How do you get Salesforce to automatically tell sales reps which leads are most likely to buy, what products to recommend, and what action to take next — all without writing any code? Einstein AI Tools — specifically Einstein Prediction Builder, Recommendation Builder, and Next Best Action — are the no-code features that turn your business data into intelligent predictions and actionable suggestions, helping customer-facing teams work smarter and faster.
Jump to a section
A simple way to picture Einstein AI Tools: Prediction, Recommendations, and Next Best Action
When you open your favourite food delivery app in the evening, it first predicts what you will order next based on your history — maybe a Friday night pizza because you always order pizza on Fridays. This prediction then leads the app to recommend three specific pizzas from local restaurants you have not tried yet, tailored to your taste. Finally, the app shows you a "Next Best Action" button: "Add a garlic bread for just £1 more" — a suggestion designed to improve your meal right now.
In a Salesforce business environment, a sales manager uses Einstein Prediction Builder to forecast which deals will close this month, just like the app predicts your pizza order. The manager can then use Recommendation Builder to surface products the customer is most likely to buy next, similar to the app recommending new pizza flavours. Finally, Next Best Action tells the sales rep exactly what to do next — like sending a follow-up email or offering a discount — to increase the chance of closing that deal. This three-tool system turns raw sales data into a personalised, step-by-step guide for every customer interaction, making the sales process as intuitive as ordering your favourite meal.
Salesforce Einstein is a set of artificial intelligence (AI) features built directly into the Salesforce platform. AI means a computer system that can learn from data to make decisions or predictions, rather than being explicitly programmed with every rule. For the AI Associate exam, you need to understand three specific Einstein tools: Prediction Builder, Recommendation Builder, and Next Best Action. Each tool solves a different business problem, but they work together to create a smarter, more proactive customer experience.
Let us start with Prediction Builder. This tool allows you to create a custom AI model that predicts a specific outcome relevant to your business, with no coding required. A model is a mathematical representation of patterns found in your data. For example, imagine you are a sales manager at a company selling software. You have historical data about past deals: which ones closed successfully and which ones were lost. You also have data about each opportunity — things like deal size, product type, the sales rep involved, and the industry of the customer. Prediction Builder uses this historical data to learn which factors most strongly correlate with a successful close. Once trained, the model can analyse a current open opportunity and give it a prediction score, like "85% likely to close". This prediction helps your sales team prioritise their time on the deals with the highest chance of success.
Why does this matter? Before AI, sales teams would rely on gut feeling, manual spreadsheets, or simple rules like "focus on deals over £10,000". Prediction Builder automates this analysis, using all available data to make a more accurate forecast. It is like having a data scientist on your team, but without the cost or complexity. The exam tests that you know Prediction Builder is for creating custom prediction models — it is not limited to pre-built predictions. You can choose the object (like Opportunity or Lead) and the field you want to predict (like "Won" or "Lost").
Next is Recommendation Builder. This tool helps you create personalised recommendations for your customers, again without code. A recommendation is a suggestion for a product, service, or piece of content that is tailored to a specific individual, based on their past behaviour and preferences. Think of how Netflix recommends shows you might like, or how Amazon suggests "Customers who bought this also bought...". Recommendation Builder does the same thing inside Salesforce.
How does it work? The tool analyses data from your CRM — like past purchases, products a lead has viewed on your website, or support cases they have opened — and identifies patterns. For instance, a customer who bought a laptop from your company within the last six months is very likely to be interested in a laptop bag. Recommendation Builder can then surface that suggestion when a service agent is chatting with that customer. The recommendation appears in a small box within the Salesforce interface, and the agent can click to add it to the order or share the suggestion with the customer.
Before Recommendation Builder, companies had to manually create product bundles or rely on sales reps to remember to upsell. This tool automates that process, using AI to find the most relevant recommendations in real time. The exam will expect you to know that Recommendation Builder uses a company's own data — not external data — and that it can be used for products, articles, or any other standard or custom object.
Finally, we have Next Best Action. This tool goes one step further than a recommendation. Instead of just suggesting a product, Next Best Action tells the user what specific action they should take next in a customer interaction. An action could be "Send a follow-up email", "Offer a 10% discount", "Schedule a phone call", or "Add the customer to a loyalty program". The tool is powered by business rules and AI, meaning you can combine simple if-then logic ("If customer is a VIP, then offer priority shipping") with predictive scores ("If customer is predicted to churn, send a retention offer").
Next Best Action is implemented using something called "strategies". A strategy is a set of rules and conditions that define when a particular action should be shown. For example, a strategy might say: "If the customer's support case is older than 5 days AND the case priority is High, then the next best action is to escalate to a senior agent." These strategies can be stacked, so the system shows the most relevant action first. The result appears as a recommendation in the Salesforce console — often a button or link the user can click to execute the action instantly.
What did businesses do before Next Best Action? They relied on manual playbooks — long documents that told staff what to do in different scenarios. Next Best Action digitises and automates that guidance, making it dynamic and personalised for each customer. The exam will test that you know Next Best Action is not the same as Recommendation Builder: it is about actions, not just product suggestions. Also, remember that strategies can be versioned and A/B tested, meaning you can compare which strategy performs better.
In summary, these three tools form an intelligent progression. Prediction Builder tells you what will happen. Recommendation Builder tells you what the customer might want. Next Best Action tells you what you should do right now to get the best outcome. Together, they turn raw data into a guided, automated customer journey.
Identify the Business Problem
Before using any Einstein tool, you need to decide what you want to achieve. For Prediction Builder, you need a specific outcome to predict, like 'Will this lead convert?' For Recommendation Builder, you need a goal like 'Increase upsell of phone cases.' For Next Best Action, you need a scenario where a specific action helps, like reducing customer churn. This step is about mapping the business requirement to the correct tool.
Access the Einstein Tool in Setup
In Salesforce, navigate to Setup using the gear icon. Type 'Einstein' in the Quick Find box. You will see options for 'Einstein Prediction Builder', 'Einstein Recommendation Builder', and 'Einstein Next Best Action'. Click to enter the respective configuration page. You need the 'Manage Einstein' permission to see these options.
Configure the Model or Strategy
For Prediction Builder: choose the object (e.g., Lead), the field to predict (e.g., Converted), and the fields to use for training (e.g., Lead Source, Industry). Then click 'Run' to train the model. For Recommendation Builder: select the source data (e.g., Order History), define the recommendation type (e.g., Frequently Bought Together), and set the target object (e.g., Product). For Next Best Action: create a new strategy, define the conditions (e.g., 'Last purchase > 60 days'), and link to a flow or template for the action (e.g., 'Send Discount Email').
Activate and Assign Permissions
After training the model or building the strategy, you need to activate it. Activation means the tool goes live and starts producing results. You also need to assign the appropriate user permissions or profiles so that specific users (like sales reps) can see the predictions, recommendations, or actions in their Salesforce view.
Monitor and Refine
AI models are not static. You should regularly check the performance of your Prediction Builder model (e.g., its accuracy score) and retrain it with new data. For Recommendation Builder and Next Best Action, you can run A/B tests on different strategies to see which one leads to better outcomes. This step ensures your tools remain effective over time.
Let us walk through a realistic scenario at a mid-sized company called "GadgetFlow", which sells consumer electronics both online and through a telesales team. GadgetFlow uses Salesforce Sales Cloud and Service Cloud. The company has hired you (as an aspiring AI Associate) to help them use Einstein AI tools to improve their customer engagement.
The first problem GadgetFlow faces is that their sales team spends too much time on leads that never buy. The sales director asks you to build a model that predicts which new leads are most likely to make a purchase within the next 30 days. You open Einstein Prediction Builder in Setup. You choose the Lead object as your target, and select "Converted" as the outcome field (meaning the lead became a customer). The tool automatically asks you to choose which data fields to include — you pick Lead Source (e.g., Webinar, Website, Referral), Industry, Company Size, and Number of Employees. You click "Run". After a few minutes, Einstein returns a model with a score. The model shows that leads from webinars are 3x more likely to convert than those from cold calls. Now, every new lead gets a "Conversion Likelihood" score from 0% to 100%. Your sales reps are instructed to call leads with a score above 70% within the first hour. This is Prediction Builder in action.
Next, GadgetFlow wants to increase average order value. They decide to use Recommendation Builder. The company has historical data showing that customers who buy a smartphone often buy a screen protector and a case within the next week. You open Recommendation Builder and create a new recommendation strategy based on the Product object. You set the source data as "Order History" and the recommendation type as "Frequently Bought Together". The system processes the last 12 months of orders and generates a rule: "Customers who buy a smartphone from the X200 series are 80% likely to also buy a screen protector." Now, when a customer adds a smartphone to their online cart, the website instantly shows a pop-up: "Add a screen protector for £9.99 — 9 out of 10 buyers do." If the customer is on a call with a telesales rep, the recommendation appears in Salesforce, and the rep can read it out loud to the customer.
Finally, GadgetFlow wants to reduce customer churn — customers who stop buying. A customer named Priya has not purchased anything in 60 days. AI Next Best Action analyses her profile: she has an open support case for a faulty charger from her last phone order. The system runs a strategy that says: "If last purchase date > 60 days AND open support case status = 'In Progress', then the next best action is to escalate the case and send a 15% discount voucher for accessories." The tool surfaces this action to the service agent handling Priya's call. The agent clicks the button, the case is escalated, and the voucher is automatically emailed. Priya is happy, and she stays a customer. As an AI Associate, you would set up this strategy using the Next Best Action configuration in Setup, defining the criteria and the action templates (like "Email Discount", "Escalate Case", "Schedule Callback"). The exam will expect you to recognise this workflow — prediction, recommendation, and next action — as a complete intelligent customer journey.
The Salesforce AI Associate exam (objective 4.2) tests very specific knowledge about what each tool does and, crucially, what it does not do. The trick is that the exam will give you scenarios and ask you to pick which tool to use. You will not need to configure the tools in the exam, but you must be able to distinguish between them cleanly.
The most common question type is a multiple-choice scenario. For example: "A sales manager wants to know which of their current sales opportunities are most likely to close this quarter. Which Einstein tool should they use?" The answer is Prediction Builder. The trap is that some candidates choose "Next Best Action" because they think "action" means "closing a deal". Remember: Prediction Builder predicts an outcome (like close probability), while Next Best Action tells you what to do next (like send a follow-up).
Here are the exact concepts the exam loves to test:
Prediction Builder is for predicting a binary or numeric outcome (e.g., will an opportunity close? what will the deal amount be?). It uses historical data from your Salesforce org.
Recommendation Builder is for suggesting items (products, articles, content) based on patterns like "frequently bought together" or "customers also viewed". It cannot predict outcomes.
Next Best Action is for suggesting specific business actions (send an email, assign a task, start a flow). It uses strategies that combine rules and AI scores.
The exam will ask about the difference between recommendations and next best actions. A recommendation suggests a product; a next best action suggests a step to take.
Trap patterns to watch out for:
"Can Prediction Builder predict customer churn?" Yes, if you have data on churn (e.g., a field called 'Churned') and you train the model on that field.
"Can Recommendation Builder be used to recommend a next phone call?" No — that is an action, not a product. Use Next Best Action.
"Does Next Best Action require coding?" No — all three tools are no-code, point-and-click.
"Do these tools require external data?" No — they all use data already in your Salesforce org (standard, custom objects, or even data from connected apps via Einstein Data Insights).
"Can you use all three tools together?" Yes — a common scenario is: Prediction Builder predicts a risk, Recommendation Builder suggests a product, and Next Best Action offers the action to email the product recommendation.
Key definitions to memorise for the exam:
Model: a trained AI algorithm that makes predictions.
Score: the predicted value (e.g., 85% conversion likelihood).
Strategy: a set of rules and conditions used by Next Best Action.
Recommendation: a suggested product or content item.
Action: a suggested task or step (e.g., call, email, update record).
The exam may also ask about related concepts like "Einstein Discovery" (which is a separate tool for deeper data analysis and insights). Be careful not to confuse Discovery with Prediction Builder. Discovery explains why something happened; Prediction Builder says what will happen next. Finally, know that all three tools appear in the "Einstein" section of Salesforce Setup, and they require the appropriate user permissions (usually "Manage Einstein" or specific permission sets).
Prediction Builder predicts a specific outcome (like deal closure) using your Salesforce historical data — no code required.
Recommendation Builder suggests products or content to customers based on patterns like 'frequently bought together' or 'customers also viewed'.
Next Best Action tells you what step to take right now (e.g., send an email, escalate a case) using strategies that combine rules and AI scores.
All three Einstein AI tools use data already in your Salesforce org and are configured through Setup with point-and-click wizards.
The main difference between Recommendation Builder and Next Best Action is that recommendations suggest items, while next best actions suggest steps or tasks.
These tools can be used together in a workflow: predict a risk, recommend a solution, and then take the next best action.
The AI Associate exam expects you to match the correct tool to a business scenario — not to configure them.
Next Best Action uses a concept called 'strategies' which are sets of conditions and actions that can be versioned and tested.
These come up on the exam all the time. Here's how to tell them apart.
Prediction Builder
Predicts an outcome like 'will this deal close?'
Outputs a numeric score or percentage
Uses a trained AI model based on historical data
Next Best Action
Suggests a step to take like 'call the customer'
Outputs a specific action button or text
Uses strategies combining rules and AI scores
Recommendation Builder
Suggests a product or content item
Uses patterns like 'frequently bought together'
Output is a list of items to show the customer
Next Best Action
Suggests a task or step
Uses rules and conditions in strategies
Output is a specific action the user should take
Prediction Builder
Focuses on forecasting a future outcome
Requires a defined outcome field (e.g., Won/Lost)
Creates a single prediction model per configuration
Recommendation Builder
Focuses on suggesting relevant items in the moment
Requires historical transaction data (e.g., orders)
Creates multiple recommendation strategies
Mistake
Einstein Prediction Builder can only predict whether an opportunity will close or not.
Correct
Prediction Builder can predict any binary or numeric outcome from any standard or custom object, as long as you have historical data. For example, it can predict the number of days until a support case is closed, or whether a lead will become a contact.
Beginners often see examples about opportunities and assume that is the only use case. The exam deliberately tests this by giving scenarios with other objects like Cases or Leads.
Mistake
Recommendation Builder can recommend actions like 'Send an email' or 'Call the customer'.
Correct
Recommendation Builder recommends items (products, articles, content). Actions are the domain of Next Best Action. A recommendation is a thing; an action is a step you take.
The word 'recommendation' sounds generic, so beginners assume it covers all suggestions. The exam tests this distinction directly.
Mistake
Next Best Action only works for sales scenarios, like upselling.
Correct
Next Best Action works across Sales, Service, Marketing, and any custom app. For instance, it can suggest 'Escalate the case' in Service Cloud or 'Send a re-engagement email' in Marketing Cloud.
Salesforce is known for Sales, so many new learners forget that Service and Marketing also use these tools.
Mistake
You need to install a separate app to use Prediction Builder because it is not part of standard Salesforce.
Correct
Prediction Builder, Recommendation Builder, and Next Best Action are all part of Salesforce Einstein, which is included with many Salesforce editions (e.g., Enterprise, Performance, Unlimited) with additional licensing. They are configured through Setup without external installations.
People assume AI tools require complex setup or third-party software, but Salesforce has built these in as native features.
Mistake
All three tools require you to write code to train the AI models.
Correct
All three are no-code tools. Prediction Builder has a simple wizard that guides you through selecting objects and fields. Recommendation Builder uses point-and-click strategy creation. Next Best Action uses declarative strategy builders or flow-based actions.
The term 'AI' often makes beginners think it requires programming. Salesforce designed these tools for administrators without coding skills.
Reveal each answer, then mark whether you got it right. Score 60%+ to unlock the next chapter.
Yes, as long as your Salesforce org has a field that stores whether a customer churned (e.g., a checkbox called 'Churned'), you can use Prediction Builder to create a model that predicts churn for future customers. No coding is required.
Recommendation Builder suggests items like products or articles (a thing to offer). Next Best Action suggests a step to take, like 'Send a follow-up email' or 'Escalate a case' (an action to perform). They are complementary but solve different problems.
Einstein Prediction Builder is available with Salesforce Enterprise Edition and above, but it requires an additional Einstein add-on license or is included in certain Einstein packages. Check your org's license management in Setup to see if you have access.
No, Next Best Action recommends actions, not products. To recommend a product based on a prediction, you would use Prediction Builder to identify the customer, then Recommendation Builder to suggest the product, and then optionally Next Best Action to prompt the user to 'Add the recommended product to the order'.
Salesforce recommends having at least 500 records (e.g., opportunities) and a minimum of 50 positive outcomes (e.g., 50 won opportunities) and 50 negative outcomes (e.g., 50 lost opportunities) to train a reliable model. The more high-quality data you have, the better the predictions.
Yes, Recommendation Builder can recommend any standard or custom object, including articles from Salesforce Knowledge. This is useful in Service Cloud, where a service agent might see a suggested article to share with a customer who has a specific issue.
You've finished Einstein AI Tools: Prediction, Recommendations, and Next Best Action. Continue through the AI Associate study guide to build a complete picture of the exam.
Done with this chapter?