How Small Firms Are Buying AI Tools Without a Technical Hire

More UK small firms are buying AI tools than ever, but a growing share are also abandoning them. The British Chambers of Commerce found that active AI use among UK SMEs rose from 25% in 2024 to 35% by September 2025, a genuine progress on paper. Set against that, research into why AI projects fail keeps landing on the same root cause, and it isn’t the technology itself.

A 2025 survey by data management firm Informatica found the top obstacles to AI success were data quality and readiness, cited by 43% of respondents, and a lack of technical maturity, also 43%, with a shortage of skills and data literacy close behind at 35%. None of those is a reason to avoid AI. There are reasons to buy and implement it differently than most small firms currently do.

This piece is about that buying and implementation process, specifically: how to choose the right tool, evaluate a vendor properly, budget for the real cost rather than the sticker price, and set it up successfully, when nobody on the team has a technical background to lean on.

Why the Failure Pattern Isn’t About the Tool Itself

McKinsey’s 2025 research on AI adoption found a specific, actionable pattern among the organisations reporting genuinely significant financial returns: they were twice as likely to have redesigned their actual workflow before selecting which AI technology to use. In other words, the businesses getting real value didn’t start by picking a tool and then figuring out where to use it. They started by understanding exactly what was broken, and only then went looking for something to fix it.

That ordering matters enormously for a small business with no technical hire to catch a bad decision before it’s made. Analysis of failed AI implementations consistently finds that projects starting with a technology decision, rather than a defined business problem, are dramatically more likely to be abandoned. The pattern is recognisable: someone sees a demo, reads an article, or gets pitched by a vendor, comes back enthusiastic about the possibilities, and only afterwards tries to work out where in the business that enthusiasm actually applies. That’s backwards, and it’s expensive to get backwards.

What Getting It Wrong Actually Costs

It’s worth being concrete about the financial stakes here, because “we’ll just try it and see” is a more expensive strategy than it sounds. Industry analysis of failed AI implementations puts the typical cost of a rushed, poorly evaluated vendor selection at somewhere between £15,000 and £75,000, once failed setup work, wasted subscription time, and the cost of unwinding the decision are all counted, a meaningful sum for a small business to absorb on a purchase that delivers nothing.

A large share of that cost isn’t the software itself. Data preparation alone is estimated to consume 60 to 80% of a typical AI project’s budget, with integration work adding a further 40 to 60% on top, meaning the advertised subscription price is often a small fraction of what implementation actually ends up costing once a business accounts for getting its own systems ready to work with the tool. Budgeting only for the monthly fee, without setting aside meaningful time and, in some cases, cost for data preparation and integration, is one of the most common ways a small business ends up spending considerably more than expected on a tool that still doesn’t work properly.

Define the Problem Before You Look at Any Tool

The starting point isn’t “which AI tool should we buy?” It’s a specific, measurable answer to what’s actually broken: which task takes too long, where the business loses money, and what customers repeatedly complain about. A vague goal like “we should use more AI” gives a vendor nothing to solve, and gives you no way to judge afterwards whether the purchase actually worked.

A useful test is whether the problem can be stated as a number. “Invoice processing takes four hours a week” or “we lose roughly a fifth of enquiries because nobody replies fast enough outside office hours” are problems a tool can be judged against. “We want to be more efficient” is not, and buying a tool against that kind of goal is how a subscription ends up running quietly in the background for a year, doing very little, because nobody defined what success was meant to look like in the first place.

A worked example: consider a small independent retailer with a website receiving a steady stream of customer enquiries by email, most asking about stock availability, delivery times, or returns. The founder’s instinct is “we need a chatbot.” Pushed to be specific about the actual problem, it turns out roughly 40% of enquiries are the same three questions, and the average reply currently takes six hours during busy periods, by which point many customers have already bought from a competitor instead.

That reframing changes the evaluation entirely. The tool doesn’t need to handle every possible customer question intelligently; it needs to answer three specific, predictable questions instantly and hand everything else to a person quickly. That’s a far narrower, cheaper, and more achievable brief than “get us a chatbot,” and it gives the business a genuine number to judge the pilot against afterwards: has the average reply time on those three question types actually dropped, and has that translated into fewer lost sales during busy periods?

Why Data Readiness Matters More Than Most Founders Expect

This is the part that catches non-technical buyers out most often, because it has nothing to do with the AI tool itself and everything to do with what’s already sitting in the business. Informatica’s research identified data quality and readiness as the single most common reason AI projects fail, tied with technical maturity at 43% each. In practice, this usually looks unglamorous: a customer listed under three slightly different names across two systems, sales figures exported from an old platform that don’t match the format the new one expects, or basic records that were never kept consistently enough for a tool to learn anything useful from them.

Separate analysis has found that poor data quality alone can drain a meaningful share of annual revenue for affected businesses, not through any single dramatic failure, but through the accumulated cost of decisions made on faulty information and the wasted time spent working around inconsistent records. Before evaluating any vendor, it’s worth a genuinely honest look at whatever data the tool would actually need to work with. If that data is scattered, inconsistent, or incomplete, no amount of vendor capability will fix that on day one, and a demo that looks impressive using the vendor’s clean sample data can be genuinely misleading about how the tool will perform on your own messier version.

What to Ask When Choosing AI Tools With Nobody Technical in the Room

A non-technical buyer is at a real disadvantage in a sales conversation with a vendor whose job is to make their product sound essential. A short, consistent set of questions closes most of that gap without requiring any technical background.

Does It Integrate With What You Already Use

A tool that can’t talk to your existing accounting, scheduling, or customer records software creates more administrative work, not less, however impressive its own individual features look in a demo. Asking specifically how integration works, not whether it’s possible in theory, but what setup it actually requires, is one of the highest-value questions a non-technical buyer can ask.

What Happens to Your Data

This matters for two separate reasons: data protection obligations and simple business sense. Ask plainly whether the tool trains its underlying model on your business data, how long data is retained, and what happens to it if you stop using the service. A vendor who can’t answer this clearly and specifically is worth treating with real caution.

What Does Onboarding and Support Actually Involve

Ask who helps set the tool up, what happens when something goes wrong, and whether ongoing support is included or billed separately. A tool that’s simple to demo but genuinely complicated to configure properly is a common source of the “we bought it and never really used it properly” outcome that quietly wastes a subscription for months.

Can You Actually Leave If It Doesn’t Work Out

This is the question non-technical buyers ask least often and need most. Understand what happens to your data and your workflow if you want to switch providers later: can you export your data in a usable format, is there a contract minimum, what does the exit process actually involve. Vendor lock-in is far easier to avoid before signing than to escape afterwards.

Ask for a Reference From a Business Like Yours

A vendor with genuine small business customers should be able to point to one in a comparable sector and size, not just a large, well-known logo used for marketing. If they can’t, that’s informative in itself.

Budgeting for the Real Cost, Not the Advertised One

Given how much of the cost of an AI project sits in data preparation and integration rather than the subscription fee itself, it’s worth building a rough budget that accounts for all three before comparing vendors on price. A tool that looks cheaper on its monthly fee can end up considerably more expensive overall if it requires extensive data cleanup or custom integration work that a competing, slightly pricier tool would have avoided by connecting more easily to what the business already uses.

It’s also worth asking each vendor directly what a realistic total first-year cost looks like, including setup, training time, and any integration work, rather than relying on the headline subscription price alone. A vendor unwilling or unable to give an honest answer to that question is, again, informative in itself.

Running a Small Pilot Before Committing Fully

Rather than adopting a tool business-wide immediately, a short, deliberately limited pilot, commonly four to six weeks, in a single area of the business is a far lower-risk way to test whether a tool actually delivers against the specific, measurable problem it was bought to solve. Setting one clear goal for the pilot, such as a defined reduction in the time a specific task takes, gives a non-technical team an objective way to judge the outcome rather than relying on a general impression of whether the tool “feels” useful.

Assigning one person to actively review the tool’s output during the pilot, rather than assuming it’s working correctly in the background, is a simple step that catches problems early and builds genuine confidence in the tool before it’s relied on more widely.

Who Should Own This Decision When Nobody’s Technical

Rather than treating an AI purchase as purely a technology decision handed to whoever’s most comfortable with software, the businesses that implement AI successfully tend to involve the staff who’ll actually use the tool day to day in choosing it, rather than presenting it to them as a finished decision after the fact. Change management research on failed AI projects consistently finds that staff who feel a tool is being imposed on them, particularly where job security feels threatened, are considerably more likely to quietly work around it or under-report problems, which undermines the pilot’s ability to reveal genuine issues.

In practice, this doesn’t require a dedicated hire. It requires naming one person as the clear point of contact for the decision and the pilot, ideally someone who’ll actually use the tool rather than a founder evaluating it from a distance, and being explicit with the wider team that the goal is reducing tedious work, not reducing headcount.

Common Mistakes Worth Naming Directly

Beyond the core process, a handful of specific, recurring mistakes show up across failed small business AI purchases. Adopting several AI tools simultaneously, rather than proving one works before adding the next, tends to overwhelm a small team’s capacity to properly learn or troubleshoot any of them.

Choosing a solution built for a much larger organisation, simply because it looked more impressive in a sales pitch, often means paying for capabilities the business will never use, while still requiring the same basic setup effort as a right-sized alternative. And expecting a tool to fix a process problem on its own, without any change to how the business actually works, tends to produce the same result as before, just with a subscription fee attached.

Getting a Second Opinion Without Hiring Anyone

For a genuinely significant purchase, it’s worth getting an outside perspective before signing, even without a technical hire on staff. This doesn’t have to mean an expensive consultant: a peer in a similar business who’s already bought a comparable tool, a local enterprise support scheme, or even a paid hour with an independent IT advisor to review a shortlist and ask the harder technical questions can catch problems a non-technical buyer might miss.

The cost of an hour or two of outside advice is small next to the £15,000 to £75,000 a poorly evaluated purchase can end up costing, and a second opinion is often enough to catch an unrealistic vendor claim or a data readiness gap before it becomes an expensive lesson.

For businesses in Northern Ireland or the Republic of Ireland, this kind of outside opinion is often available free or at low cost through existing support schemes, rather than needing to be paid for privately. In the Republic, Local Enterprise Offices run a Digital for Business service with expert digital consultants available to review a technology decision before it’s made, and the Grow Digital Voucher specifically supports the cost of adopting new digital tools.

In Northern Ireland, Go Succeed offers free mentoring to early-stage businesses, while Invest NI’s mentoring scheme connects supported clients with experienced business mentors for exactly this kind of strategic decision. None of these requires a technical hire on staff; they exist precisely to give a non-technical founder a second, informed opinion before committing.

Red Flags Worth Walking Away From

A small number of warning signs are worth treating as a hard stop rather than a minor concern. A vendor promising the tool solves every problem instantly, or fully deploys within days for a genuinely complex use case, is a common marker of overpromising rather than realistic expectation-setting. A vendor who can’t clearly explain what happens when the tool gets something wrong, or can’t describe any scenario where human review is still required, hasn’t thought through the failure modes of their own product. And a proposal with no clear, specific problem statement attached, just a general pitch about efficiency or competitiveness, tends to produce exactly the vague, unmeasurable outcome that makes a tool hard to judge six months later.

Frequently Asked Questions