AI Is Changing the Economics of Serving Small Businesses

A small business can be an excellent customer in almost every respect except the one suppliers tend to notice first: how much money it has. The owner needs useful software, reliable advice and somebody who will answer a question without first establishing which department should receive it. They cannot necessarily fund an account manager’s mortgage.

This creates a familiar arrangement. Large organisations receive implementation support, tailored systems and people who learn how they work. Smaller businesses receive a login and an invitation to visit the knowledge base. The difference is described as a product tier, although it is also a fairly precise account of who can afford attention.

AI could alter that calculation. The important possibility is not simply that small businesses can generate more marketing copy. It is that services previously too expensive to deliver to them might become commercially practical.

The market is large. The accounts are small.

The UK government estimated 5.64 million small businesses at the start of 2025, defining them as businesses with fewer than fifty employees. They represented more than 99% of the private-sector business population.

That sounds like an irresistible market until somebody examines the cost of serving one customer. A large number of small businesses does not automatically produce a profitable business selling to them. Each still has to discover the product, understand it, get started and obtain enough value to remain.

Some of those costs do not shrink neatly with the customer’s size. Explaining an import problem to a sole trader can require as much attention as explaining it to a larger company. The invoice at the end is considerably less ambitious.

The resulting temptation is to move upmarket. Bigger customers can be more demanding, but at least their demands sometimes arrive attached to a budget.

The expensive part is often the translation

Consider a small contractor trying to introduce a new system. Their information exists across emails, photographs, messages and a spreadsheet whose column headings made sense when it was created.

The difficulty is not necessarily that the business is badly run. It is that its working knowledge has accumulated in forms convenient for the people doing the work. Converting that knowledge into somebody else’s software takes effort.

Traditional software often makes the customer perform the conversion. Choose the category. Complete the fields. Learn our terminology. Decide whether the thing you call a job is what we call a project, an opportunity or an engagement.

AI offers a way to reduce some of this translation work. A system can propose structured information from an email, organise a set of notes or help someone describe what they need in ordinary language. These are capabilities to implement and test, not permission to assume every extracted detail is correct.

The commercial significance is that an interaction which previously required a person on the supplier’s side may become cheaper to support. The customer no longer has to arrive already speaking fluent software.

There is evidence behind the possibility

Research on support work gives this argument a firmer foundation. In Generative AI at Work, published in 2025, Erik Brynjolfsson, Danielle Li and Lindsey Raymond found that AI assistance increased issues resolved per hour by 15% on average. Gains were larger among less experienced and lower-skilled workers.

That was a study of customer-support agents in a particular organisation. It was not evidence that a small-business software company can remove its support team or improve its margins by precisely the same amount.

It does demonstrate a mechanism that matters: useful guidance can help people resolve problems more efficiently. If that transfers to a supplier’s own work, the amount of human attention required per customer can fall without simply refusing to provide it.

That is a more interesting proposition than a chatbot designed to make reaching a human sufficiently exhausting that the customer abandons the attempt. A support cost has not disappeared merely because it now occurs in somebody else’s evening.

Small businesses are already experimenting

The demand is not entirely hypothetical. An OECD survey of more than 5,000 SMEs across seven countries, including the UK, found generative AI in use at 31% of surveyed businesses. Among users, 65% reported improved employee performance, while 26% reported increased revenue.

These were reported experiences from a survey conducted in late 2024, not experimentally established gains. They also concern SMEs’ own use of AI, rather than the profitability of companies serving them.

Nevertheless, the difference between performance and revenue is instructive. Making a task easier does not immediately create additional sales. It may instead give the owner an hour back, reduce a backlog or allow something previously neglected to happen.

For a small business, that can be sufficient value. The person buying the service may want to finish the paperwork before dinner, not produce a case study about transformation.

A better service, not a busier owner

The strongest opportunity may therefore be to sell more complete assistance at a price smaller firms can sustain. Help prepare the work, identify missing information and make the next decision straightforward.

Imagine an owner forwarding an enquiry and receiving a draft response, the relevant customer details and one question about the date. Compare that with being asked to configure a workflow, define a taxonomy and attend an onboarding webinar.

The second arrangement is sometimes necessary. But software suppliers should be careful about presenting their own setup requirements as evidence that the customer is insufficiently sophisticated.

The economic test is whether the supplier can deliver the simpler experience reliably, including the exceptions. If a useful draft costs little but correcting an invented commitment takes an hour, the attractive demonstration has concealed an expensive service.

The owner does not need an assistant with boundless initiative. They need one that knows when to ask.

The costs do not evaporate

There are limits to this opportunity. Customer acquisition still costs money. Integrations still break. Payments fail, records conflict and people disagree about what was promised.

Some decisions also carry consequences that cannot sensibly be treated as text-generation tasks. A proposed quote is not an agreed price. A draft reminder is not permission to contact every customer. A plausible explanation is not a verified answer about the business’s records.

Good products must distinguish preparation from commitment. They need dependable records, clear permissions and a way to recover when an action’s outcome is uncertain. The conversational interface does not excuse the machinery underneath.

This is why cheaper model usage alone will not determine the winners. A supplier might spend very little generating an answer and still lose money resolving the consequences. The relevant cost is serving the customer successfully, not producing the response.

Attention becomes affordable differently

If the economics improve, the result need not be a fully autonomous service. A small team with better tools may support customers who previously could not justify the team’s time.

That could favour specialists. Someone who understands how a particular trade operates may need fewer questions, recognise an exception sooner and design a service around the actual working day. Broad technical capability does not automatically supply that judgement.

It could also favour existing providers with trusted relationships. A bookkeeper or software supplier may use AI to extend what they offer rather than surrender the customer to a new company with a more animated homepage.

The opportunity is open, but it is not ownerless territory. Small businesses already have ways of getting things done. A new service must improve on those arrangements, including their familiarity.

The customer is not the implementation team

The most consequential change would be a reduction in the amount of work a small business must do before somebody can help it. More services could become economical without requiring customers to become larger, more standardised or better at administering software.

That is an argument about what AI can enable, not a claim that the outcome is guaranteed. Suppliers can use efficiency to improve service, reduce prices or simply remove the last reachable person. The technology does not choose the business model.

The useful test is ordinary. Can a business with limited money and no spare administrator obtain something that previously required both?

If the answer is yes, the market changes. If the answer is another dashboard the owner must learn to maintain, we have mainly found a cheaper way to assign them homework.