Why Banks Need to Start Personalization Before the Offer
By David Longobardi
Simple Subscribe
Subscribe Now!
Under competitive pressure from fintechs and neobanks, traditional financial institutions are struggling to execute personalization. While banks and credit unions have spent years automating processes, turning those efficiencies into targeted, growth-driving offers remains difficult.
Zafin’s 2026 State of Offers research found that 48% of traditional-FI offers use “sophisticated mechanics” — such as partnership bundles, behavioral incentives, or dynamically tailored rewards — compared with 65% of fintech offers. The gap is even sharper among community institutions, where only 27% rely on sophisticated mechanics. Notably, two-thirds of traditional-FI offers are not highly targeted at all, requiring no more than a simple sign-up or infrequent engagement by the accountholder.
The imbalance is primarily a function of institutional constraints, including legacy technology and regulatory caution. At many institutions, changing a product, rate or offer still requires updating a decades-old core system. According to Carson Kotnyek, Zafin’s head of industry advisory, the time to market for a campaign can drag on for three to nine months.
The question today is whether agentic AI can finally enable banks and credit unions to break through. Implemented correctly, Kotnyek said, AI agents can bypass infrastructure bottlenecks — enabling more sophisticated targeting and reducing execution costs without requiring institutions to rip and replace their existing core technologies.
Key insight: The opportunity is to use AI to turn better intelligence into better banking decisions, and ultimately into products, rates, and offers that reach the market much faster. But to do so, financial institutions must rethink some of their assumptions, both about AI and about personalization.
1. Start With the Opportunity, Not the Offer
When most financial institutions think about personalization, they tend to start with the product or offer: identify something to promote and then determine which customers are most likely to respond. Agentic AI opens the door to reversing that sequence. The starting point can instead be the customer behavior that signals an unmet need or an opportunity to deepen the relationship.
Kotnyek offers the example of accountholder money movement. AI can identify customers who routinely transfer funds to investment platforms such as Robinhood or Vanguard, including amount, destination, and frequency of transfers. The same analysis can reveal customers making mortgage or credit-card payments to competing institutions. These signals have long existed in bank data, but identifying and acting on them systematically has been difficult.
Key insight: “What AI gives us,” Kotnyek says, “is the ability to surface those opportunities without knowing what to look for in advance.” Rather than analysts running queries against predefined hypotheses, AI can scan customer activity and identify patterns that merit attention.
2. Personalization Does Not Equal ‘One Offer Per Person’
Many institutions pursuing personalization are held back by the assumption that the goal is to create a unique offer for every customer — in effect, letting perfect be the enemy of good. Kotnyek argues for a less-granular approach in which AI defines opportunities by mapping customer behaviors to various needs. Then, accountholders can be presented with a combination of opportunities that match their behaviors.
Consider 20,000 customers who regularly transfer money to an investment platform. Creating 20,000 different competitive wealth-management offers would add little value. But each of those customers may also fall into other opportunity buckets — perhaps consolidating funds in advance of a home purchase, or making progressively high credit card payments. Each of these accountholders will get the same wealth-management offer, but within a sequence of other offers that reflect the accountholder’s needs.
Key insight: Kotnyek describes this as an opportunity matrix. Each individual component can be reused across a broad cohort, but the offer mix presented to the customer is specific to an individual relationship. AI helps determine which interventions are relevant to each customer and assembles them into a more individualized experience.
3. Personalization Isn’t Your Strategy; Your Strategy Is Your Strategy
One of AI’s strengths is its ability to generate more opportunities, and at a faster rate, than human analysts can. But surfacing more opportunities creates a second challenge: deciding which ones are worth putting forward to customers in light of an institution’s strategic objectives.
“Determining which set of opportunities is best suited to achieving a bank or credit union’s larger goals can be a heavy lift,” Kotnyek said. “AI can connect the dots far more efficiently.” An agent can ingest the opportunities AI has identified and optimize them against all of an institution’s goals, whether customer acquisition, deposit growth, or credit card issuance.
And in fact, strategic intent is a key driver of offer design. Zafin’s research sorted offers into five intent categories — balance growth, acquisition, retention, behavioral activation, and relationship deepening — and then compared the techniques they applied. Here are some of the report’s findings:
- 53% of relationship-deepening offers incorporated sophisticated targeting, such as segment-based reward tiers, life event-triggered offers, or real time personalization driven by individual transaction behavior.
- 54% of acquisition offers required a qualifying action such as direct deposit or salary transfer.
- 78% of behavioral activation offers applied emerging or advanced engagement mechanics, such as interactivity or gamification.
Key insight: Rather than treating every customer signal as a call to action, institutions must make strategy the filter through which AI interprets the field of opportunities — deciding which are worth pursuing and what kinds of offers to build around them.
Leaner Operating Model
According to Kotnyek, AI agents take on work that previously required employees to move manually through multiple platforms — drafting the offer, preparing it for approval and carrying out the steps needed to put it into production. Agentic AI enables financial institutions to “short-circuit a bunch of things that theoretically they could have done before, but practically they didn’t,” he said.
Using natural language, a bank marketer might instruct an agent that its objective is to acquire mass-affluent customers, while inputting other parameters such as budget. The agent would then interrogate the institution’s data, identifying the subset of opportunities that meet that goal and proposing the most effective offer design.
The key is that agents decouple execution from the underlying core system, enabled by an integration and orchestration layer that translates data from the core into executable formats and then translating the result back again. The same layer can coordinate what happens after each campaign is launched, tracking offers in market, qualifying signups, and triggering followups.
Bottom line: As financial institutions advance from LLMs to agents, Kotnyek expects financial institutions to next look toward a “factory” model of AI deployment.
Rather than buying a series of purpose-fit AI applications, institutions would instead build an environment in which they create, deploy, and govern their own agents. This environment — the factory — provides access to multiple underlying models, which can then be called up as needed. So, simpler, lower-cost models would handle routine work and more capable models would be reserved for higher-stakes decision-making use cases. The factory would also comprehensively provide the guardrails and governance needed to control how those agents operate.
Such a model has potential beyond offer management. Kotnyek points to agents that can help rationalize sprawling product catalogs, create products once a new structure has been approved, and manage rate changes as market conditions shift.
In each case, the underlying principle is the same: give AI a business objective, put the right controls around it, and allow agents to carry more of the work from decision through execution.
