, because a busier quarter pushes fuel and wages up regardless. Cost per job is a fairer comparison than the raw total.

Then sort each line into three piles: confirmed (cash visibly lower, keep going), unproven (no change yet, give it one more quarter with better measurement) and failed (no change, or costs rose, so stop or rethink). One confirmed line is enough to justify the next round. The landscaper above would expect the overtime, bookkeeping and software savings to confirm first, because they show up as smaller bills within a month or two.

At the 90-day mark, the landscaper's sorted review might look like this (illustrative):

Notice that the fuel line uses cost per job, not the total, and that the bad-debt line got worse. That is useful rather than embarrassing: it shows the reminders are only as good as the contact details behind them.

Follow-up questions about AI savings and your P&L

Is it safe to upload my profit and loss to an AI tool?

A summary P&L with line totals is low risk on a business plan that doesn't train on your content, such as ChatGPT Business or Claude Team. Remove payroll detail by name, bank account numbers and customer names first. On a personal plan, check the model-training switch in privacy settings before uploading anything financial, and never upload the full nominal ledger to a free account.

What percentage of costs can AI realistically save a small business?

There's no reliable single figure, and published percentages rarely match a small firm's cost structure. In hands-on trades most spending is materials and site labour, which AI barely touches, so savings come from a few admin-heavy lines. The illustrative landscaping example here lands near 1.4% of turnover in year one. Work out your own figure line by line rather than borrowing one.

Should I build AI savings into next year's budget?

Only savings you have already seen for at least one full quarter, and only the cash kind: a cancelled subscription, overtime that stopped, a smaller bookkeeping bill. Leave capacity gains out of the budget until the freed hours are visibly doing billable work. Budgeting on projected savings is how owners end up cutting a role before the automation is reliable.

Further reads

• How to Calculate AI ROI for Your Business (Worked Example) — Turn the worksheet totals into a proper return figure.
• How Much Time Can AI Really Save a Small Business Each Week? — Realistic weekly hours by task, before you estimate.
• Can AI Help a Small Business Grow Revenue, or Only Cut Costs? — The revenue side your P&L review will miss.
• Chasing Unpaid Invoices With AI: Polite Reminders for Trades — The quickest win on the bad-debt line.
• How to Review an AI Tool After 90 Days: Keep, Fix or Cancel — Decide keep, fix or cancel once the numbers are in.
• AI Receipt Capture and Bank Categorisation: A Bookkeeper's Setup — How receipt capture cuts bookkeeping hours.
• How Much Should a Small Business Budget for AI? — Itemised AI budget lines, three budgets by team size, a ceiling formula based on hours saved, and a kitchen fitter's first year month by month.
• What AI Costs a Small Travel Agency and What It Saves — The monthly AI bill for a one-to-three adviser travel agency, set against a task-by-task ledger of the hours it can give back.
• Budget vs Actual: How to Explain Monthly Variances With AI — A materiality rule, price and volume splits, a context-notes template and an AI prompt, worked through a craft brewery's month line by line.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: ChatGPT Business and Zapier pricing as listed on the vendors' pricing pages (checked September 2026). All business figures are illustrative.