Book a demo

Insights

How AI selects the next best action for each customer

Eligibility first, then a ranked decision, then measurement. A walk through the mechanism, without the hand-waving.

By The GoEngage AI team7 min read

Key takeaways

  • Eligibility is a hard filter applied before anything is ranked — entitlement, consent, region, risk, budget.
  • Remaining actions are scored by predicted contribution to the objective for this person, right now.
  • "Do nothing" is always an eligible action; silence often wins.
  • A holdout that receives no action is what lets the platform attribute real lift and correct itself.

Under the marketing language, next best action is a ranking problem with a safety layer. The system has a set of actions it could take for a customer; it needs to pick one — or none — that best serves the goal, without doing anything it is not allowed to do.

The four stages

  • 1. Eligibility filter

    Hard rules remove actions that must not happen: no consent for this channel, product not available in the region, customer over a risk threshold, discount budget exhausted. This is deterministic and auditable.

  • 2. Ranking

    Each surviving action gets a score for how much it is expected to move this person toward the objective — informed by their history, similar customers, recency and fatigue. "Do nothing" is scored alongside the rest.

  • 3. Act

    The top-ranked action is delivered on the channel the customer responds to, at a moment that fits their pattern, within frequency caps. If the chosen channel is unavailable, the next viable one is used.

  • 4. Learn

    A randomised holdout receives no action. The difference in outcome between treated and control is the lift attributable to the decision, and it feeds the next ranking.

A worked example: a customer 21 days after their last order

Candidate actionEligible?Why the agent might rank it low or high
10% off, via emailYesLow: this customer buys full-price and opens email rarely; a discount trains the wrong behaviour
Back-in-stock alert for a watched item, via pushYesHigh: strong intent signal, responsive on push, no margin cost
Loyalty tier nudge, in-appYesMedium: relevant but not time-sensitive
No actionAlwaysHigh if nothing clears the bar — avoids fatigue and protects the channel
SMS promoNoFiltered: no SMS consent on file

Why "do nothing" matters

The most common mistake in next-best-action programmes is treating every decision cycle as a send opportunity. An agent whose objective is net incremental outcome will often choose silence, because a marginal message costs attention, opt-in and deliverability. Making "no action" a first-class, ranked option is what keeps the programme from degrading into over-contact.

References

Keep reading

See it decide on your data.

A short working session on your customers, your channels and one outcome you want to move. No slideware.

Book a demohello@goengage.ai
How AI Selects the Next Best Action for Each Customer | GoEngage AI