SOLUTIONS · MARKETING & SEO
AI lead qualification, trained on which of your leads actually became customers
Qualify inbound leads with a private AI model trained on reviewed CRM outcomes. Route predicted categories through your workflow and prioritize uncertain cases for review.
AI lead qualification suggests a category for each new lead using the outcomes your team defines. StayCharted AI Model Trainer does it with a private model trained on your own CRM history: the leads sales accepted, the ones that became opportunities and the ones that didn't. Each new lead gets a qualification and a confidence. Your configured CRM or automation workflow can route approved categories, while a person reviews uncertain cases.
Use your own won/lost history or sales-accepted outcomes as candidate labels, then review them against your current qualification rules. A model learns patterns in those labels; it does not determine why a deal was won or lost.
Lead qualification trained on your CRM data
You can ask ChatGPT or Claude whether a lead looks good, and it will give a reasonable answer about job titles and company size. Without your examples and account context, it does not automatically know what a good lead looks like for you:
- which industries actually buy;
- which job titles turn into champions, and which never reply;
- which "free trial" messages come from students;
- which sources bring partners rather than customers.
That knowledge is in your CRM: past leads and their recorded outcomes. A model trained on that history learns your pattern, including the parts nobody has written down.
What it reads, and what it answers
It reads the fields you have when the lead arrives:
| Field | Example |
|---|---|
| Job title | "Head of Revenue Operations" |
| Company size, industry, country | 201–500 · Logistics · Germany |
| Lead source | Webinar · Pricing page · Partner referral |
| The message, if there is one | "We route 2,000 tickets a week and want…" |
It answers in your own categories, learned from what happened to past leads, for example:
- Sales-ready: became an opportunity.
- Nurture: a real buyer, not yet.
- Partner: wants to resell or integrate.
- Support: an existing customer in the wrong place.
- Not a fit: students, job seekers, vendors pitching.
Every answer comes with a confidence. Use it to prioritize review within each predicted category. High confidence in “Not a fit” does not make that lead valuable; confidence is not a lead-quality score or conversion probability. Your team sets the sales priority and routing rules.
How to set it up
1. Export past leads with what happened to them
A period such as the past twelve months can be a starting point, depending on your sales cycle. One row per lead, with the fields above and the outcome: lead status, “became opportunity”, or the final stage. Use completed, reviewed outcomes and keep unresolved leads out of the negative class. A recently arrived lead that has not converted is not necessarily “Not a fit”. Use the categories your team already uses; you don't have to invent new ones.
Leave out any field that's filled in later, such as owner, last activity, or deal amount. Those give the answer away, and new leads won't have them.
2. Upload and check
Upload the file (Excel, CSV or JSON), or train straight from Google Sheets with the StayCharted add-on. Choose the fields the model reads and the outcome column. StayCharted checks the data first:
- duplicate input rows;
- matching inputs with conflicting labels;
- outcomes with too few examples;
- personal information it can mask, such as phone numbers in messages.
3. Train and read the report
Train an AI Classifier Model. It reads meaning, so "VP RevOps" and "Head of Revenue Operations" can be represented by similar meaning. The report shows how often it was right on leads it never saw, outcome by outcome.
Test it on a later period, too. Train on leads up to three months ago, then fill the last three months and compare with what actually happened. Keep leads from the same company together when preparing your evaluation, and make sure outcomes in the test period have had time to mature. StayCharted’s automatic held-out report is not a substitute for this later-period, company-aware evaluation.
4. Inbound lead routing through your existing workflow
- Through an integration: for supported CRM triggers and actions, a Zapier or Make step sends each new form entry to the model and writes the qualification and confidence to a field in your CRM. Your CRM's own routing rules then assign it: sales-ready with high confidence to an account executive, the rest to an SDR to check.
- In batches: fill a spreadsheet of the week's leads, or use the Google Sheets add-on.
- From your own software: call the model through the API.
5. Retrain on new outcomes
Every quarter, export the leads that now have an outcome and retrain. Compare the updated model with the current version before publishing it. Retraining can help it adapt to changes, but does not guarantee improvement.
Start in suggestion mode and review a sample of confident answers as well as uncertain ones. API-based workflows require Essentials or Business; the Google Sheets add-on is available on every plan. Follow the current Zapier or Make setup guide and check that your CRM exposes the fields and actions you need.
What it doesn't do
- It doesn't look anything up. It reads only the fields you send. There's no enrichment and no web search.
- It isn't a points-based lead score. It gives a category and a confidence. Confidence measures certainty in that predicted category, not the commercial value of a lead.
- It doesn't contact leads or change your CRM by itself. Your integration writes the answer; your team and your rules decide what happens.
- Choose fields with care. Use relevant business fields you are permitted to process, and review whether past outcomes reflect outdated policies or uneven follow-up. Don't train on personal characteristics, or on anything you wouldn't want deciding who gets a call back.
Common questions
Can AI qualify leads automatically?
A trained model can suggest qualification categories for new leads. A configured integration can then apply your routing rules, with uncertain cases sent for review. StayCharted does not change your CRM or contact leads by itself.
How is this different from asking ChatGPT to qualify leads?
An assistant can use your rules and examples when you provide them. A trained model offers batch classification, confidence scores and held-out evaluation using the CRM inputs and outcomes you choose. Compare both on reviewed leads from a later period.
How is it different from the lead scoring in my CRM?
StayCharted predicts your chosen categories and returns confidence, rather than a points-based score or conversion probability. Connect supported CRM workflows through Zapier, Make or the API, or use spreadsheet exports. Check the integration capabilities and plan requirements.
How many past leads do I need?
Around 20 reviewed examples per outcome is a starting point, not a guarantee. Overlapping outcomes may need substantially more. Use mature outcomes, assess errors by category and evaluate on a later period before relying on the model.
Can it route leads to the right team as well?
Yes. Train a second model on which team handled each lead (sales, partners, support), or combine both in one set of outcomes.
Is our CRM data used to train anyone else's model?
No. The model is private to your workspace. See Security.