How to predict churn, upgrades, and sales using your existing crm data (no code needed)

Your CRM is full of useful data: plans, payments, logins, tickets. You could be using that data to see who’s likely to cancel, upgrade, or buy.

Here’s a no-code system that turns your data into predictions.

Step 1: Pick one thing you want to predict

Before you touch any tools or data, stop and ask yourself: “What do I wish I could see coming with my customers?”

It could be something like:

• Who’s likely to cancel
• Who’s likely to upgrade
• Which leads are most likely to buy

Choose one question to focus on. That’s what your system will predict.

Step 2: Find old customer info

Go to where you keep customer data — like Stripe, Airtable, your CRM, or a spreadsheet.

Download a list of customers you’ve had before.

Each row should be one customer. Try to include:

• Things you know about them (like their plan or signup date)
• What they did (like how many times they logged in)
• What happened (did they leave? Did they buy something?)

Put everything into a new spreadsheet.

Make sure there’s one column that says what happened — like:“Churned: Yes or No”“Upgraded: Yes or No”

That’s all. Even 100–200 rows is enough to get started.

This is what the AI will learn from to make future predictions.

Step 3: Train your prediction model

Now that you’ve collected and cleaned your customer data — and you’ve got one clear outcome to predict — it’s time to train your first machine learning model.

We’ll use BigML for this walkthrough.

Here’s exactly what to do:

Create a free account at BigMLGo to bigml.com and sign up. You’ll get a 14-day free trial.

Upload your dataFrom the dashboard:

1. Click “Create → Source”
2. Choose “Local file” to upload your CSV, or connect a cloud provider
3. Once the file is uploaded, BigML parses it automatically
4. Review the detected field types (e.g. categorical, numeric)
5. Click “Create Dataset”

This transforms your raw data into a structured dataset BigML can work with.

Train your model

1. Open your dataset from the dashboard
2. Click “Configure → Supervised Model” (or just click the “1-click model” icon)
3. Select your target field (your outcome column)
4. Click “Create Model”

BigML will choose the appropriate algorithm (e.g. decision tree or logistic regression) depending on data type.

Review what the model learnedOnce training finishes (usually under 1 minute), you’ll see:

• A visual decision tree showing how the model makes predictions
• The top fields driving outcomes (like login frequency, email opens, or support tickets)
• A confidence score for each prediction

You can click on any branch in the tree to explore how it came to that conclusion.

Save the modelClick “Actions” → “Publish” to save and reuse the model later. You can now upload new customer data and get instant predictions.

Key details to get right:

• You must have at least one outcome column with historical labels (e.g. Churned = Yes/No, or Upgraded = Yes/No). This is what the model learns from.
• Make sure there’s enough variety in the outcome. If 95% of your customers didn’t churn, the model won’t learn much. You want a good mix of Yes and No outcomes.
• Don’t overthink it. If your data’s a bit messy or limited, that’s fine. This is about momentum, not perfection.

Step 4: See what causes what

Once the AI is trained, it tells you what things lead to certain results.

For example, it might tell you that:

• Customers who don’t log in within 7 days are 3x more likely to churn.
• Customers who open 3+ emails in the first week are 5x more likely to upgrade.
• Customers who contact support within 2 days of signup are high-risk.

These are clues. Write them down.

Step 5: Start running predictions on new customers

Now the fun part: Use your model to predict outcomes for current customers.

Do this:

1. Open your model in BigML.

2. Click “Predict.”

3. Upload a new CSV file with customer data (same format as before — just leave out the column that says if they churned, upgraded, etc.).

4. BigML will look at each customer and tell you what it thinks will happen.

   • Will they churn?
   • Will they upgrade?
   • Will they convert?

It also gives you a confidence score — how sure it is.

Example:

• Row 1: Might churn — 91% sure
• Row 2: Might upgrade — 88% sure
• Row 3: Won’t convert — 74% sure

You can download these predictions and add them to your spreadsheet or CRM.

Now use the predictions:

Set simple rules for what to do:

• If churn risk is over 80%, send them a check-in email
• If upgrade chance is over 90%, show them a special offer
• If they’re likely to convert, move them to the top of your sales list

At this point, keep it manual. Just do it once a week. Look at the predictions. Take action. Learn what works.

Once that feels easy, you can automate it.

Step 6: Automate it with no-code tools

Once you trust your predictions, connect everything with tools like Make, Zapier (with Webhooks and Code), or n8n.

Example: Auto-churn alerts for new signups (using Make or Zapier)

Capture new users

• Add their data to a Google Sheet called New Signups

Send data to BigML

• Use an automation tool like Make or Zapier (with Webhooks and Code) to send each new row to BigML’s API.
• Trigger: New row added in New Signups
• Action: Format the data to match your model
• Action: Call BigML's Batch Prediction API (or use the Single Prediction API) with your trained model
• BigML returns a prediction and confidence score

Filter results

• If churn = Yes and confidence > 80%, continue
• Else, skip

Take action

• Send an email (Gmail/Mailchimp)
• Post alert to Slack
• Tag in CRM

You’ve now automated your first prediction-driven workflow.

Clone this for other outcomes

You can build the same system for:

• Upgrade potential: Add to an upsell campaign
• Conversion score: Notify sales
• Low engagement: Trigger onboarding email

Same steps. Different predictions.

When to run it

• Real-time: trigger the workflow every time new data comes in
• Daily batch: export from Stripe or your CRM → run predictions → trigger workflows
• Weekly digest: send yourself a summary of high-risk or high-opportunity users

Step 7: Refine the model over time

Your AI model is like a new hire. It gets smarter with experience — but only if you retrain it.

Every month or two:

• Export new customer data
• Add outcome labels (did they churn? upgrade?)
• Feed it back into your model and retrain

This closes the loop. Over time, your predictions get sharper. Your system gets faster. Your results get better.

Want to automate this?

You can. Here’s a simple version:

1. Set up a weekly data sync. Use Make( or Zapier) to automatically export updated customer records to a Google Sheet.
2. Auto-label the outcomes. Use tags or filters (like “Plan Cancelled” = Churned) to label the rows. You can also script this if needed.
3. Trigger retraining via BigML API. BigML lets you automate uploads and training. Set a weekly schedule using a tool like Make, and keep your model improving in the background.

One last thing: Use your judgment

Don’t turn your brain off.

Before you automate messaging or decisions:

• Sanity check the logic
• Adjust confidence thresholds
• Monitor model accuracy (especially after product changes)

Automation is powerful, but only if it’s built on sound logic and good data.