A few months ago I spoke with the owner of a property maintenance firm. She had two distinct revenue streams: sixty or so standard boiler services a month, priced consistently for years, and three or four large refurbishment contracts a year, each negotiated around the client, the scope, and the relationship. An AI pricing tool had caught her eye. It promised revenue uplifts with no extra headcount. Her question was direct: would it work for both types of job?
She already suspected the answer was no.
What choice are you actually facing?
You probably aren’t choosing between AI pricing and no system at all. You’re deciding which decisions to hand to a model and which to keep in the room. The real split is between repeatable, data-rich work, where AI can test price points faster than any human, and bespoke, relationship-driven contracts, where the model has no useful signal and guessing is expensive.
Around 35 to 39 percent of UK SMEs were actively using AI tools by mid-2025, according to research from the Mole Valley Chamber and the Business AI Alliance, but pricing-specific adoption clusters heavily in sectors with high transaction volumes: accommodation, e-commerce, utilities, standardised professional services. For a business with a mix of work, the sensible move is to treat the two revenue streams as separate problems rather than applying a single tool across both.
On one side are the jobs that look the same each time. On the other are the contracts where the client, the context, and the stakes vary significantly. AI is genuinely useful on the first side. On the second, it can give you a number that looks credible while missing everything that actually determines the price.
When does AI do the heavy lifting well?
AI earns its place in pricing when you have enough transaction history for a model to learn from and when your market is transparent enough to benchmark against. UK businesses in travel and hospitality report revenue uplifts of ten to twenty percent from AI-assisted dynamic pricing when it is supervised rather than left to run unsupervised. Volume and data quality are the deciding factors.
Three situations make AI the stronger choice. First, you complete enough of the same transaction each month for the model to detect patterns, estimate demand, and find price points you would not have tested manually. Second, your service can be compared directly with competitors, so the model has a visible benchmark to work from: standard consulting retainers, routine maintenance, and commoditised professional services all fit this profile. Third, you need to run scenario tests, such as what happens to margin if you raise prices by five percent, or what annual billing does to churn compared with monthly, faster than any analyst could produce them.
Promotional pricing is another area where AI tends to earn its keep with lower risk. Letting a model optimise discount offers, trial pricing, or seasonal rate adjustments is less exposed than handing over your base list price entirely. Several UK CRM platforms and e-commerce tools already include AI-assisted testing for this type of work, typically available for around twenty pounds per user per month.
When should a human make the call?
Human judgement needs to dominate when each deal matters individually, when the relationship drives more value than the rate, or when you serve people whose circumstances put them at disadvantage. In these situations the data is thin, the stakes are high, and a model cannot price in what you read in the room. Negotiation, trust, and strategic positioning are not variables an algorithm can quantify.
The pattern is visible in project-based services. If a single contract represents twenty percent of your annual revenue, mispricing it by twenty or thirty percent can erase most of your profit for the year. AI trained on a handful of past engagements is working from too little information and too much noise to be the decision-maker.
There are also regulatory dimensions worth understanding. The Competition and Markets Authority has warned that algorithmic pricing tools can produce tacit coordination between competitors without any explicit agreement, essentially by all firms following the same automated logic and arriving at the same prices. If your pricing tool uses competitor data to set rates, you need to understand how it does so and whether the CMA would see the result as softly collusive.
Under UK GDPR, personalised pricing that significantly affects an individual requires explainability and, in some cases, human review. The ICO’s guidance on automated decision-making makes this clear: accountability stays with the business, not the algorithm. Walking away and letting a model decide is not a defensible governance position.
What does it cost to get this wrong?
Getting the split wrong tends to cost in one of two directions depending on which way you err. Over-automate in the wrong context and you risk systematic underpricing that quietly erodes margin, reputational damage if customers compare notes on differential prices, or a regulatory challenge you did not anticipate. Under-automate where AI would genuinely help and you leave money on the table month after month.
UK SMEs that use AI for commercial decision-making report cost savings of twenty to forty percent in the relevant workflows and meaningful improvements in working capital, according to industry research. The upside of appropriate automation is real. So the question is not whether to use AI in pricing, but where.
During the early months of the Covid-19 disruption, the CMA investigated thousands of complaints about apparent price gouging on online platforms, a number of which involved automated or semi-automated pricing that had not been adequately supervised. The reputational damage was significant, even for businesses that had not deliberately set out to exploit the situation.
The NCSC raises a less visible but equally important concern: feeding your rate cards, discount rules, or client-specific terms into consumer-grade AI tools creates confidentiality risk. If those tools use your input for model training, your commercial pricing logic stops being confidential. Enterprise-grade accounts with explicit data-use terms are the minimum standard for any pricing work.
What to ask yourself before you decide?
Before you deploy AI in your pricing, four questions will tell you whether you are in the right territory. They cover the data you have, the customers you serve, the sector you operate in, and the governance you can actually sustain. Answer them honestly and the decision about where AI earns its keep tends to resolve itself.
Do you have the data? A model needs at least twelve to twenty-four months of consistent pricing history, structured by channel, segment, and outcome. Thin or inconsistent data produces outputs that look credible but are not reliable enough to act on.
Are your customers in a position where different prices create a fairness concern? If similar clients would see materially different quotes with no plain-language explanation, that is a signal to slow down. The ICO’s guidance on AI and data protection makes explainability a requirement, not an option.
Are you in or near a regulated sector? Financial services, health, education, and utilities all carry regulator expectations around automated pricing decisions. The FCA’s 2022 discussion paper on AI in financial services confirmed that accountability stays with the firm regardless of what the model recommends.
And can you sustain meaningful human review when it matters? If the answer is yes to all four, AI-assisted pricing is worth building into your process. If not, keep AI in the analysis seat and keep humans at the controls until the conditions are right.
If you want to think through where this applies in your business specifically, Book a conversation and we can map it out.
