A static model takes how people behave today and projects it forward at a fixed rate. It treats the firm as a spectator. Whatever the firm does next, a price change, a retention offer, a campaign, a strategic move, never enters the math, because the math holds behavior constant. The model can grade the past and score the present. It cannot say what to do next, because in the world it describes, nothing the firm does changes anything.
The flaw is that the model puts the firm's actions outside the system. A customer's value comes out as a number to read, not an input the firm can move. A model with that shape can describe a customer well. It cannot represent the firm reaching in to change the customer's path.
The same flaw runs through churn models that treat attrition as a property of the customer rather than of how the firm treats them, pricing that forecasts demand as if the price had not moved, marketing-mix work calibrated on a market assumed to hold still, and strategy set as a plan against a future that will not respond. In each, the firm's own actions are the input left out.
The fix is a model of a different kind, one that treats the firm's actions as choices to make. That takes three things: a goal, the firm's own actions as the levers, and a loop. You set the goal, solve for the action that moves it most, act, watch what comes back, and solve again. Without a goal there is nothing to optimize, only something to describe. The loop is what turns a number into a decision.
The obvious objection is that firms are not flying blind. They measure their own effect all the time, campaign lift, holdout tests, attribution. That is true, and it is the right place to look. The trouble is the clock. Each measurement is supposed to trigger a fresh solve, the loop running again on what changed. A mix model that re-estimates once a quarter, a campaign whose conversions arrive over thirty days, a clean incrementality test that runs six weeks, each closes that loop long after the moment it could have changed. Every correction is worth less the longer it waits.
A worked example in banking
Customer value shows it plainly. The standard lifetime-value method takes a customer's current margin, holds it flat at an assumed retention rate, and discounts to present value. Retention enters once, as a single constant, the same for a customer leaning out as for one who just opened a second account. We made the case in our work on customer value optimization that a customer's worth moves with behavior, so the firm's next action, which relationship to deepen and which to leave, is what the decision rides on. The method is set out in a 2014 patent.
The dynamic version puts that action into the math, and the score becomes a decision. You set the goal, the worth of the relationship over time, treat the bank's actions as the levers, and solve for the ones that move the goal furthest from where the customer stands today. A customer's value is current value plus potential value, the gain available across the products the bank could deepen. In the worked example, a single retail customer carries $241 in potential value, the difference the bank's own product decisions move. A score sorts customers. An optimization chooses the bank's next move.
Why the static habit holds
The reason the static habit holds, even where the better method is understood, is not a shortage of data. Most large banks already run customer-360 systems and propensity scores. The visibility problem is largely solved, and the dynamic figure is within reach on infrastructure that already exists.
The constraint is organizational. The right objective crosses product lines and spans more than a quarter, while authority and incentives are drawn by product and by quarter. No single P&L owner is measured on the worth of the whole customer over time, so when actions are optimized at all, they are optimized against the goal each owner answers for, a product in a quarter, not the customer over years. The firm optimizes against the wrong objective. Until that number has an owner whose budget and horizon reach across products and across years, the static default holds, whatever the data shows and whatever the model can do.
The shape repeats well outside banking, and so does the market's response. Wherever the better model crosses the lines that authority and incentives are drawn on, the static version wins by default. Many firms now sell a fix for one element of this, faster data, a faster model, sharper attribution, a better optimizer, each real and each partial. What none of them sells is the binding constraint, an owner who runs the whole loop against the right goal, on a clock that lets the answer change something. The fix is not a better model. It is an owner.
Strategy is set the same way, and for the same reason. The plan is authored by one group, on one clock, and graded before the firm's own later moves have made it stale. The static assumption lives in the model, and beneath the model, in who owns the plan and when they answer for it. And the firm is not the only player the plan holds still. Competitors respond to the firm's moves and to each other's, and each responds from where it stands against its own goal. Where an industry is heading follows from those responses, and modeling them is the subject of a coming piece.