AI agents for business finance: what they can and can't do
What AI agents can do in bookkeeping, close, collections and compliance today, where a person must stay in charge, and how to judge a product.
The short answer
An AI agent can prepare and send quotes, but the price should be set by rules a person has approved: a list price, a discount band the agent may use without asking, and a ceiling above which a named person decides. The agent should record every discount and its reason, so margins and promises stay visible.
Pricing is where AI in sales is most useful and most dangerous. Useful, because quoting is slow and repetitive and customers expect a number quickly. Dangerous, because a wrong price sent in writing is hard to take back, and a habit of easy discounts quietly erodes margin. The answer is not to keep the agent away from pricing but to give it a clear frame.
The three parts of a pricing frame
- List price
- Your standard price for each product or service, per currency, with a date. The agent quotes from this and nothing else.
- Discount band
- The range the agent may apply without asking, for example up to 5% for orders over a set size, or a fixed early-payment discount.
- Ceiling
- The point above which the agent stops. A request for 12% when the band is 5% goes to a named person with the history attached.
Some businesses add conditions to the band: discounts only for annual contracts, only for repeat customers, never on certain products. Keep the rules short enough that you could explain them to a new employee in two minutes. If you cannot, the agent will struggle too.
How to set a sensible band
Start from margin, not from what competitors do. If a service has a 40% gross margin, a 10% discount removes a quarter of the profit on that job. A simple way to choose the band:
- Work out the gross margin on each main product or service.
- Decide the lowest margin you will accept without discussion.
- Set the band so that the maximum discount keeps you above that floor.
- Set conditions that make discounts earn something: volume, upfront payment, longer commitment.
- Review the band each quarter against what was actually given.
| Item | List price | Gross margin | Band | Margin at maximum discount |
|---|---|---|---|---|
| Monthly bookkeeping (illustrative) | £450 | 55% | Up to 7.5% | 51% |
| Website build (illustrative) | £6,000 | 35% | Up to 5% | 32% |
| Hardware resale (illustrative) | £1,200 | 12% | None | 12% |
In the table, hardware has no band because the margin is too thin to discount without approval. That is a perfectly reasonable rule for an agent to follow, and one that tired people often forget late on a Friday.
What the agent should do above the ceiling
When a customer asks for more than the band allows, the agent should not refuse the customer, and it should not quietly give in. It should:
- Acknowledge the request and say it will come back by a set time.
- Send the request to the named approver with the deal value, history, margin at the requested price and any competing quote mentioned.
- Suggest alternatives inside the band, such as a smaller scope, longer term or faster payment.
- Record the decision, who made it and why, once it is made.
The record is not bureaucracy. Discounts given once tend to be expected again at renewal. Knowing that a 15% discount was a one-off for a launch, approved by the owner, makes next year's conversation easier.
Risks to watch
Two risks stand out. The first is the agent quoting a price it was never allowed to quote, through a misread rule or a confusing request. Courts and tribunals have been willing to hold businesses to what their automated systems told customers; the 2024 Air Canada chatbot case in British Columbia is the example most often cited. The second is drift: dozens of small, in-band discounts that together cost more than one large approved one. A monthly summary of discounts given, by size and reason, deals with both.
Consumer law also matters if you sell to individuals. In the UK, for example, the Consumer Rights Act 2015 and related rules govern pricing information and contracts with consumers. Check the rules where you sell and take advice on your terms.
A worked example
Illustrative. A Manchester IT support firm lets its agent quote managed-service contracts with up to 5% off for annual prepayment. A prospect asks for 15% to match a rival. The agent replies within the hour that it will confirm by the next morning, and sends the owner the request: £14,400 a year at list, 38% margin, 23% margin at 15% off, with the rival's quote attached. It suggests 5% for prepayment plus a reduced onboarding fee. The owner offers 8% on a two-year term. The agent sends the revised quote, logs the 8% with the owner's name and reason, and adds a note to the renewal record that the discount was tied to the two-year term.
The owner spent five minutes on the decision because the numbers he needed were already in the request. Over the quarter, the monthly discount summary showed he had approved four above-band requests and declined two, and that in-band discounts averaged 3.1%, a figure he was comfortable with.
Where Pryme fits
The Pryme Sales agent drafts proposals and quotes from your live pricing rules. Discounts inside the band go out; anything above stops and waits for a person, and the promise made to win the deal is written to a register. Pricing and contractual terms sit outside its scope. The Chief of Staff brings pending approvals to you alongside other decisions. For drafting, see checking AI proposals, and for decisions that belong with people, human in the loop.
Questions people ask
Can an AI agent change my prices?
It should not. It should quote from a list a person maintains and apply only the discounts your band allows.
What is a discount band?
The range of discount the agent may apply without asking, often with conditions such as volume or upfront payment. Requests above it go to a named person.
How do I stop discounts creeping up?
Review a monthly summary of discounts by size and reason, and tie discounts to something the customer gives back, such as a longer term.
What if a customer accepts a wrong quote?
Whether you are bound depends on the law where you trade and the circumstances. Take advice, and reduce the risk by keeping the agent inside clear rules with a record of every quote sent.
Sources
- Moffatt v. Air Canada, 2024 BCCRT 149 (CanLII) · accessed Oct 2026
- Consumer Rights Act 2015 (legislation.gov.uk) · accessed Oct 2026
This article is general information, not financial, legal or tax advice. Products described are provided under the terms on this site and are subject to regulation and eligibility.