News, with a practical perspective.

We link to the original sources below. Sections labelled “Sandycroft analysis” contain our interpretation and practical suggestions. Product and service links are commercial suggestions from Sandycroft, separate from the source reporting.

The news

The British Chambers of Commerce published an AI economy analysis on 19 November 2026. It says BCC surveys found AI use among UK firms rose from 25% in 2024 to 54% in early 2026, but argues that headline adoption does not show how much work the technology actually performs.

The analysis cites BCC and University of Essex research in which 45% of AI-adopting firms used only generic tools and 10% had built something of their own. The BCC’s March summary of that research also said 95% of SMEs using AI reported no change in workforce size over the previous year and 86% said job roles were unchanged. These are survey findings, not a guarantee of savings or a forecast for an individual business.

Choose one workflow with a clear finish line

Do not begin with a broad instruction to “use AI more”. Choose one repetitive process with a defined input, output and owner. Good candidates are sorting quote enquiries, turning meeting notes into a task list, drafting a first response from an approved template, or checking that an order record contains every required field.

Write down the current steps and the normal time taken before introducing a tool. The trial should have a beginning and an end: for example, an enquiry arrives, required details are extracted, missing information is flagged, and a staff member approves the reply. If the boundary is unclear, neither time saved nor mistakes can be measured reliably.

Put human gates around costly decisions

Keep a named person responsible for prices, contracts, employment decisions, customer promises and production approval. AI can prepare or flag information, but it should not invent a price, confirm stock, approve artwork or send a binding response without a controlled check.

Protect customer and staff data as well. Use only approved systems, remove information that the task does not need, and decide what must never be pasted into a prompt. Record the source document and final approver so that a mistake can be traced and corrected.

Test the workflow against real work

For an embroidered-workwear enquiry, a trial could extract garment, colour, size, quantity, logo position and deadline into a checklist. A person then checks availability, pricing, artwork and delivery before sending anything. For business printing, the trial could flag missing bleed, low-resolution artwork, an absent delivery date or a mismatch between approved quantity and purchase order.

Those examples use AI to reduce retyping and omissions, not to replace embroidery knowledge, print checks or customer service. Start with a small batch of completed and live jobs, including awkward cases. Compare the AI-assisted result with the existing process and keep the original route available during the trial.

Measure useful outcomes for four weeks

Track minutes per job, correction rate, missed fields, customer response time and staff intervention. Also record failures: confident but wrong text, duplicated work, privacy concerns or extra checking that cancels the saving. A polished draft is not a productivity gain if someone must rebuild it.

After four weeks, decide to stop, adjust or scale. Scale only when the workflow has a documented owner, approved inputs, human checkpoints and a measurable improvement. The BCC evidence supports a sober conclusion: value is more likely to come from changing how a task is completed than from adding another licence to an unchanged process.

Sources & dates