Next best action is an approach in which a system chooses, for each customer at a given moment, the one thing most worth doing next. That might be an offer, a helpful message, a service call, a reminder or, often, nothing at all. The choice balances what the customer is likely to want with what the business is trying to achieve.
How next best action works
A next best action system has four ingredients:
- Customer data: what you know about the person, such as their lifecycle stage, purchases, recent behaviour, open complaints and marketing preferences.
- A menu of actions: every offer, message or intervention the business could make, including “do nothing”.
- Rules: eligibility and limits, such as who qualifies for an offer, how often someone can be contacted and what consent they have given.
- A ranking: a way to choose between eligible actions, usually the likelihood the customer will respond multiplied by the value of that response, adjusted for cost and business priorities.
Likelihoods often come from a propensity model, such as the chance someone buys a product, upgrades or leaves. The chosen action is then delivered wherever the customer next appears: an email, a website banner, an app screen or a prompt on a call handler’s screen. Next best offer is a narrower version that chooses only between offers.
Two examples show the range. A broadband provider sees that a customer’s contract ends next month and that churn prediction rates them as likely to leave. The best action is a retention offer, not an upgrade pitch, and if they have an open complaint, the best action is to resolve it and send no marketing at all. A small online shop can do a simple version with rules: someone who bought a coffee machine a month ago is shown descaling tablets; someone who bought tablets last week is shown nothing.
Why it matters
Without it, each channel and team pushes its own message. The same customer gets an upsell email, a discount banner and a survey request in one day, some of which contradict each other. Next best action puts the customer at the centre and lets the business speak with one voice. Done well, it is a disciplined form of personalisation, which also means fewer, more relevant contacts.
In the UK there are regulatory angles too. Choosing actions based on personal data is profiling under UK GDPR, so it belongs in your privacy notice, and people can object to profiling for direct marketing. Financial services firms must also meet the FCA’s Consumer Duty, which expects good outcomes for customers, so a system that pushes products ahead of customer needs is a risk there as well as a poor experience.
Common mistakes
- Treating it as a sales engine only, with no service or “do nothing” actions on the menu.
- No frequency limits, so customers are contacted every time they appear.
- Ranking only on short-term conversion, so discounts win every time and margin suffers.
- A model nobody can explain, which makes errors hard to spot and decisions hard to defend.
- Building it before the customer data is clean and joined up.
How to act on it
- Start with rules, not machine learning: five to ten actions, clear eligibility and a contact limit.
- Include service actions and a “do nothing” option.
- Suppress marketing for anyone with an open complaint or no marketing consent.
- Hold back a small control group so you can measure incrementality rather than assuming the system works.
- Add propensity scores once the rules are proven and the data is reliable.
Working out which customer actions matter and how to coordinate channels is part of my digital marketing strategy and consulting work.