Automation and AI

Next Best Action

Also called NBA, next best offer

A recommendation, made per customer, about which single step is most likely to move them forward now.

Quick facts: Next Best Action

Category
Automation and AI
Also called
NBA, next best offer
Level
Advanced
Affects
Message selection, sales scripts, campaign sequencing
Where to see it
CRM and marketing automation decision rules, customer data platform journey builders, custom models
In this article4
  1. How next best action works
  2. Why next best action matters
  3. Where next best action goes wrong
  4. How to act on it

How next best action works

You define a list of things you could do for a customer: send a guide, invite them to a call, offer a trial, ask for a review, or wait. For each customer the system estimates the effect of each option and recommends the one with the best expected result. Underneath, that is usually several propensity models compared against one another, plus business rules that remove some options entirely.

Constraints do as much work as the model. Eligibility rules stop the system offering a product the customer already owns, frequency rules stop it contacting someone too often, and margin rules stop it recommending a discount every time because discounts convert. Without those, the recommendation collapses into whichever action is easiest to say yes to, which is rarely the one worth taking.

Why next best action matters

Most marketing programmes are organised around campaigns, so everybody in a segment gets the same thing this week. Next best action inverts that and organises around the person, which is much closer to how a good salesperson works and to how customers experience being sold to.

The practical benefit is fewer wasted contacts. Instead of one monthly email to the whole list, the people ready for a conversation get a call, the ones still reading get something to read, and the ones who need to be left alone are left alone. That protects both your sending reputation and their patience.

Where next best action goes wrong

Short-term optimisation is the trap. A system rewarded for immediate response learns to send the most aggressive offer to everybody, because that maximises today’s number while eroding margin and goodwill. Recommendations should be judged against a longer outcome, such as retained revenue, rather than the next click.

The second failure is an action list nobody can service. If the recommendation is to call within the day and there is no one free to call, the system produces a queue of ignored advice and the team stops opening it.

The third is opacity. When a salesperson cannot see why a customer was surfaced, they treat the list as noise. Attaching a short reason to each recommendation changes adoption more than any improvement in accuracy does.

How to act on it

Start manually. Write down the decision your best salesperson would make in each common situation, encode those as plain rules, and run them for a while. That does much of what a model would do, exposes the data you are missing, and forces the action list to be realistic before anything is automated.

Then measure against the honest comparison, which is what would have happened under your old campaign schedule. Keep a group on the old approach so you can tell the difference. Decide early where the human sits, whether approving, overriding or simply informed, because these systems fail on adoption far more often than on accuracy. Fitting the logic into an existing marketing automation setup is usually easier than building a separate decision layer beside it.

Do and do not

Do

  • Give the system a real menu of actions
  • Include doing nothing as a valid recommendation
  • Measure whether the recommendation beat the default

Do not

  • Recommend an action your team cannot deliver
  • Optimise the next click over the relationship
  • Hide the reasoning from people acting on it

Questions people ask about this

Is next best action the same as personalisation?

They overlap but answer different questions. Personalisation decides how a message looks for a given person, including the name, the product shown and the wording. Next best action decides what should happen at all, including whether to make contact. You can personalise a badly chosen message, which is why the decision usually matters more than the presentation.

Do I need a model, or will rules do?

Rules do more than people expect, especially when the list of possible actions is short and your team already knows the patterns. Models earn their place when there are many actions, many signals, and enough history to learn from. Start with rules, note where they feel wrong in practice, and let those gaps decide whether modelling is worth the effort.

How do I stop it recommending discounts constantly?

Give the system the cost of each action as well as its success rate, so a discount is judged on margin rather than on conversion alone. Add hard rules limiting how often discounting can be recommended to the same customer, and judge the whole programme on retained revenue over a period rather than on immediate response.

Related terms

Found this useful?

Share it, or ask an AI to summarise it

Back to the glossary

Knowing the term is the easy part

Applying it to your own site and budget is the work. Book a call and I will tell you what actually applies to you.