Sometimes the Best Intervention Is No Intervention

Every pop-up, coupon, and reminder has a cost. A decision engine earns its place when it can say: leave this visitor alone.

Published 2026-09-28 ยท 5 min read

Sometimes the Best Intervention Is No Intervention

Tools Built To Act

Most e-commerce tooling has a bias. It is bought to do things: show the pop-up, send the email, apply the coupon. Its reports count what it did and the revenue that followed. A tool that did nothing has nothing to show.

So the default becomes action. A trigger fires, an intervention goes out, and whoever buys afterwards is counted as a result.

What this misses is that doing nothing is also a decision, and for a large share of visits it is the correct one.

Three Visits That Should Be Left Alone

The buyer who is already buying. She added the item, went to checkout, and is filling in the address. Any interruption here, even a generous one, adds a step to a purchase that was already happening. A discount at this point is pure lost margin.

The researcher. He is comparing models for a purchase he will make next month. He reads specifications and leaves. Pressure today will not move the date. It will only teach him that this store is the one that insists.

The returning customer in a routine. She buys the same category every few weeks and follows the same path each time. She does not need help, and an offer trains her to wait for the next one.

In all three cases the visit looks like an opportunity to a trigger-based system. In all three, acting makes the outcome worse or more expensive.

The Costs That Do Not Appear In The Report

Over-intervention has costs that are easy to miss because no dashboard line shows them:

  • Margin given away. Discounts paid to people who would have bought at full price.
  • Trained behaviour. Customers who learn that leaving a cart produces a coupon will leave carts on purpose.
  • Attention spent. Each irrelevant message makes the next relevant one less likely to be read.
  • Brand perception. A store that interrupts constantly feels insecure about its own prices.

    • These costs accumulate slowly. By the time they are visible in margin or repeat-purchase data, the habit that caused them is months old.

    Deciding Whether Before Deciding What

    This is why we describe the work as three questions in order: whether to intervene, how, and when. The first question is the one most systems skip.

    To answer it, an engine needs to estimate two things for each visit: how likely the person is to buy without help, and how likely an intervention is to change that. When the first is high or the second is low, the best score goes to no action.

    Fluence scores each possible intervention for a visit and picks the best one. Sometimes the best one is none. That decision is recorded like any other, with its reason, so it can be audited later.

    How To Know If Restraint Is Working

    Restraint is only credible if it is measured. The check is a comparison: a group that receives the context-based decisions, including the decisions not to act, against a group that receives the old blanket flow.

    Look at conversion, and then at what it cost: discount spend per order, margin per order, and repeat purchase in the following weeks. If conversion holds while discount spend falls, the engine is doing its job in the visits where it stayed quiet.

    A good intervention system should be judged partly by what it chose not to send.

    Read how we measure the impact of a decision.