What a customer is worth, and what they will be worth

Any business that keeps customers over time has to put a value on them, and most settle on one number: what a customer is worth today, multiplied by how long they are likely to stay. Banking is where this gets interesting, because the data to do better is already on hand. Most banks still use the one number. It is simple to compute and simple to act on, which is why it is the standard. It is also a single figure, and a single figure hides the things a bank can do something about, the product to put in front of the customer and the direction the relationship is taking.

A more useful number is built from those parts, and the first cut is across time. A customer is worth what they contribute today, the current value, and what the relationship will be worth over the next several years if it develops, the potential value. Banks have scored propensity to buy for years, and most already hold that number. The harder question is what to do with it, and the decisions that carry weight turn on the second, the customers worth investing in and the ones that are not. Managed on current value alone, a bank optimizes the quarter.

Two numbers, current and potential

Current value is what a customer contributes today. Potential value is what the relationship is worth over time, given where the customer is heading. Most customer management runs on the first number, which is why the customers who look profitable today get the attention and the rest get default treatment.

The customers who matter most are often the ones who do not clear the bar today and show the signs of becoming valuable if they stay. A young account with a thin balance and a rising trajectory can be worth more over five years than an account that reads better this month and is going nowhere. The same logic runs the other way. A customer who looks fine today can have no future worth building toward. A bank managing on current value alone treats these cases the same. The second number tells them apart.

How the number is built

The pieces are not new, and the math is straightforward. For each product, estimate the customer's probability of buying it, the balance they would carry, and the margin that balance earns, then subtract the cost to acquire. That is the product's value to the relationship. The margin comes from the customer's segment, so each product is priced to that customer's economics. Sum across the products for the customer's expected value, then project it forward and discount it to the present, the way any future cash flow is valued.

Two things make this more than a tidy sum. The parts carry the meaning the total throws away. A lifetime value of $241 tells a bank to keep the customer. The same $241 broken into its parts tells the bank what to do with the customer, which product to lead with and which one loses money and should be left alone. A single number ranks the customer. The decomposition reads as instructions, and on some products it runs below zero, which is the bank's signal to stop offering them.

The harder work is keeping the numbers connected and current as they move. A propensity is not fixed. It shifts as a customer opens an account, moves a balance, or responds to an offer, and the expected value shifts with it. A figure calculated last quarter is stale the moment the customer acts. The models that produce these estimates are far better now than when this method was first built, and they can be rebuilt again as methods improve; the framework does not turn on which one is used. It turns on holding the moving parts together in one picture that updates as the customer does.

Per candidate product, probability of buy times projected balance times margin, less acquisition cost, summed across products and priced by the customer's segment. The loop updates the inputs as the customer acts.

What changes when you manage potential value

The hard part of retail banking comes at the start. Almost every new customer is unprofitable at first, because acquisition and onboarding cost more than a new, thin relationship returns. The relationship only pays if the customer stays and buys more over time, a loan, a card, a second account. Many never do. The work is picking the ones that will, early, while they still look like everyone else. Current value sees a new customer as a cost. Potential value asks the question the bank actually cares about, which of these costs becomes a relationship that pays.

With the second number in view, today's action looks different. The action that matters is the one that carries the customer toward higher value, and it is not always an offer. Sometimes it is a product, sometimes a change to the service, sometimes a call to check in, and sometimes nothing at all. When it is a product, the one to put forward is the one that deepens the relationship, even when a different product is more likely to close this month. What the framework picks is the action that bends the path, whether or not anything is being sold.

Retention and concessions are investments too. Waiving an overdraft fee for a customer with low current value and high potential value is money well spent, because the fee given up today is small against the relationship it protects. Run this way, a bank holds a standing answer to who is most valuable today and who will be most valuable later, with the path between the two laid out.

The hard part is the prediction

This is where the framework meets its real test, and it is a fair one. Potential value is a multi-year forecast, and acting on it means giving up real money now, the waived fee or the retention spend, against a model's claim about year three. A CFO is right to ask how often that claim holds and what it costs when it misses. The trap is that the signals of future value are also the signals of a customer who was going to grow on their own, and the two look the same early.

The way through is to name the quantity the decision actually rides on. It is the effect of the action on the customer's value, which is not the same as the value. Value ranks who is worth keeping. The effect, the lift, is what tells you where the spend moves the trajectory. A customer with high value and near-zero lift, the one who would have stayed regardless, is a poor place to spend even though the value number is large. Only a lift view separates that customer from the one whose path bends when the fee is waived.

That number comes from inside the framework. The expected value already updates with what the bank has done. Line a treated customer up against similar customers who got nothing, and the gap is the lift. The same machinery that makes the value dynamic is what tells "grew because of us" from "grew anyway." It has to be run as a comparison, treated against untreated, which is the test every estimate here has to pass. The model is a claim about behavior, and a claim the market contradicts is wrong, whatever its fit on past data. The decision rests on two things at once, what the data projects and what the market confirms, and it is trusted only where the two agree.

The payoff is that the same number that ranks customers also sizes the bet. A cheap concession on a weak signal is a fine bet, because the cost is small against the upside across a large book, so the hit rate barely matters. An expensive one, heavy retention spend, is gated on more, a strong signal and evidence that the action changes the outcome. The value estimate has to carry how solid it is, since a high but shaky number is a different bet from a high and grounded one. The prediction problem stops being the soft spot and becomes the thing the framework is for.

What it's worth at scale

The discipline is worth the effort, and the industry numbers show the size of it. McKinsey's figures put the typical impact of analytics in customer management at a 25% lift in revenue per customer and a 20% reduction in churn, with a 20 basis point gain in deposit margin alongside. The margin figure is the one that scales to the whole industry, because the public data carries both the rate and the base.

The deposit base comes from The U.S. Banking Industry Data Explorer, our free tool that reads the FDIC call reports directly, which puts industry deposits near $19 trillion (FDIC-insured banks and savings institutions, so credit unions are not in the figure). A 20 basis point margin gain on that base is roughly $40 billion a year. That is a ceiling, not a forecast. No one captures a benchmark across every bank, and adoption is uneven, heaviest among the largest institutions and thin among the smallest. It is also gross, not incremental, so only the share a program actually causes counts, treated against untreated. Even a fraction of $40 billion is a large number, which is the point.

Both inputs, industry deposits and margin, are pulled from the Explorer, which refreshes every quarter as new call reports post, so a reader can check the math and rerun it there as the data moves.

What the framework tells you to stop

Selecting which customers to back has a second half: it tells you which ones to stop backing. The same number that flags the new customer worth a bet flags the one whose bet has not paid, the relationship that has had its chance to deepen and has not. A bank that runs the framework only to find winners and fund them leaves that half unused, and it is where the cost discipline lives.

For that customer, the instruction is not to collect less. The margin they bring is real, and the bank should keep it. The instruction is to stop paying to grow a relationship that will not grow. A model that only scores current value cannot give it, because on its books every paying customer looks worth chasing. The second number is what says this one is as deep as it will get, and the money set aside to deepen it belongs with a customer who will return it.

That is the part most programs skip. Lifetime value has ranked customers for years, and every deck says to focus on the best ones. Naming the customers to stop spending on, and acting on the name, is the part almost no one operationalizes.

Why it holds together

This is a way to manage a book of customers over time. The components, the propensity scores, the profitability figures, the discounting, are each familiar on their own. Holding them together so they move with the customer is what turns a set of models into a framework a bank can run on, and it is what the work was built around. That structure was patented and put into production more than a decade ago, on transaction data from thousands of institutions. The models inside it have moved on since; the way they connect has not.

The method is specific to banking. The way of arriving at it is not. The model gives a number, the market test says whether to believe it, and neither is trusted alone. That is how we work on any decision that is expensive to get wrong, with the discipline to act on what the evidence shows, including where the answer is to stop. Customer value optimization is one place that test plays out.

The cost of ignoring the second number rises as acquisition gets more expensive. A bank managing on current value alone wins customers it then fails to recognize, and keeps spending on relationships that will never grow. A bank managing both numbers knows which relationships to build and which to stop funding. If you are working this question inside your own customer base, get in touch.