Why bank loyalty programs need an AI feedback loop

Shawn Conahan is chief revenue officer of Wildfire Systems, a white-label cashback and shopping rewards loyalty platform.

BCG's Loyalty 2.0 framework lays out a compelling economic case: the earn-and-burn loyalty program model many banks still run is broken. Points tied to card spend are expensive to fund, increasingly commoditized, and largely invisible to the customers who nominally participate in these programs. 

BCG’s research makes clear that programs must be self-sustaining, ideally with affiliate commissions earned from merchants helping to fund the rewards paid to customers, creating a program that pays for itself.

That logic is hard to argue with. But putting the pieces in place is a whole other story.

The missing piece is often relevance. Banks may have the offers and merchant relationships needed to create a self-funding loyalty program, but the economics only work when customers consistently engage. AI can help close that loop by using customer behavior to deliver more relevant offers, learn from the response and improve the next interaction.

Many financial institutions have internalized the Loyalty 2.0 mentality and invested in shopping portals, browser-based offer tools, and card-linked offers. These are the right building blocks. But the self-funding flywheel BCG describes only gains momentum when banks deliver personalized offers that connect with customers. Greater engagement drives merchant referrals, which fund richer rewards and encourage further engagement. 

Relevance determines engagement

BCG’s report points to data indicating a strong consumer appetite for personalized savings and offers: 75% of consumers are more interested in earning cash back because of inflation pressure and 73% expect companies to understand their unique needs and expectations. These findings suggest that customers are primed for offers. But wanting them, and responding to the ones a bank actually serves, are two different things.

The problem is that most bank offers are still built around “rules” like card spend tiers, broad demographic segments and seasonal promotions. Under this legacy model, Customer A, actively comparing flights and hotels for a family vacation, and Customer B, who just bought a high-end stroller, see the same reward offer, because they're technically in the same card spend tier. The bank knows both use their cards extensively. But it doesn't know what either of them actually needs next.  

Consumer expectations for personalization are extremely high due to years of conditioning by retail and tech brands. Netflix and Ulta are two good examples of this. Netflix personalizes home screen recommendations for each of its members. More than 95% of Ulta Beauty’s sales come from members of its loyalty program, giving the retailer a substantial pool of customer data it can use to personalize experiences. These experiences have set a bar that “spray and pray” offers, which get sent to broad groups of banking customers, simply can’t hit. When offers miss the mark, the downstream effects compound. Customers disengage. Merchant referral volume declines. Commission revenue slows. The program gets more expensive for the bank to sustain on interchange alone. The flywheel stalls out.

The feedback loop that changes the economics

What AI-driven personalization introduces isn't just better offer targeting. It's a continuous feedback mechanism that turns each offer interaction into a data signal. These signals, taken together, make the next offer incrementally more relevant, and the one after that even better still.

In the example of Customers A and B above, with that continuous feedback loop in place, the travel shopper gets a discount offer from one of the hotel brands she's been browsing. The new parent gets a cash back offer at buybuyBaby. Same bank, same program, but a completely different experience. The AI is working from a constantly updating picture of each customer and that customer’s intent, instead of a static segment assigned months ago.  

The key input here isn’t just historical transaction data. That tells banks what already happened. Behavioral insights inferred from shopping actions, such as browsing product pages, activating coupons or offers and making purchases, are far superior at understanding an individual’s preferences. 

With this in mind, banks that embed shopping rewards into the shopping experience, through browser extensions, in-app offer surfaces, or other commerce integrations, gain access to something card transaction data alone can't provide actual shopping intent. 

When a customer is actively comparing refrigerators or researching mortgage rates, that behavioral signal is a far stronger predictor of purchase likelihood than fitting someone into a broad demographic segment or trying to infer intent from their historical spend pattern. An offer delivered in response to demonstrated intent is much more likely to convert. In fact, McKinsey reported that personalization can drive a 5-15% revenue lift. 

AI closes the loop by making those shopping behavior signals actionable in near-real-time. Rather than refreshing offers according to a bank’s rules or promotional calendar, a personalization engine is continuously learning which offer types resonate with which customers, adjusting delivery timing, refining category matching, and deprioritizing offers that show declining engagement before they erode program performance.

Building the infrastructure that enables the flywheel

None of this happens through a portal refresh or a new points calculator. Several capabilities need to be in place to make this work. Banks first need a data layer that can integrate shopping signals like browsing and buying into the offer decisioning engine. This means moving beyond the traditional card-linked offer model, where the trigger is a completed purchase, toward systems that can act on pre-purchase intent.

Using browsing and shopping-intent data also creates privacy responsibilities. Banks must be transparent about which information they collect, obtain appropriate customer consent and give customers meaningful control over how that information is used. Personalization that customers perceive as intrusive can undermine the trust and engagement the program is intended to build.

Then, the personalization model needs to “see” across the full product relationship. A customer with three products at an institution often gets no more recognition than someone with one. Unifying the data foundation to reflect the total relationship is what allows AI to surface offers that feel genuinely relevant, because they're informed by the full picture of who that customer is.

The compounding case for getting this right

BCG cites the research showing that U.S. consumers are enrolled in roughly 17 loyalty programs on average but actively engaged in only seven. The gap between enrollment and engagement is where most bank loyalty programs currently fall short. Getting a customer to activate is table stakes; building a program they return to because the offers feel like they were made for them is where the flywheel actually starts to turn.

AI alone will not repair the economics of bank loyalty. But when it connects customer intent, relevant merchant offers and continuously improving personalization, it can provide the feedback loop that turns a collection of rewards into a program customers regularly use.