Almost every article about AI for small business is a list of fifteen tools. It is the wrong shape of advice, because the tool is the last decision, not the first, and buying software before you have identified the task is how businesses end up paying three subscriptions for things nobody uses. Here is a different approach, and some UK numbers worth knowing before you start.
Where the UK actually stands
The government's SME Digital Adoption Taskforce reports that 43% of SMEs surveyed by the British Chambers of Commerce in 2024 had no plans to use AI at all, down from 48% the year before. Movement, but slow movement. The same report puts the prize in context. Raising SME productivity by one per cent over five years would add £94 billion a year to UK GDP, and the research it cites finds firm-level productivity improvements of 7 to 18 per cent per technology adopted, depending on the product. Federation of Small Businesses research found innovative SMEs recorded 14.8% revenue growth. Meanwhile the UK ranks 25th worldwide for future digital readiness in the IMD 2024 index, and Be the Business found UK SMEs invest less in technology and management than their G7 peers. Read together: the gains are real and measurable, most of your competitors have not moved, and the barrier the taskforce identifies is not cost so much as it appearing too difficult and too risky to start.
Find the task before you find the tool
For one week, write down every task that meets three conditions: you do it more than twice a week, it follows the same steps every time, and it produces a predictable output. Do not filter, just log. At the end of the week you will have somewhere between ten and thirty entries. Multiply each by how long it takes and how often it happens. The top three are your automation candidates. Everything below is a distraction for now. This matters because the tasks that feel most annoying are rarely the ones costing the most time. Chasing an invoice feels worse than retyping delivery addresses, but the retyping is probably eating four times the hours.
What typically comes out on top
Across the UK businesses we have worked with, the same handful keep appearing. Retyping data between systems. An enquiry arrives by email, gets typed into a spreadsheet, then typed again into the quoting system, then again into the accounts package. Every retype is a chance to introduce an error and a job nobody wants. This is not really AI, it is plumbing, and it is usually the highest return thing on the list. First response to enquiries. Not replacing you, acknowledging receipt. An immediate reply that confirms the message arrived and says when a human will respond keeps you in the running against competitors who answer within the hour. Layer in answers to the five questions you get asked constantly and a meaningful share never needs your attention. Quote and invoice preparation. If your quotes are assembled from the same components with different quantities, that is a template problem, and templates are cheap to build. Chasing payment. A polite reminder at 7, 14 and 21 days past due, sent automatically, recovers money that otherwise sits unclaimed because nobody enjoyed making the call. Under UK late payment legislation you are entitled to statutory interest and a fixed recovery charge on overdue commercial invoices, and simply stating that in the third reminder tends to accelerate things. Turning notes into documents. Site visit notes into a report, a phone call into a summary, meeting notes into actions. This is where the current generation of AI tools genuinely earns its keep, because it is language work with a human checking the output.
What not to automate yet
Anything where being wrong is expensive and hard to spot. Pricing decisions, anything that goes to a customer without a human reading it, anything involving personal data you have not thought through, and anything where the process itself is still changing month to month. Automating an unstable process just makes the instability faster. Also be careful about publishing AI-written content unreviewed. It reads as generic, which costs you the credibility that makes small businesses worth choosing in the first place.
The UK data protection part, briefly
If customer data goes into a tool, UK GDPR still applies. Check where the provider processes and stores data, whether it trains models on what you submit, and whether you need to update your privacy notice. The ICO publishes guidance aimed at small organisations that is genuinely readable. This is not a reason to avoid automation, but it is a reason to check before you connect your customer list to something. None of this is legal advice, and if you handle sensitive data it is worth a short conversation with someone who does that professionally.
Start with one thing, and measure it
Pick the single task at the top of your list. Time how long it takes now, across a fortnight. Automate it. Time it again. If it did not save meaningful time, stop and try the next one rather than adding a second tool on top of a first that is not working. Businesses that get value from automation almost always got it from two or three things done properly, not from a stack of subscriptions.
Related reading
- What the tasks above look like once they are built with language models rather than rules: LLM agents and where they cut operational cost.
- Automating the first response only pays if enquiries are arriving at all: five things we check when a site has traffic but no enquiries.
- The retyping problem in a trading business usually traces back to the accounting package: Busy, Tally, QuickBooks or custom.
Not sure what to automate?
Tell us how your week runs and we will tell you which parts are worth automating and which are not. If the honest answer is that your processes are not stable enough yet, we will say so rather than sell you something. Free consultation, no commitment. Book a free consultation at devsioservices.com/contact, or read about our AI automation and workflow integration work at devsioservices.com/services/ai-automation.
