What problem does it solve?
Banks and card issuers run many offers (card linked merchant deals, rewards points, fee waivers, product upgrades) and mostly send them as campaigns to broad segments. Customers ignore what is not relevant, points go unredeemed and the rewards budget buys little loyalty. Staff in branches and contact centres have no view of which conversation matters most for the customer in front of them.
The opposite failure is just as real: aggressive targeting that pushes credit at customers who are struggling, or offers that systematically skip some groups. The job is to choose the one relevant thing for this customer now, including "nothing to sell, here is help instead", and to make it explainable.
How does it work?
- Build the candidate list. All offers, rewards and service messages the customer is eligible for, filtered by business rules, consent and suitability.
- Score and choose. Propensity and value models rank the candidates; an arbitration layer picks the next best action for the customer across all channels, so the app, the contact centre and the branch show the same priority.
- Deliver in context. The action appears where the customer is: an in app card, a message after a relevant purchase, or a prompt on a colleague's screen during a call.
- Converse about rewards. An assistant explains the points balance, what a reward is worth, how to redeem and what is needed for the next tier, and completes the redemption through approved APIs.
- Learn. Responses (accepted, ignored, dismissed) feed back into the models, and outcomes are monitored for fairness and customer harm.
- Audience
- Customer facing
- Autonomy
- Autonomous
- Adoption
- Mainstream
- Channels
- Mobile app, Web chat, WhatsApp, Agent desktop
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Interactions handled | Not pooled | at least 3 billion | 1 | 1 organization |
Value drivers: Revenue growth, Customer experience.
Indicative value
A card issuer with 1 million active cardholders
USD 75,000 to USD 1.1 million
Additional margin from personalized offers per year
How this is calculated
Formula: cardholders * reachedShare * extraAcceptance * marginPerAcceptance. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Active cardholders cardholders, customers | 1,000,000 | 1,000,000 | The reference issuer. |
| Share of cardholders who see personalized offers each year reachedShare, fraction of cardholders | 0.3 | 0.5 | Editorial assumption, depends on app usage and consent. |
| Additional offer acceptances per reached cardholder versus generic campaigns extraAcceptance, acceptances per reached cardholder per year | 0.005 | 0.015 | Editorial assumption; no deployment on this page discloses a conversion uplift. Measure it against a control group. |
| Margin per accepted offer marginPerAcceptance, USD per acceptance | 50 | 150 | Editorial assumption, replace with your own offer economics. |
What it leaves out: Incremental offer margin only. It leaves out loyalty and retention effects, rewards costs saved by better targeting, merchant funded revenue, and the cost of the decisioning platform and data work.
Who already uses it?
3 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
DBS Bank
Singapore · Banking · 2026
DBS runs two generative AI virtual assistants on its own AI platforms: DBS digibot for individual customers in Singapore, Hong Kong and Taiwan, and DBS Joy for corporate and SME customers. In July 2026 DBS Joy became agentic in Singapore and now answers questions such as payment status and fees from the customer's own transaction and account data. DBS digibot answers card, refund, fee waiver and remittance questions today; DBS plans to add agentic tasks such as checking card usage, tracking reward points and blocking or replacing cards in the fourth quarter of 2026, for logged in customers only.
- Containment rate: about 90%, DBS digibot, first half of 2026, queries resolved without a follow up call
"In the first half of 2026, DBS digibot successfully resolved nine in every 10 queries digitally, without customers needing to make a follow-up call."
Claimed by: organization - Contact deflection: 7%, DBS Joy in Singapore, first six months of 2026, calls or emails to customer service
"Active users increased by 61%, contributing to a 7% reduction in calls or emails to customer service."
Claimed by: organization - Satisfaction uplift: 17%, DBS Joy in Singapore, first six months of 2026
"Customer satisfaction scores for DBS Joy rose by 17% over the same period."
Claimed by: organization
Bank of America
United States · Banking · 2025
Erica, launched in 2018, is Bank of America's virtual financial assistant in its Mobile Banking app. Beyond answering questions it delivers proactive, personalized insights: BankAmeriDeals cash back deals based on the client's spending, where balances are trending over the next seven days and eligibility for the Preferred Rewards program. It also gives guidance on investment topics for Merrill clients and hands off to people by scheduling appointments. The bank reports that clients have received and interacted with more than 1.7 billion of these insights, and that most users find the information they need, which it links to lower call centre volume. Bank of America says Erica selects answers from a predefined set and does not use generative AI or large language models.
- Users served: about 50 million, since launch in 2018, as of August 2025
"assisting nearly 50 million users since launch, surpassing 3 billion client interactions, and now averaging more than 58 million interactions per month"
Claimed by: organization - Interactions handled: at least 3 billion, client interactions since launch in 2018, as of August 2025
"surpassing 3 billion client interactions"
Claimed by: organization
Commonwealth Bank of Australia
Australia · Banking · 2022
Commonwealth Bank's Customer Engagement Engine (CEE), built on Pega Customer Decision Hub, suggests in real time the next best conversation to have with each customer, whether in the branch, on the phone, online or on a mobile device. Beyond suggesting conversations, the bank uses it to match customers to government benefits and rebates they may be missing (Benefits finder) and to reach customers hit by natural disasters with same day support such as a loan deferral. The same decisions feed digital channels and prompts for branch and contact centre staff.
No outcome disclosed.
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Transaction and product holding data per customer
- Offer catalog with eligibility rules, costs and expiry
- Marketing consent and contact preferences per channel
- Response history with control groups
Systems to integrate
- Decisioning or next best action engine
- Rewards and loyalty platform (balance, redemption)
- Card linked offer provider where used
- App, messaging and agent desktop channels
- Consent management platform
Complexity: Medium
Decisioning engines are mature. The effort is in clean eligibility and consent data, one arbitration across channels, measurement with control groups and the fairness and suitability rules that keep targeting safe.
- 1
Define the action catalog, including service
List every offer and every service action (a fee refund, a hardship check in, a reward reminder) and give each eligibility, suitability and consent rules. Service actions must be able to win against sales.
- 2
Arbitrate in one place
Use one decisioning layer for all channels so the customer does not get three different offers from the app, email and a call. Commonwealth Bank's engine serves branches, the contact centre and digital channels from one decisioning engine.
- 3
Measure with control groups
Hold out a random control group from the start, or no one will be able to say what the engine added.
- 4
Add the rewards conversation
Let customers ask about points, value and redemption in the app assistant, with balances and redemption through the loyalty platform's APIs.
- 5
Review fairness and harm regularly
Check who is shown and who is excluded from offers, and suppress credit offers for customers with signs of financial difficulty.
Guardrails
- Eligibility, suitability and consent checked by rules before any model ranks an offer
- No credit offers to customers with financial difficulty or vulnerability markers
- Offers are clearly labelled as offers and respect opt outs on every channel
- Reward values and redemption terms come from the loyalty platform, never generated text
- Periodic bias review of targeting outcomes across customer groups
KPIs to instrument
- Acceptance rate versus a control group, per offer
- Incremental revenue or margin per treated customer
- Reward redemption rate and points balance age
- Opt outs and complaints about offers
- Offer exposure by customer segment
Human in the loop
Marketing and product owners approve every offer, rule and model change. A conduct review checks targeting outcomes each quarter. Colleagues in branches and contact centres decide whether to raise a suggested conversation at all.
Common failure modes
- Optimising for the wrong customers
- Models learn that struggling customers accept credit offers. Exclude them by rule and monitor outcomes.
- Invisible uplift
- Without a control group, gains cannot be separated from seasonality. Hold out from day one.
- Channel conflict
- Each channel runs its own targeting and customers get contradictory offers. Arbitrate centrally.
- Unexplainable targeting
- A customer or regulator asks why an offer was shown or withheld and nobody can answer. Log the rules and scores behind each decision.
What are the risks and rules?
EU AI Act
Depends on design
Ranking offers is generally minimal risk and the conversational part carries the Article 50 transparency duty. Using AI to evaluate creditworthiness for a credit offer is high risk (Annex III point 5(b)), and Article 5 prohibits techniques that exploit vulnerabilities due to a person's social or economic situation to distort their behaviour in a harmful way.
Rules that apply
Guidance
- Direct marketing and privacy and electronic communications (Information Commissioner's Office, Europe). The ICO's hub for UK direct marketing rules under PECR and data protection law, covering the lawful basis and consent for marketing messages, with its detailed direct marketing guide.
- Article 5: Prohibited AI practices (European Union, Europe). Bans manipulative techniques and the exploitation of vulnerabilities, including those due to a person's economic situation.
Controls to put in place
- Decisioning models in the AI inventory with owners, validation and periodic bias review
- Documented suitability rules for credit and high cost products
- Consent and opt out enforcement across all channels
- Decision logs that explain why each offer was shown or suppressed
- Complaint and outcome monitoring for customers in vulnerable circumstances
Frequently asked questions
- Is next best action the same as a recommendation engine?
- It is broader. A next best action engine chooses among offers, service messages and doing nothing, across channels. Commonwealth Bank said in 2022 that its engine made over 35 million decisions a day and used it to suggest the next best conversation to have with each customer, including same day support for customers affected by natural disasters and matching customers to government benefits through its Benefits finder.
- Can an assistant help customers use their rewards?
- Yes. Bank of America's Erica highlights cash back deals based on the client's spending and notifies clients of their eligibility for its Preferred Rewards program, and DBS plans to add reward point tracking to its digibot assistant in a later phase of its agentic rollout.
- What is the main compliance risk?
- Unfair or harmful targeting, such as pushing credit at customers who are struggling or systematically excluding groups. Rules for suitability and consent, control groups and regular bias reviews are the defence.
How to cite this page
Blits.ai AI Use Case Library, "AI agent for personalized offers and rewards", last verified 27 September 2026, https://www.blits.ai/ai-use-cases/offers-and-rewards-agent. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 27 September 2026: First published