Next best action

Next best action extends personalisation beyond the storefront into every touchpoint. Instead of asking "what should we recommend on this page?", it asks "what should we do for this customer now, across all channels?" Candidates might include a reorder reminder on WhatsApp, a discount, an educational email, a support call, or deliberately no contact. Each is scored for expected value given the customer's state, and business constraints such as contact frequency, margin and regulatory suitability filter the result.

The scoring combines predictive models, such as churn prediction and propensity to buy, with uplift estimates that ask whether the action changes behaviour or merely coincides with it. The output feeds a decision service that channels such as CRM, marketing automation and the app consult before acting.

It is not a marketing calendar with segments, and it is not a recommendation widget. The distinguishing feature is that "do nothing" is always a candidate and that the same engine serves marketing, retention and service decisions, so the customer receives one coherent sequence rather than competing campaigns.

• Founder / CEO: it turns customer engagement from batch blasts into individual decisions, which is where retention and lifetime value are won.
• Product manager: the action catalogue and its constraints are a product design task; the model only ranks what you allow.
• Data lead: it needs a unified customer state across channels, which usually exposes identity and event gaps first.
• Compliance officer: in BFSI and insurance, suitability and consent rules must be encoded as hard constraints, not left to the model's judgement.
• CFO: because it optimises expected value rather than volume, it tends to cut discount spend while holding revenue, if the uplift models are honest.

Next best action exists because customers experience one company, not a marketing team, a retention team and a support team each running their own campaigns. Uncoordinated outreach over-contacts good customers, wastes discounts on people who would have bought anyway and misses the moment when a nudge would have mattered. Deciding per customer, per moment, with a single value function fixes that. The trade-off is organisational as much as technical: teams must agree on one objective and hand control of timing to a shared engine, and the models must be validated by controlled tests or the "value" is fiction.

• A D2C brand choosing between a reorder nudge, a bundle offer or silence for each customer on WhatsApp.
• A bank deciding whether to offer a credit-limit increase, a savings product or a service call after a salary credit, within suitability rules.
• A SaaS company selecting the onboarding nudge most likely to move a trial account to activation.
• An insurer timing renewal reminders and cross-sell by predicted lapse risk rather than a fixed calendar.
• An ed-tech platform deciding when a struggling learner should get a mentor call versus an automated hint.

Technique / practiceA decisioning technique built from predictive models, uplift estimates and constraint rules. Eazyware delivers it under personalisation engines, often integrated with a client's CRM and messaging stack.

Eazyware service that covers it: Personalization Engines. Starting prices are on the pricing page.