EASE 9.0 · Excellence · Advances & Deposits GrowthThe Same AI That Wins a Deposit Can Walk It Out the Door
Scaling origination and cross-sell without a proportional rise in headcount.
And without relaxing credit quality to do it.
Growing a loan book and a deposit base at the same time, without a proportional increase in underwriting and relationship-management headcount, sounds like a contradiction. Agent-based decisioning breaks that link.
Why Growth Used to Require Proportional Headcount
In a manual underwriting model, doubling loan volume does require close to double the underwriting staff, because each application consumes a fixed amount of a credit officer's time regardless of how routine it is. Deposit growth has a quieter version of the same problem. A surplus balance sitting in a low-yield savings account, or a fixed deposit approaching maturity, is a cross-sell moment, but only if someone notices it before the customer's relationship manager gets to it during a routine review, which for most customers is rarely.
The Path to Progress
The marginal cost of the ten-thousandth application in a month is close to zero once the decisioning infrastructure exists.
The Business Case
The Customer Experience Transformation page details a PSU bank's full reported numbers for this exact mechanism — a 3× cross-sell lift at the salary-credit moment, alongside higher digital engagement and lower dormancy. The growth argument that matters specifically for this category is what that mechanism implies for deposits, which the credit-side numbers alone don't capture: the same real-time trigger that fires a cross-sell offer at salary credit can just as easily fire a deposit-retention offer when a fixed deposit approaches maturity, or a savings-to-FD sweep suggestion when a surplus balance sits idle. These are moments a relationship manager covering hundreds of accounts structurally cannot monitor for every customer, every day.
Underlying figures: Appice client engagement data, EASE 9.0 & AI client briefing, 2026; see Customer Experience Transformation for the full reported results. Indicative of results reported across these campaign-management engagements, not independently audited; bank name withheld at client request.
That mechanism has an independent precedent at scale, outside any Appice engagement. Commonwealth Bank of Australia's Customer Engagement Engine, in production for seven years and built on real-time decisioning technology from Pega, is regarded as one of the most mature implementations of its kind anywhere. The bank's data and analytics chief has reported the engine generating ten times more home-loan leads than its previous approach, at a lead quality some 300% better, by the same real-time-trigger logic this page describes rather than a periodic campaign. The engine now makes 53 million calls a day on inbound traffic alone, drawing on hundreds of millions of data points across 1,000 machine-learning models, and its Benefits Finder tool has proactively surfaced more than a billion dollars in government payments customers were entitled to but had not claimed, an average of $710 per customer who used it. None of that runs on Appice's platform; it is independent confirmation, at real scale, that real-time trigger-based decisioning produces exactly this kind of growth and engagement lift, not a result specific to one vendor's own client data.
Source: Mi3 Australia, “Commbank's Customer Engagement Engine Helping Redefine Boundaries of Banking Services” (2023), reporting Commonwealth Bank of Australia executive statements. Figures as publicly reported, not independently audited.
What's Actually Deployed Elsewhere, and the Risk No One Is Pricing In
The number above shows what a deposit-side trigger already does; global deployments show how far the category still has to run, and a risk worth taking seriously regardless of geography. Every deployed system elsewhere doing this today is a recommendation engine, not yet a fully autonomous one; the distinction is worth stating plainly, since overclaiming it is easy to catch. Personetics, used by Truist, BMO and other global banks, has delivered over a billion personalized savings and spending insights and reports real relationship-balance lift among engaged customers. Kasisto's conversational assistant drove a 27% increase in new certificates of deposit at First Financial Bank. Both surface a nudge for the customer or a banker to act on; neither yet autonomously decides an offer, prices it against live competitor rates, and executes the retention call end to end. FIS has named a fully autonomous deposit-retention agent as a 2026 roadmap item in its partnership with Anthropic, a useful marker of where the frontier actually sits — proven recommendation layers today, fully agentic retention systems still arriving.
The more urgent case for building toward one is defensive, not incremental. A Dallas Federal Reserve research paper published in August 2026 warns that agentic AI configured on the customer's side to automatically chase the best available deposit rate could erode the customer inertia that has always been a bank's cheapest, stickiest source of funding, modelling a reduction in aggregate bank lending capacity of up to $700 billion if this scales. That reframes the question this category poses: the strategic case for a deposit-retention agent is not primarily growth — it is whether the bank keeps any pricing power at all once its own customers' AI starts shopping their deposits automatically, a risk that does not stop at any one country's border.
Sources: Personetics and Kasisto published case studies; FIS/Anthropic 2026 roadmap announcement; Dallas Federal Reserve research on agentic AI and deposit stickiness (Aug 2026), as reported. Figures as publicly reported by each company, not independently audited.
Where AI Agents Fit
This is where the category is headed, not a claim about the deployment cited above: today's live results come from Appice's real-time campaign layer, not from an autonomous Decisioning Agent holding credit or offer authority. Extended into full agentic decisioning, two named agents would divide the work, and their sequence would matter: a Decisioning Agent scoring the moment itself, whether that is a salary credit landing, a surplus balance appearing, or a term deposit approaching maturity, and deciding whether a credit or deposit offer fires, against Appice's under-200-millisecond platform target. An Optimize Agent would then tune which offer and channel converts best for that customer over time, without a human resetting targeting parameters. The customer would experience only the offer; the scoring and continuous tuning behind it is what would make that offer arrive at the moment attention is highest, instead of in a generic monthly campaign that treats a salaried professional and a retiree with idle savings identically.
The Skeptic's Answer
This page's own strongest piece of evidence cuts against its own recommendation. A Federal Reserve paper warns that agentic AI shopping for deposit rates on the customer's behalf could erode bank funding stability by up to $700 billion, and the advice that follows is still to build more agentic AI on the bank's side. That looks like arming the same technology the page just called a threat.
It is the same asymmetry every bank already lives with in fraud and cybersecurity: the threat does not wait for the defence to be ready. If customer-side AI is going to shop deposit rates automatically regardless of what any single bank does, the bank without a real-time retention agent watching for exactly that signal is not avoiding the arms race — it is simply the first one to lose share when it starts. Building a deposit-retention agent is not participating in the erosion the Dallas Fed describes; it is the minimum required to notice it happening to this bank's own book before the balance is already gone.
One Decisioning Layer, Both Sides of the Balance Sheet
Most banks still run loan growth and deposit mobilization as separate initiatives, reporting up through separate departments with separate targets. The balance sheet does not see that separation. A customer's idle surplus and their next credit need are two views of the same relationship, often a few weeks apart on the same account. On Appice's platform, whichever side moves first, the same Decisioning and Optimize Agents simply redirect their attention to it. A bank still organized by product line is scoring half of every customer relationship at a time, and calling it the whole picture.