Which processes should a small business automate first?

Start with the work your team repeats every week, that follows clear rules and that already lives in email, documents or spreadsheets: admin, routine replies, data entry and reports. These tasks are easy to measure before and after, and a mistake there is cheap to catch. Leave judgement calls, sensitive decisions about people and anything you cannot check for last.

More businesses are trying this now. The UK government’s AI adoption survey report 2026, published on 29 October, found that 32% of UK businesses with five or more employees used at least one AI technology in 2026, up from 16% in 2025. Most of them are still at the early stage, though, and the same survey shows where they got stuck.

What did the 2026 government AI survey find?

AI use doubled in a year, but it is concentrated in a few everyday functions and held back mainly by skills and an unclear business case. The survey is based on 3,503 telephone interviews carried out between 7 January and 21 April 2026, so it reflects the first half of this year rather than the last few months.

The figures most useful to a smaller business owner:

• Adoption was higher among larger firms: 52% of businesses with 250 or more employees used AI, compared with 32% across all businesses surveyed.
• Among businesses using or planning to use AI, the most common area was administration (81%), followed by marketing (71%) and IT (66%).
• Efficiency and productivity was the main reason for adopting, given by 71%.
• The typical tasks were drafting content, summarising information, research, writing emails and producing documents.
• 67% of businesses already using AI, and 78% of those planning to, said limited AI skills or knowledge held them back. Among businesses not using AI, 65% pointed to the lack of a clear business case or proven return.
• 56% of adopters said they were concerned about bias and accuracy.

The Office for National Statistics measures the same trend in a different way. Its article Artificial intelligence in UK businesses: 2023 to 2026 shows the share of firms with 10 or more employees using AI rising from about 12% in September 2023 to about 35% by June 2026. The average number of AI technologies per adopting firm only went from about 1.4 to 1.6, which the ONS reads as limited transformation so far. Most adopting firms also reported no change in headcount.

So most businesses are using AI for a narrow set of office tasks, and the main obstacle is knowing where to point it. An earlier ONS study of management practices found the same thing in 2023: the most common barrier, cited by 39% of firms, was difficulty identifying activities or business use cases.

How do you decide which process to automate first?

Pick a process that scores well on four questions: how often it happens, how predictable it is, how easy the result is to check and how much it costs you when it goes wrong. You can answer all four on paper with the people who do the work.

Frequency comes first because savings multiply. A task that takes ten minutes and happens forty times a week is a better candidate than a two-hour task you do once a quarter. Predictability matters because automation follows rules; if three people in your office would handle the same email in three different ways, the process needs agreeing before it can be automated.

Checkability is easy to overlook. If a member of staff can glance at the output and spot an error in seconds, you can run the automation with a person approving each result and remove that step later. If an error only surfaces weeks later in a customer complaint or a VAT return, the process is a poor first project.

Finally, think about the cost of a mistake. A wrongly filed supplier invoice is an annoyance. A wrong quote sent to a customer, or an automated decision about a job applicant, is a different level of risk.

Which processes are good first candidates?

For most small and medium businesses, the best first projects sit in administration and customer communication, the same areas where the government survey found AI was most used. The table compares common starting points by the four questions above.

Process	How often	How easy to check	Cost of a mistake	Good first project?
Reading supplier invoices into your accounting system	Daily or weekly	Easy: totals and supplier names can be matched	Low if a person approves before posting	Yes
Sorting and drafting replies to routine customer emails	Daily	Easy: staff read the draft before sending	Low with approval, higher if sent automatically	Yes
Weekly sales or stock reports built from spreadsheets	Weekly	Easy: compare with last week’s manual version	Low	Yes
Order taking over WhatsApp or web forms	Daily	Moderate: needs product and price data to be accurate	Medium: wrong orders cost money	Often, once product data is clean
Answering customers with a chat assistant	Daily	Harder: answers are generated in real time	Medium to high: wrong answers reach customers directly	Later, with clear limits
Screening job applicants or deciding on credit	Varies	Hard	High, with legal duties attached	No

Invoice and document processing is a sensible first choice because the inputs are structured and the output is easy to verify. Email triage is similar: the AI sorts incoming messages and drafts a reply, and a person presses send. Reporting automation often needs little AI at all, since pulling data from your systems into a regular report is mostly an integration job, and it can remove a lot of copying and pasting.

How do you run a first automation project?

Keep the first project small and easy to undo. A sensible sequence looks like this:

1. Write down the current process step by step, including who does it, how long it takes and where errors happen. Time it for a week or two.
2. Choose one process using the four questions above, and agree what success means in numbers (minutes saved per week, error rate, response time).
3. Check what data the process touches. If it includes personal data about customers or staff, read the data protection section below before going further.
4. Build the simplest version that works, with a person approving every output at first.
5. Run it alongside the manual process for a few weeks and compare the results against your baseline.
6. Remove the approval step only for the cases where the automation has been reliably right, and keep a person in the loop for the rest.
7. Only then pick the next process.

Step 1 is often the most useful. Mapping the process often shows that part of the problem is a missing rule or a duplicated spreadsheet, and fixing that costs nothing. It also gives you a baseline, which answers the “no clear business case” worry that 65% of non-adopters raised in the government survey.

Where does AI automation fall short?

AI tools that generate text can be confidently wrong, so any task where accuracy matters needs a check built in. More than half of adopters in the government survey worried about bias and accuracy, and the practical answer is process design: approval steps, limits on what the system can do on its own and logs you can review.

Some work is also a poor fit. Tasks that depend on relationships, negotiation or judgement about an unusual situation still need a person. Processes that change every month will break any automation built around them. And if your data is spread across paper, personal inboxes and three versions of the same spreadsheet, the first job is getting it into one place, which may be a custom software or integration project before it is an AI project.

What about customer data and UK GDPR?

If an automation handles personal data, data protection law applies to it in the same way it applies to the rest of your business. The Information Commissioner’s Office publishes guidance on AI and data protection, covering principles such as data minimisation, accuracy and security. Read it, or ask your data protection adviser, before connecting an AI tool to customer or staff records. Starting with processes that involve little or no personal data, such as supplier invoices or stock reports, keeps your first project simpler.

Frequently asked questions

What is the easiest process for a small business to automate with AI?

Usually document handling and routine email: reading invoices into an accounting system, or sorting incoming emails and drafting replies for staff to approve. Both are frequent, follow clear rules and are easy to check.

How many UK businesses use AI in 2026?

The government’s AI adoption survey found that 32% of UK businesses with five or more employees used at least one AI technology in 2026, up from 16% in 2025. The ONS puts the figure for firms with 10 or more employees at about 35% in June 2026.

Do I need technical staff to start automating?

No, but someone in the business has to own the process and judge whether the results are right. Skills were the most common barrier in the government survey, so plan for training or outside help.

Should AI make decisions without a person checking?

Not at first. Keep a person approving outputs until the automation has a track record, and keep decisions that significantly affect people, such as hiring, under human control. The ICO’s guidance explains the data protection side.

How much does AI automation cost?

It depends on the process and the systems it connects to. OpusGate publishes its starting prices on the pricing page and gives a fixed quote before any work begins.

If you have a process in mind and are not sure it is the right place to start, book a free call. We will look at it with you and tell you honestly whether it is worth automating, what it would cost and how long it would take. You can also read more about our AI and automation work.