# Next best action in sales: what a good recommendation looks like (an action plus the reason)

https://salesbond.app/blog/next-best-action-in-sales/
Published: 2026-11-03 · Updated: 2026-11-13

Next best action in sales is the one step a rep should take next on a deal or a customer, shown together with the reason it is next. “Call Ivanov, because the customer has been waiting for a reply for three days” is a next best action. “Call Ivanov” on its own is a task, and a recommendation without a reason is just another task nobody trusts. The reason is what makes it usable: the rep can check it against the CRM, agree or disagree, and act in under a minute.

Most writing under this phrase covers marketing, a different job and a fair one. For a rep, whether inside or out in the field planning a day of visits, the questions are where the action comes from, where a recommendation ends and a task begins, why reps ignore cards without a reason, and how many cards a day is too many. Outside sales sharpens every one of them. A field rep reads the list in the morning and builds the route of visits from it, so each customer has to be one card, and the reason has to be short enough to reread in the car before walking in.

## NBA in marketing vs NBA for a sales rep: not the same job

Most definitions of the term come from customer engagement platforms. Pega describes next best action as a customer engagement approach that combines artificial intelligence with real-time interaction data to create relevant customer experiences ( [Pega: What is Next Best Action](https://www.pega.com/next-best-action)). The CDP.com glossary calls it a real-time decisioning strategy that uses customer data, business rules and AI to pick the most relevant action for each customer at any given moment ( [cdp.com: Next best action](https://cdp.com/glossary/next-best-action/)). Both describe a system that decides, for thousands of customers at once, whether to send an offer, a support message, a piece of content, or nothing.

The decision maker in that picture is the platform. No human looks at each decision.

A sales rep with forty open deals is in a different situation. The decision maker is the rep; the software, if there is any, is an adviser. The customer never sees the recommendation only what the rep does after reading it. And the rep has a whole morning to decide.

So the same words mean two different products. Marketing NBA picks a message per customer at scale. Sales NBA ranks one person’s day and explains the ranking. If you are evaluating a tool, ask which of the two it does. A vendor that answers “both” usually means the first one.

## An action and a reason: anatomy of a useful card

Here is what we think a card should contain, in order, and nothing else.

![Schematic recommendation card with four rows: action, reason, stake and the buttons Accept, Skip, Not now](https://salesbond.app/wp-content/uploads/2026/11/next-best-action-in-sales-2.png)

**The action**, as a verb and an object. Call Ivanov at Northwind. Send the revised quote to Contoso. Not “follow up”, not “review deal”, not “engage account”. If the rep cannot picture themselves doing it, it is not an action.

**The reason**, as a fact from the CRM the rep can verify. Customer asked for a revised quote on Monday; no reply has been logged since. Or: deal has had no next activity for twelve days. One sentence, past tense, with a date or a count in it. The test is simple. Could the rep open the deal and see that the sentence is true?

**The stake**, briefly. Deal is in Proposal, close date in nine days. This is why the card sits near the top rather than the bottom. It doesn’t need a number; it needs to answer “and so what”.

**A way to say no.** Accept, skip, not now. A card you can only accept is a command, and commands from software get resented, then ignored.

That’s the whole card. Confidence meters, close probabilities and email templates make it longer and make the rep less likely to read it.

## Next best action examples: the CRM signals behind each card

A good recommendation starts from something already true in the CRM and translates it into a verb. If you build these by hand, in a saved filter or a spreadsheet, these signals reliably produce an action worth taking:

- No next step: the deal is open and nothing is scheduled. The action is to agree the next step with the customer, and the empty field is the reason.
- The customer is waiting for a reply. Their last message asked for something, and nothing from your side has been logged since. Do the thing they asked for. Our piece on [the sales follow-up email](https://salesbond.app/blog/sales-follow-up-email/) works through this one signal in detail.
- An overdue activity: a call was due, the date passed, no outcome logged. Do it or reschedule it.
- Close date already behind you while the deal is still open. Confirm the timeline with the customer instead of quietly moving the date.
- A late reorder. The customer buys on a cycle, the cycle has run out, and there is no new deal. Contact them before the gap turns into a lost account. This is the one signal that lives in the customer base rather than the pipeline, and most deal-centred setups miss it.
- A new lead nobody has touched. First call; the reason is the lead’s age.

Notice what is not on the list: the rep’s mood, the customer’s sentiment, an opaque score. Each signal is a field, a date or a count, which is exactly why teh reason can be written as a checkable sentence. The [deal health score](https://salesbond.app/blog/deal-health-score/) article shows how signals like these roll up into one number per deal for a manager. Here they stay as single facts, because a rep acts on facts.

## Recommendation vs task: who decides, who writes to the CRM

A recommendation is advice: no owner, no due date, and it never goes overdue. A task is a commitment the manager can see and count. Our position is that the engine proposes, the rep decides, and only the rep’s acceptance writes a task into the CRM, because a task is a promise to a customer and the person who talked to that customer should be the one making it. How this plays out in Einstein, and the weekly check a manager can run on it, is in [Salesforce Next Best Action vs the rep’s own next step](https://salesbond.app/blog/salesforce-next-best-action-vs-rep-next-step-manager-check/).

## Why recommendations without a reason get ignored

Put yourself in the rep’s seat on a Tuesday morning. A panel says “Reach out to Contoso”. You remember Contoso: you sent a proposal last week, they said they would discuss it internally. So why is the system telling you to reach out? Because of the proposal? A date you set? A score you can’t see?

You have two choices. Open the deal and reconstruct the reasoning yourself, which takes longer than the call would. Or dismiss the card and do what you were going to do anyway. After a week every rep on the team has learned the second option, and the panel is furniture.

Now the same card with a reason: “Reach out to Contoso: you sent the proposal on the 24th, they asked for a decision by the 31st, and nothing has been logged since.” Nothing to reconstruct. The rep either agrees (it’s the 2nd, I should call) or disagrees (I spoke to them yesterday and forgot to log it, skip). Both outcomes are good. The first produces a call. The second produces a logged activity and a corrected CRM, so the next card will be right.

The rule we give managers: if a recommendation cannot be written as “do X, because Y happened on date Z”, keep it as an internal signal that affects the order, and don’t show it as a card.

## Limits: a day’s worth of cards, a pause after “no”, one card per customer

Volume first. A rep can act on a certain number of things in a day, and a list longer than that is a backlog with a new name. Wherever the cut is for your team, the feed should stop there and the rep should see the whole list at once. The reasoning behind a daily cut is in [task prioritization for sales reps](https://salesbond.app/blog/task-prioritization-sales-reps/).

Then respect for a no. If a rep skips a card, the same card should not be back the next morning with the same reason. Either the reason changed (waiting five days now instead of three) or the the system waits. A tool that repeats itself trains people to click skip without reading, and then the accept button means nothing either.

And one card per customer. The same account can trip several signals at once: a late reorder, a stalled open deal, a quote they are waiting on. That is one conversation with one person, so show it as one card with the signals listed inside the reason. Three cards for one customer is how you end up with three calls to one customer in a week.

## How Sales GPS builds its cards

We build SalesBond, a sales management layer that works on top of a team’s existing CRM. The part that does this job is called Sales GPS, and it’s fair to say what it does and doesn’t do.

Each card is an action and a reason, in the shape above. The signals come from Deal Health scores and rule violations, from Repeat Sales statuses and buying cycles, and from the context of each deal. A customer who is overdue, leaving or due to buy soon gets one card, with that customer’s deal signals attached to it.

Cards are ordered by importance and urgency, the same two axes as in [the prioritization matrix for deals, leads and customers](https://salesbond.app/blog/prioritization-matrix-for-sales-deals-leads-customers/), and every card explains the reason behind its place in the list.

Sales GPS never creates tasks in the CRM on its own: accepting a card creates one task, and nothing else does. It doesn’t predict churn with machine learning, it never writes to customers, and it personalises nothing for them. It ranks a rep’s day and says why.

If you want to see which deals and customers in your own CRM would raise these signals, the free CRM audit is the quickest way. SalesBond looks at the last 30 days of the connected CRM and sends a PDF report: money at risk, deal health, the pipeline bottleneck, repeat customers, and up to five things to fix, sorted by money. Seven days later a second PDF shows what changed. It works with Pipedrive, HubSpot, Salesforce, Zoho CRM, Microsoft Dynamics 365, Freshsales, Close, monday CRM, Copper, Zendesk Sell, Nutshell, Insightly, Attio, Bitrix24 and Kommo (amoCRM). [Check your CRM with SalesBond](https://salesbond.app/).

## FAQ

**What is a next best action model?**
 In marketing, it is the logic that scores the eligible actions for each customer and returns the highest-value one; the CDP.com glossary notes that rules-only NBA is possible, while predictive models push the ranking further ( [cdp.com](https://cdp.com/glossary/next-best-action/)). In a sales team the model can be much simpler: a fixed set of CRM signals, a fixed way of ordering them and a written reason on each card. It doesn’t need to predict anything. It needs to be checkable by the rep.

**What is next best action in Salesforce?**
 Einstein Next Best Action shows recommendations in a panel on the record, picked by a strategy an admin builds. What that means for the rep’s actual tasks is covered in [our Salesforce article](https://salesbond.app/blog/salesforce-next-best-action-vs-rep-next-step-manager-check/).

**How is next best action different from lead scoring?**
 Lead scoring ranks records: this lead is hotter than that one. Next best action ranks steps: do this on this deal, for this reason. A score tells you where to look; an action tells you what to do once you are there. For how a score is built from deal criteria, see [how Pipedrive scores deal priority](https://salesbond.app/blog/pipedrive-scores-deal-priority-criteria/).

If you would rather see the cards than build the filters, [start with SalesBond](https://app.salesbond.app/auth/signup).
