Talk to us

AI · 8 min read

AI automation for UK SMEs: where to start in 2026

By the Hot Rocket team

Most UK SMEs should start AI automation with one repetitive, text-heavy admin task, such as inbox triage, document data entry or chasing, and keep a person approving anything customer-facing. ONS figures show adoption rising fast but staying shallow, so a single well-measured process is a sensible first step.

The short version

AI use in UK businesses has roughly tripled in three years, but most firms are still using it lightly. For a small or medium-sized business, the best first step is one repetitive, text-heavy process that you can measure before and after. Keep a person approving anything that affects a customer or an employee.

What the official figures say

The Office for National Statistics asks businesses about AI in its Business Insights and Conditions Survey (BICS). The figures are official statistics in development, so the ONS advises caution, but they are the best regular UK measure available.

MeasureFigureSource
All businesses using at least one AI technology, September 202630%ONS BICS, 8 Oct 20261
The same measure when first asked, September 20239%ONS BICS, 8 Oct 20261
Businesses with 10 or more employees using AI, September 202638%ONS BICS, 8 Oct 20261
Businesses with fewer than 10 employees using AI, September 202629%ONS BICS, 8 Oct 20261
Most used: text generation with large language models (10+ employees)21%ONS BICS, 8 Oct 20261
AI users (10+ employees) saying admin and clerical roles were most affected38%ONS BICS, 8 Oct 20261
Average AI technologies per adopting business, late 2023 to 20261.4 → 1.6ONS, AI in UK businesses2
Information and communication firms using AI, versus construction58% vs 13%ONS, AI in UK businesses2

Two things stand out. First, adoption is now mainstream: almost a third of all businesses report some use. Second, it is shallow. The ONS notes that adopters use barely more than one or two AI technologies on average, and that improving operations is the most common purpose.2 Most businesses have tried a chatbot. Far fewer have rebuilt a process around AI.

Where automation pays back first

The ONS finding that admin and clerical roles are the ones most affected matches what we see in practice. The best early candidates share four traits:

  1. High volume, low variety. The same kind of input arrives many times a week.
  2. Text or documents in, structured data out. For example emails, PDFs, forms or call notes.
  3. A clear right answer that a person can check quickly.
  4. A measurable cost today in hours, delays or errors.

Typical examples for UK SMEs:

  • Inbox triage. Classify incoming email, draft a reply and route it to the right person.
  • Document data entry. Read supplier invoices, delivery notes or application forms into your accounts package or CRM.
  • Chasing and follow-up. Draft reminders for overdue quotes, missing documents or unpaid invoices for someone to approve.
  • Knowledge lookup. Answer staff questions from your own policies, price lists and past jobs, with links to the source.

Use our automation ROI calculator to sanity-check any of these with your own numbers. It shows every assumption.

The legal position in 2026

The Data (Use and Access) Act 2025 reformed the UK GDPR rules on automated decision-making. Most of its data protection provisions came into force on 5 February 2026, according to the ICO.3 Article 22 of the UK GDPR was replaced by new Articles 22A to 22D.4

In plain terms:

  • A significant decision (one with a legal or similarly significant effect on a person) taken solely by automated means must come with safeguards. The person must be told, must be able to make representations, must be able to get human intervention and must be able to contest the decision.4
  • Stricter limits remain where special category data, such as health data, is involved.4
  • What counts as "meaningful human involvement" matters. The ICO consulted on updated automated decision-making guidance in spring 2026.5

None of this stops an SME automating admin work. It does mean you should design the process so that a person makes or genuinely reviews decisions about people, such as hiring, credit or eligibility. Record that design in a short data protection impact assessment.

This is not legal advice. We design systems with these rules in mind and document them, but for a decision that affects people's rights, take advice from a data protection specialist.

A sensible first project

  1. Pick one process and measure it for two weeks: volume, time per item and error rate.
  2. Prototype on real, anonymised examples. If the model cannot do the task reliably on paper, stop there.
  3. Build it with a review step. The AI drafts and a person approves. Log everything.
  4. Measure again after a month. Keep it, widen it or switch it off.

That is deliberately unglamorous. It is also how automation earns trust, and trust is what lets you take on the second and third process.

Sources

  1. Office for National Statistics, "Business insights and impact on the UK economy: 8 October 2026" (BICS Wave 165). Official statistics in development. ons.gov.uk ↩ ↩2 ↩3 ↩4 ↩5 ↩6

  2. Office for National Statistics, "Artificial intelligence in UK businesses: 2023 to 2026", 20 July 2026. ons.gov.uk ↩ ↩2 ↩3

  3. Information Commissioner's Office, "Statement on the commencement of the Data (Use and Access) Act (DUAA)", 5 February 2026. ico.org.uk ↩

  4. Data (Use and Access) Act 2025, section 80 (automated decision-making). legislation.gov.uk ↩ ↩2 ↩3

  5. ICO, "ADM guidance consultation". citizen-space.ico.org.uk ↩

T+99Next step

Talk it through with us

A free 30-minute first chat, with no obligation. Tell us what is getting in the way; we tell you honestly what we would do, roughly what it would take, and whether it is worth doing at all.

Book a free first chat →Request a call back