Outlook
Part of Loyalty marketing outlook for England as rewards schemes face a reset
Loyalty marketing AI applications without the hype cycle
A practical guide to loyalty marketing AI applications in England, covering personalisation, referral fraud scoring, consent rules and a pilot scoring rubric.
What to take away
Start with one AI application, not three. For a typical England scheme, run a six to twelve week pilot, depending on purchase frequency.
- A 40,000 member scheme might eventually run offer personalisation, churn scoring and referral fraud checks. Prove one before you buy the next.
- Consent and lawful basis are the gating items. The Data (Use and Access) Act 2025 changes affecting marketing data use are the reference point your team should read before any model touches member data.
- Read the DUAA 2025 provisions that insert Annex 1 to UK GDPR on recognised legitimate interests. Read the ICO's guidance on AI and data protection and its guidance on explaining decisions made with AI.
- Policy direction favours controlled experimentation rather than heavy pre-approval, as set out in the government's AI policy direction relevant to personalisation and loyalty automation.
- Platform change is continuous. Google's own Google Ads feature announcements showing current platform change are a reminder that the ad and offer layer moves faster than your scheme rules.
- Score every pilot against a written rubric before you scale it. A model that lifts redemption but breaks consent is not a win.
Where AI fits: four applications with UK examples
Most England loyalty teams are not building models. They are buying features. Personalised offer selection, next-best-action prompts in the app, churn probability scores and referral fraud detection are the four applications that appear most often in vendor roadmaps.
Four AI applications and owners
Application
- Personalised offer selection
- CRM
- Next-best-action prompts
- CRM
- Churn probability scores
- CRM
- Referral fraud detection
- Risk
Decision owner
- Personalised offer selection
- Next-best-action prompts
- Churn probability scores
- Referral fraud detection
Name one decision owner per application before the pilot starts. Then test each application against a concrete case.
For offer personalisation, Eagle Eye, a UK loyalty platform, can select which voucher a member sees in a retailer app using past basket data.
For next-best-action, Peak, a Manchester AI vendor, can prompt a member to add a forgotten item or use a nearby store.
For churn scoring, Peak or Eagle Eye can flag members whose visit frequency has fallen and trigger a win-back voucher.
For referral fraud, Ravelin, a London fraud platform, and Featurespace, a Cambridge analytics vendor, can score a referral link for fake accounts before the referrer is rewarded.
AI does not replace scheme design. The economics still depend on margin per redemption, so a cost model comes before any pilot. For illustration, a scheme with a 50p margin per redemption and a 20p cost per redemption has 30p left before other costs.
Teams budgeting for this should read the Loyalty marketing: trends and outlook for England in 2027 outlook before committing spend.
Consent, disclosure and the 2025 data rules
The Data (Use and Access) Act 2025 sits alongside UK GDPR. For loyalty teams, the practical questions are whether inferred preferences count as personal data in your context, and whether your privacy notice describes profiling clearly enough.
Consent and disclosure checks
- Inferred preferencespersonal data in your context?
- Privacy notice describes profiling clearly enough
- Retention schedule says how long scores are kept
- Lawful basis for referred friend's data flow
- Human review for automated decisions recorded
If a model infers that a member is likely to churn, that inference is about an identifiable person. Your notice should say so, and your retention schedule should say how long scores are kept.
Worked consent example: A supermarket wants to send personalised vouchers based on past purchases. If it uses purchase history to infer preferences, it can often rely on legitimate interests after a balancing test, a privacy notice update and an opt-out.
If it uses special category data, such as health or religious belief inferred from purchases, or makes solely automated decisions with legal or similarly significant effects, explicit consent is the safer basis. For referral fraud scoring, legitimate interests can cover fraud prevention, but the referred friend still needs a notice.
Referral marketing adds a second layer. A referred friend's data arrives through a member's action, not directly. Check the lawful basis for that flow before you score the referrer.
Where a model decides on its own, with no human in the loop, the bar rises. Keep a person able to review the outcome and record that review. For most loyalty use cases, a human check on high-value offers and referral rejections is enough.
Questions to ask your data protection lead
Questions for data protection lead
- Does the privacy notice name profiling and automated decision-making where they occur?
- How long are model scores retained, and who can access them?
- What is the process if a member objects to profiling?
Questions to ask your platform vendor
Questions for platform vendor
- Where is inference run, and does member data leave the UK?
- Can you export scores, or are they locked in the vendor's dashboard?
- What happens to scores if we leave the contract?
Scoring rubric for an AI loyalty pilot
Use a fixed rubric so pilots are compared on the same terms. The weights below are illustrative, not a standard.
Pilot scoring rubric weights
- Lawful basis and notice25%
- Measurable lift20%
- Data portability15%
- Explainability15%
- Cost per active member15%
- Operational fit10%
| Criterion | What good looks like | Weight |
|---|---|---|
| Lawful basis and notice | Profiling is described in the privacy notice and retention is defined | 25% |
| Measurable lift | A control group shows a defined change in redemption or retention | 20% |
| Data portability | Scores can be exported and reused if you change vendor | 15% |
| Explainability | Staff can explain why a member saw a given offer | 15% |
| Cost per active member | Modelled against margin, not just licence fee | 15% |
| Operational fit | Fits the weekly delivery cycle without new manual steps | 10% |
Score each pilot out of five per row. Anything below three on lawful basis should stop, whatever the lift.
Worked example for a churn scoring pilot at a 40,000 member scheme. Scores out of five: lawful basis 5, measurable lift 3, data portability 4, explainability 3, cost per active member 2, operational fit 4. Weighted score: 1.25 + 0.60 + 0.60 + 0.45 + 0.30 + 0.40 = 3.60 out of 5.
The pilot passes the lawful basis stop rule, but the cost score of 2 is a warning. It should not scale until cost per active member reaches at least 3.
Running these applications changes what the team needs to know. Someone must read model output critically, and someone must own the vendor relationship. The Loyalty marketing skills forecast in England sets out which roles are shifting fastest.
Common questions
Do we need explicit consent to personalise offers?
Not always. It depends on your lawful basis and whether you rely on legitimate interests. Document the assessment and describe the profiling in your privacy notice.
Can AI improve referral marketing performance?
It can help with fraud scoring and referrer ranking. The gain is usually in reducing bad referrals rather than increasing volume.
