# Predictions, Measures, and KPIs in Customer Insights

> Chapter 17 of the Courseiva MB-910 curriculum — https://courseiva.com/learn/mb-910/customer-insights-predictions-and-measures

**Official objective:** 5.3 — Identify key analytics, measures, KPIs, and AI-driven predictions available in Dynamics 365 Customer Insights to understand customer behavior.

## Introduction

How do you know which customers are about to leave you, and which ones are your most valuable, before you check any sales reports? That is the core problem that Predictions, Measures, and KPIs in Dynamics 365 Customer Insights solve: they transform confusing customer data into clear, actionable signals without needing a data science degree. For the MB-910 exam, understanding how these three tools work together is essential because the test will ask you to distinguish between a basic calculation (a Measure), a goal-based comparison (a KPI), and an AI-generated forecast (a Prediction), and you need to know the exact name and purpose of each one.

## The Head Chef's Dashboard Analogy

A busy professional kitchen, 45 minutes before the dinner rush begins. The head chef isn't cooking; she's standing at a large screen showing a live dashboard of the entire restaurant's operation.

The screen displays three key things. First, 'Measures' — raw numbers from the past hour: '42 orders taken,' '17 portions of salmon left in the fridge,' '3 tickets waiting longer than 15 minutes.' These are simple, factual counts, like the basic statistics in Customer Insights that tell you how many customers you have or how many purchases they made last month. They are the starting point, not the whole story.

Next to those numbers, the chef sees 'KPIs' (Key Performance Indicators) — colour-coded targets. The 'Average time from order to plate' is currently 12 minutes, but the KPI target is under 10. The gauge shows red. The 'Number of customer complaints per night' is zero, which is green. KPIs take those raw Measures and compare them to a business goal, showing success or failure at a glance. In Customer Insights, a KPI might be 'Customer Lifetime Value' or 'Churn Rate,' telling you if your customer relationships are getting healthier.

Finally, the dashboard shows 'Predictions.' A small box reads: 'Based on current order velocity and staffing levels, the kitchen will hit a bottleneck at 7:15 PM — 85% probability. Suggest adding one more prep cook.' This isn't a guess; it's an AI-driven forecast using past data. In Dynamics 365 Customer Insights, predictions work the same way: the system analyses customer behaviour patterns to forecast what they will do next — like predicting who is likely to stop buying your product in the next 30 days.

The chef uses all three levels together. Measures tell her the pulse right now, KPIs tell her if she's winning, and Predictions let her act before disaster hits. That is exactly what Customer Insights does: it turns scattered customer data into that chef's dashboard, so businesses can serve the right meal at the right time.

## Core explanation

Let us start with the big picture. Every business collects data about its customers: who they are, what they buy, when they visit a website, if they open an email. The problem is that this data lives in many different places — sales systems, marketing tools, customer service tickets, loyalty programmes — and it is often messy and unconnected. Dynamics 365 Customer Insights is a tool that pulls all of that customer data into one unified profile for each person. But once you have that single view, what do you actually do with it? That is where Measures, KPIs, and Predictions come in. They are the three layers of intelligence that turn raw data into business decisions.

A Measure is the simplest layer. Think of it as a straightforward count or calculation. For example, 'Total number of customers who purchased in the last 90 days' is a Measure. 'Average spend per transaction' is a Measure. 'Number of support cases opened last week' is a Measure. In Customer Insights, you can create custom Measures using a visual interface — you do not need to write code. You just specify what you want to count (or average, or sum) and over what time period. Measures are important because they give you the raw facts. They answer the question: 'What happened?' But by themselves, they lack context. Is a high number of support cases good or bad? A Measure cannot tell you that on its own.

A KPI (Key Performance Indicator) is a Measure that has been given a target or benchmark. It answers the question: 'Are we doing well or poorly?' For example, a KPI might be 'Customer Retention Rate' with a target of above 85%. If the actual rate is 90%, the KPI is green (success). If it is 75%, the KPI is red (action needed). KPIs are always tied to a business goal. In Customer Insights, you can create KPIs from your existing Measures by simply adding a goal. The system then tracks whether that measure is hitting, missing, or exceeding your target over time. KPIs are what executives look at on dashboards to get a quick health check of the customer base.

A Prediction is the most advanced layer. It uses artificial intelligence (AI) and machine learning models built into Customer Insights to forecast future customer behaviour based on past patterns. The system analyses your historical data — purchase history, web browsing, support interactions, demographic details — and identifies which factors are most predictive of a future event. There are several pre-built prediction models in Customer Insights. The most commonly tested ones in MB-910 are:

- Subscription Churn Prediction: This predicts which customers are likely to cancel their subscription in the next 30, 60, or 90 days. It ranks customers by churn risk (high, medium, low) and shows you the top reasons (like 'low engagement' or 'recent price increase').
- Transaction Churn Prediction: This predicts which customers are likely to stop making purchases (even without a subscription). Useful for retail and e-commerce.
- Customer Lifetime Value (CLV) Prediction: This predicts how much total profit a customer will generate over their entire relationship with your business. It helps you decide how much to spend on acquiring or retaining them.
- Product Recommendation: This predicts which products a specific customer is most likely to buy next. It powers personalised 'recommended for you' features.

How do these predictions actually work? The system takes a portion of your historical data (say, data from the last 12 months) and trains a machine learning model to find patterns. For churn, it might learn that customers who have not opened an email in 45 days AND who have called support more than twice in a month are 80% likely to churn. Once the model is trained, it applies that pattern to your current customers and scores each one. The output is a list of customers sorted by probability of churning, along with the top contributing factors. You can then take action: send a special offer to high-risk customers, or route them to a retention team.

A crucial distinction for the exam: Measures and KPIs look backward (they tell you what has already happened). Predictions look forward (they tell you what is likely to happen next). Measures and KPIs are calculated from data you already have. Predictions are generated by AI models that learn from that data to make forecasts.

Why should a business use Customer Insights instead of a spreadsheet? Spreadsheets can calculate sums and averages (Measures) and you could manually colour-code cells to indicate targets (KPI-like). But spreadsheets break when you have millions of customers, data from dozens of sources, and you want AI predictions without hiring a data scientist. Customer Insights automates the entire pipeline: data ingestion, identity resolution (matching a customer across different systems), calculation of Measures and KPIs, and the running of AI prediction models, all in one place. It also refreshes automatically, so your Measures and KPIs are always up-to-date, and your predictions are retrained periodically on new data to stay accurate.

## Real-world context

Let us imagine you work for a medium-sized online clothing retailer called 'StyleSphere.' The company sells via a website, a mobile app, and a physical store. They also run an email loyalty programme and have a customer service phone line. The marketing manager, Priya, has a big problem: she does not know which customers are at risk of leaving, and she is spending her entire budget on advertising to everyone equally, wasting thousands of pounds on people who were already loyal.

Priya gets access to Dynamics 365 Customer Insights. First, she connects the data sources: the e-commerce platform, the app analytics, the in-store point-of-sale system, the email marketing tool, and the customer service system. Customer Insights identifies each unique customer across these systems (this is called 'unified profile') — so 'John.Smith@email.com' who bought online is the same John Smith who returned a jacket in-store.

Now, Priya wants to understand her customer base. She starts by creating some Measures:
- 'Total Revenue Last 30 Days': a sum of all purchases.
- 'Average Order Value': total revenue divided by number of orders.
- 'Repeat Purchase Rate': percentage of customers who bought more than once in 90 days.

These Measures give her the raw numbers. She sees that Average Order Value is £45, which is okay, but the Repeat Purchase Rate is only 18%, which seems low. But is that good or bad? She does not know, so she needs KPIs.

Priya sets up KPIs:
- KPI: 'Repeat Purchase Rate >= 25%' — current value is 18%, so this KPI is red.
- KPI: 'Customer Satisfaction Score >= 4.5/5' — current value is 4.6, so green.
- KPI: 'Monthly Active Customers >= 5000' — current value is 4800, so amber.

Now she has a clear visual dashboard: red means action required, green means on track. She knows that improving the repeat purchase rate is a priority. But she still does not know which customers to target. That is where Predictions come in.

Priya runs the 'Transaction Churn Prediction' model. The system analyses the last 12 months of data and generates a list of customers with churn risk scores. It also tells her the top factors driving churn: customers who have not visited the website in 60 days, who did not open the last three promotional emails, and who returned an item recently are 70% more likely to churn. The model outputs a segment called 'High Risk of Churn' containing 450 customers.

Next, Priya runs the 'Customer Lifetime Value Prediction' model. This tells her that the top 10% of customers will be worth an average of £1200 each over the next three years, while the bottom 10% are worth only £50. She creates a segment of 'High Value Customers at Risk of Churn' by intersecting the two predictions (those who are high lifetime value AND high churn risk). This list contains 62 customers.

The IT professional's job here is to:
- Set up the data connections (the 'plumbing').
- Configure the Measures, KPIs, and Prediction models in the Customer Insights interface.
- Ensure the data is clean and properly mapped (e.g., making sure 'order date' is recognised as a date field).
- Schedule the measures and predictions to refresh daily or weekly.
- Create dashboards and reports that display the KPIs and prediction outputs for the business users like Priya.
- Explain to Priya that predictions are probabilistic, not certain — a 70% churn risk means 30% of those customers will not churn, so the business should not treat them all as lost.

Priya then uses the output: she sends a personalised 'We miss you' email with a 20% discount code to the 62 high-value at-risk customers. The result? Within two weeks, 22 of those 62 make a repeat purchase. The prediction worked, and the business saved money by not wasting discounts on customers who were never going to churn anyway.

## Exam focus

The MB-910 exam tests your recognition of what these concepts are, not how to implement them. You will not be asked to write code or configure a model. The questions are multiple-choice and scenario-based. Here are the specific patterns to expect.

First, you must be able to distinguish between a Measure, a KPI, and a Prediction. This is the most common trap. The exam will give you a scenario: 'A company wants to track the number of customers who made a purchase last month.' The answer is 'a Measure,' because it is just a count with no target. If the scenario says, 'The company wants to track whether the number of repeat purchases meets their goal of 30% of customers,' the answer is 'a KPI' because it includes a benchmark. If the scenario says, 'The company wants to forecast which customers are likely to stop buying in the next 30 days,' the answer is 'a Prediction' (specifically, Transaction Churn Prediction).

Second, they love to test the specific names of the built-in prediction models. Memorise these exactly:
- Subscription Churn Prediction (for recurring revenue models like SaaS)
- Transaction Churn Prediction (for one-off purchases, retail)
- Customer Lifetime Value Prediction (total profit over the relationship)
- Product Recommendation (what to buy next)

A typical trap question: 'Which prediction model should a music streaming service use to identify customers who will cancel their monthly plan?' The correct answer is 'Subscription Churn Prediction,' not 'Transaction Churn Prediction,' because it is a subscription model, not a transactional one.

Third, they test the concept of 'unified customer profile' as a prerequisite. You need to know that Measures, KPIs, and Predictions all work on the unified customer data that Customer Insights creates. A question might say: 'Before creating a Customer Lifetime Value prediction, what must a company first do?' The answer involves connecting data sources and creating unified profiles (also called 'customer data platform' functionality).

Fourth, watch for questions about who can create these. In Customer Insights, business users (like marketers) can create Measures and KPIs using the visual interface without coding. Predictions can also be configured by business users using pre-built templates. The exam does not expect you to know the exact steps, but it expects you to know that it is not purely for developers.

Fifth, be aware of the phrase 'key performance indicator' — they might use the full name or the acronym KPI. Both are testable. Also, note that KPIs are always tied to a business goal — if there is no goal, it is not a KPI.

Sixth, the exam may ask about the refresh frequency. Customer Insights can refresh data on a schedule (daily, weekly) or on-demand. Predictions are usually retrained periodically (e.g., every two weeks) as new data comes in. You do not need to memorise exact intervals, but know that predictions get updated automatically.

Finally, they test the benefit of predictions: they enable proactive action. A question might ask: 'What is the main advantage of using churn prediction over simply looking at past churn rates?' The correct answer is that it allows the business to intervene before the customer leaves, rather than just reporting that they have already left.

Key definitions to memorise:
- Measure: a calculated value (count, sum, average) from customer data, no target attached.
- KPI: a measure plus a target, used to gauge performance against a goal.
- Prediction: an AI-generated forecast of future behaviour, such as churn or lifetime value.
- Churn: when a customer stops doing business with you (cancels subscription or stops purchasing).
- Customer Lifetime Value (CLV or LTV): the total profit a customer is expected to generate over their entire relationship with the business.
- Unified Customer Profile: a single, 360-degree view of a customer assembled from all data sources.

## Step by step

1. **Connect Your Data Sources** — Before you can calculate anything, you need to bring customer data from all your systems (e-commerce, CRM, email, support) into Customer Insights. This step involves configuring connections to each data source. It matters because Predictions, Measures, and KPIs are only as good as the data feeding them — incomplete data leads to inaccurate forecasts and misleading KPIs.
2. **Create Unified Customer Profiles** — Customer Insights matches records across your different data sources to identify each unique customer (for example, linking the same person's email address from your website to their loyalty card number from the store). This step is essential because without unified profiles, a Measure like 'total spend per customer' would count the same person multiple times, giving wrong numbers.
3. **Define and Build Your Measures** — Using the visual builder, you create the basic metrics you need, such as 'Total Orders Last Quarter' or 'Average Support Tickets per Customer.' These are the raw building blocks. No targets are set yet — you are just calculating the facts. This matters because every KPI you later create will reference one of these Measures.
4. **Set Up KPIs with Targets** — You take one of your Measures (e.g., 'Repeat Purchase Rate') and assign a business goal to it (e.g., '25%'). Customer Insights will now track whether the actual value meets, exceeds, or falls short of the target, and show this with colour coding (green, amber, red). This matters because it turns raw data into actionable business intelligence for decision-makers.
5. **Run a Prediction Model** — You select a built-in model (e.g., Transaction Churn Prediction) and configure it with your data. The system trains a machine learning model using your historical data, then scores every current customer with a churn probability and the top influencing factors. This matters because it gives you a forward-looking view — you can now see not just who has left, but who is likely to leave next.
6. **Review and Act on the Insights** — You examine the prediction output (e.g., a segment of 'High Risk Customers with High Lifetime Value') and the KPI dashboard to decide on actions — like sending targeted offers or routing customers to a retention team. This matters because the whole point of Measures, KPIs, and Predictions is to drive real business actions that improve customer relationships and revenue.

## Comparisons

### Measure vs KPI

**Measure:**
- A raw calculation (count, sum, average) with no goal
- Answers the question: 'What happened?'
- Example: 'Total orders last month = 1,200'

**KPI:**
- A Measure with a target or benchmark attached
- Answers the question: 'Are we meeting our goal?'
- Example: 'Total orders last month = 1,200 against target of 1,500 — KPI is red'

### Subscription Churn Prediction vs Transaction Churn Prediction

**Subscription Churn Prediction:**
- Used for businesses with recurring revenue models (e.g., SaaS, streaming)
- Predicts cancellation of a subscription plan
- Looks for signals like missing payments or reduced usage

**Transaction Churn Prediction:**
- Used for businesses with one-off purchases (e.g., retail, e-commerce)
- Predicts when a customer will stop buying entirely
- Looks for signals like longer gaps between purchases or reduced basket size

### Prediction (AI-driven forecast) vs Measure (historical calculation)

**Prediction (AI-driven forecast):**
- Looks forward: forecasts future behaviour
- Uses machine learning to find patterns
- Output is a probability score (e.g., '70% chance of churn')

**Measure (historical calculation):**
- Looks backward: calculates what already happened
- Uses simple arithmetic (sum, average, count)
- Output is a definite number (e.g., '2,000 orders last month')

### Customer Lifetime Value Prediction vs Product Recommendation Prediction

**Customer Lifetime Value Prediction:**
- Forecasts the total profit a customer will generate over their entire relationship
- Used to decide how much to spend on retaining high-value customers
- Output is a monetary amount per customer

**Product Recommendation Prediction:**
- Forecasts which specific products a customer is most likely to buy next
- Used to personalise marketing emails or website recommendations
- Output is a list of product suggestions for each customer

## Diagram

_A flowchart showing the journey from raw customer data through unified profiles, Measures, KPIs, and AI Predictions, leading to actionable business outcomes in Customer Insights._

```mermaid
flowchart TD
    A[Data Sources: Sales, Marketing, Support, Website] --> B[Customer Insights Platform]
    B --> C[Step 1: Connect & Unify Profiles]
    C --> D[Step 2: Create Measures - Raw counts, sums, averages]
    D --> E[Step 3: Create KPIs - Measures with targets]
    D --> F[Step 4: Run AI Predictions - Models for churn, CLV, recommendations]
    E --> G[Dashboard: Red/Amber/Green Status]
    F --> H[Action: Target high-risk customers with offers]
    G --> H
    H --> I[Improved Retention & Revenue]
```

## Common misconceptions

- **Misconception:** A KPI and a Measure are the same thing because both involve numbers. **Reality:** A KPI is a Measure that has a specific target or goal attached to it. Without a goal, it is just a Measure. (Beginners hear 'key performance indicator' and think it is just a fancy name for any metric. They miss that the 'indicator' part means it indicates success or failure against a benchmark.)
- **Misconception:** Predictions in Customer Insights are just guesses based on gut feeling, similar to what a manager might assume. **Reality:** Predictions are generated by machine learning models that analyse thousands of data points and historical patterns to calculate a statistical probability. (The word 'prediction' sounds like fortune-telling to non-technical people. They underestimate the data science rigour behind it.)
- **Misconception:** Customer Insights can only work with data from other Microsoft products (like Dynamics 365 Sales or Power BI). **Reality:** Customer Insights can ingest data from almost any source, including third-party systems like Salesforce, Google Analytics, or custom databases, as long as they can be connected. (People assume Microsoft tools only play with other Microsoft tools. The exam tests that Customer Insights is an open customer data platform.)
- **Misconception:** If a customer is predicted to churn, they will definitely churn — the prediction is a certainty. **Reality:** A prediction is a probability, not a certainty. A 'high churn risk' label means a high likelihood, but the customer might still stay. Predictions are used to prioritise interventions, not to write customers off. (Non-technical managers often treat predictions as absolute truths. The exam tests that you understand the probabilistic nature of AI predictions.)
- **Misconception:** You need to write code or be a data scientist to create Measures and KPIs in Customer Insights. **Reality:** Measures and KPIs can be created by business users through a visual drag-and-drop interface. No coding is required. (Many beginners think all 'analytics' tools require programming. Microsoft designed Customer Insights to be accessible to marketers and sales managers.)

## Key takeaways

- A Measure is a simple calculation (count, sum, average) of customer data; it tells you 'what happened' without any judgement of good or bad.
- A KPI is a Measure with a target attached; it tells you 'are we succeeding or failing' against a business goal.
- A Prediction uses AI to forecast future customer behaviour based on historical patterns, allowing proactive action instead of reactive reporting.
- The four built-in prediction models you must know for MB-910 are Subscription Churn, Transaction Churn, Customer Lifetime Value, and Product Recommendation.
- Churn prediction models differentiate between subscription churn (cancelling a recurring plan) and transaction churn (stopping one-off purchases), and using the wrong one is a common exam trap.
- All Measures, KPIs, and Predictions in Customer Insights rely on having unified customer profiles first, which means data from different sources must be connected and matched to the same person.
- Predictions are probabilistic, not deterministic — a high churn risk score means 'likely,' not 'certain,' so businesses should use them to prioritise retention efforts, not to assume failure.

## FAQ

**What is the difference between subscription churn and transaction churn?**

Subscription churn applies to businesses with recurring payments (like Netflix or a SaaS tool) and predicts when someone will cancel their plan. Transaction churn applies to businesses where customers buy individual items (like a clothing store) and predicts when someone will stop making purchases altogether.

**Do I need to know how to code to create a prediction in Customer Insights?**

No. Customer Insights provides pre-built prediction models with a guided setup wizard. You select the model type, point it at your data, and the system handles the machine learning without any coding required.

**Can I create a KPI without first creating a Measure?**

No. A KPI is built on top of a Measure. You must first define the Measure (the raw calculation) and then add a target to it to turn it into a KPI. Customer Insights guides you through this process.

**How often do predictions update?**

Predictions are recalculated on a schedule that you configure, typically every two weeks or monthly when new data is available. They are not a one-time calculation; the system retrains the model periodically to keep it accurate.

**Can I use Customer Insights data in Power BI reports?**

Yes. Customer Insights can export its data (including Measures, KPIs, and prediction outputs) to Power BI, allowing you to build custom dashboards and reports that combine customer insights with other business data.

**What is a ‘unified customer profile’ and why do I need it for predictions?**

A unified customer profile is a single record for each customer that combines data from all your systems (e.g., name from the CRM, purchase history from e-commerce, support tickets from the help desk). Predictions need unified profiles because they analyse patterns across all this data; if a customer is split across three records, the prediction model sees three different people and makes inaccurate forecasts.

## Check your understanding

1. **A retail company wants to know if their repeat purchase rate of 22% meets their target of 30%. Should they use a Measure or a KPI?**

   Answer: They should use a KPI because it involves comparing a Measure (repeat purchase rate) against a specific target (30%) to evaluate performance.

2. **Which Customer Insights prediction model would a software company with monthly subscriptions use to identify customers at risk of cancelling?**

   Answer: They would use the Subscription Churn Prediction model, which is specifically designed for recurring revenue and subscription-based businesses.

3. **What is the main difference between a Measure and a Prediction?**

   Answer: A Measure calculates past data (e.g., 'total spend last quarter'), while a Prediction forecasts future behaviour (e.g., 'likelihood to churn in the next 30 days') using AI.

4. **True or False: You must have unified customer profiles before you can run a Customer Lifetime Value prediction.**

   Answer: True. Unified profiles ensure that all data about a customer (purchases, support, web activity) is linked to one ID, which is essential for accurate CLV predictions.

5. **An e-commerce site wants to show personalised product suggestions to each visitor. Which built-in prediction should they use?**

   Answer: They should use the Product Recommendation prediction model, which analyses purchase history to suggest items a customer is likely to buy next.

---

Interactive version with quiz and diagrams: https://courseiva.com/learn/mb-910/customer-insights-predictions-and-measures