AI Personalisation for UK Retailers: The Right Offer at the Right Time
Almost every article about AI personalisation opens with the same figure: 71% of consumers expect personalised interactions. It comes from 1,013 US adults surveyed over two days in September 2021. Now hold two more figures beside it - BCG's survey of more than 23,000 consumers found two-thirds have already had a personalisation that was inaccurate or invasive, and Forrester's 2025 survey found only 53% of online adults actually like being personalised. So the picture is not customers crying out for it. It is customers broadly accepting it, and two-thirds of them already being let down. This guide works from that, and from the data you already have: the till file, the card, the email list. It gives you six triggers that fire on what actually happened to the customer, the holdout test that proves an offer made money instead of moving sales that would have happened anyway, a break-even calculation that decides every discount, the four tests for any pricing or ranking decision, and the UK law that now applies - the Data (Use and Access) Act 2025, PECR fines that jumped from £500,000 to £17.5m in February 2026, and the CMA's unfair commercial practices rules. Nothing in it asks you to buy software.
- Price
- £4.99
- Format
- Instant digital download
- Published
Opens the secure checkout in a new tab. Instant access after payment.
What's inside
- The evidence audit: the 71% "expect personalisation" figure traced to its source - 1,013 US adults, two days, September 2021 - and the counter: two in three consumers have had a personalisation that was wrong.
- The Holdout Test: hold back 10% at random, set the bar in gross margin before you start, read the result after a full purchase cycle. Why every personalisation figure without one is unfalsifiable.
- The 95% AI-failure claim, read properly: MIT said organisations get zero measurable P&L return from CUSTOM GenAI while employee use thrives - plus BCG's rule that 10% of value is algorithms, 20% is data and 70% is people and process.
- The Offer Clock: six triggers that fire on what actually happened to the customer - replenishment, second buy, basket pair, drift warning, return, own-brand switch - with the maths to find your basket pairs.
- The break-even calculation that decides every discount: contribution before divided by contribution after. A 15% cut on a £24.99 kettle costing £13.50 needs a 48% volume uplift just to stand still.
- Personalised pricing: the CMA found no personalised advertised pricing on any significant scale, because of reputational risk. The real exposure is ranking - 61% of 160 e-commerce sites personalised their search results.
- The Four Tests for any pricing or ranking decision, and the DMCC Act: 32 banned practices, CMA fines of up to 10% of global turnover, and a vulnerable-consumer test that counts mourning, divorce and redundancy.
- UK GDPR, the Data (Use and Access) Act 2025 and PECR: the Six Records to keep on file, and why legitimate interests is now open for significant automated decisions. PECR fines jumped from £500,000 to £17.5m on 5 February 2026.
- Nine ways personalisation stalls - cold start, discount dependency, cannibalisation, discount addiction, staff friction, compliance creep - each with the symptom that appears before the money goes.
- Five of the six levers cost no margin and no legal risk: Placement, Range, Bundle, Replenishment, Winback. Price is last, and it is the only one with a legal question attached.
- Real UK practice with the footnotes read: Tesco's "up to £392" Clubcard saving is a top-25%-of-members figure over a year; Sainsbury's Nectar Prices saves £16 on an £80+ shop across ~11,000 products a week.
- A 30-day plan and a one-page decision sheet: three plays, one measured result, and five documented reasons to kill a campaign before it costs you another quarter.
How you get it
- Pay securely through Payhip, the store that processes every order.
- The download link is emailed to you the moment payment clears.
- A PDF, readable on any phone, tablet or computer.