AI lead scoring for sales calls: putting a probability on every booked meeting
AI lead scoring for sales calls means an agent researches each person who books a meeting and puts a probability of them becoming a customer next to the meeting, before you take it. Amir on our team built exactly that on Claude Code. It shows the score right on the calendar, and Emiliano was already using it by our October 9 team call.
The first thing it taught us had nothing to do with the model. It was about what you ask the model to score. On October 12, a client with 26 leads that week found that his own AI had been ranking people for how good they'd be on his show, not for whether they'd ever buy. Those are different questions, and mixing them up quietly fills your calendar with the wrong people.
Why score before the call
When the system is working, you get a lot of booked conversations. In a show-led setup, each one is a 15-minute pre-interview first, then a 30-minute recording for the strongest guests, then a sales conversation for the ones who fit. That's a lot of founder time.
A score before the first call does two things. It tells you where to spend your energy: which pre-interviews deserve your full prep and which can be a friendly 15 minutes. And over time it tells you whether your list is getting better or worse, which is the feedback loop that makes the whole engine improve.
Without it, every call looks the same on the calendar. You find out who was a fit after you've spent the time.
What the scorer looks at
The scorer runs at the calendar stage. When someone books, it researches them and their company and produces a probability of converting, placed directly on the calendar event. On our October 9 webinar and our October 16 live session we described what goes into it: how well they fit the ideal customer profile, the size of their market, fit with the offer, estimates of their resources, the likely sales cycle and close rate.
The pre-interview booking form helps. Guests answer a few questions when they book, things like monthly revenue, their ideal clients and their biggest business challenge. That gives the scorer first-party answers to work with, not just what it can find online.
None of this decides for you. It's a number and a brief. You still take the call. You just walk in knowing what you're walking into.
Podcast fit is not buyer fit
Here's the story that made this click. On October 12 we were on a weekly call with a client in the events business. That week he'd had 26 leads and 8 pre-interviews done. The quality of those pre-interviews had been low, though it was still a small sample, and his show's landing page wasn't live yet, which may have been dragging quality down too.
When we looked at how his leads were being ranked, the problem was obvious. His AI was ranking leads for podcast fit. Would this person be a good guest? Interesting story, good on camera, well known in the space?
That's a reasonable thing to want for the show. It's the wrong thing to optimize the business for. A great guest who will never buy is a nice episode. A slightly awkward guest who runs exactly the company you sell to is a customer.
You want both on the show. But the scoring that drives your targeting and your time has to be about the business, about ideal customer fit, or you'll end up with a great podcast and an empty pipeline.
Averaging scores weekly to tune the list
The plan coming off that call: deploy a scoring agent on the client's calendar, connected through his scheduling tool's API. Emiliano would send over a Claude Code snippet to score each pre-interview call. The scores would be tracked on the weekly report and averaged week by week.
The weekly average is the part I like most. One score tells you about one person. The average tells you about your list. If the average is climbing, your targeting is getting sharper. If it's flat or falling, something upstream needs work: the list, the copy, or the audience you're inviting.
On the same call we also shut off one outreach channel that wasn't producing, kept the pre-interview campaigns running, and planned more contacts and campaigns for the week because volume was running low. Seven leads came in that day. More volume plus a score on each lead is how you find out quickly which parts are working.
What we learned building it with Claude Code
Amir built the first version himself on Claude Code, and the next step on the October 9 call was setting up a shared Claude Team account so everyone on the team could run it. That's the speed now available to a small team. A scorer that once would have been a quarter-long project with a data vendor became a snippet one person wrote and shared.
The hard part wasn't the code. It was deciding what "good" means for each client. A fintech founder's best lead looks nothing like an events company's. So every scorer starts from that client's ideal customer profile, and the weekly averages tell us when the definition needs to change.
What a score can't tell you
A score is a guess with good research behind it. It's wrong sometimes. Some of the best conversations I've had came from people who looked average on paper and turned out to know exactly the person I needed. And some high scores turned out to be tire-kickers with impressive websites.
So we treat the score as a way to prioritize prep and spot trends, not as a gate that keeps people off the calendar. The pre-interview itself is still the real filter. Fifteen minutes of conversation tells you more than any amount of research. The score just means you walk in already knowing what to ask.
Start with one question
If you want to try this yourself, keep it small at first. Don't start with a complicated model. Start with one question you want answered before every call: would this person plausibly buy what I sell in the next 90 days? Have the agent answer that, with reasons, for every booking. Look at the answers after two weeks next to what happened on the calls. Then add more factors if they help.
That comparison is the whole game. A scorer you never check against real outcomes is just a confident opinion. A scorer you check every week, and adjust, becomes the most honest view you have of your own list.
Can you score bookings automatically?
Yes, if your scheduling tool lets you connect to it, which most do through an API. The pattern is simple: a booking comes in, the agent reads the booking answers, researches the person and company, writes a score and a short brief, and attaches both to the calendar event.
In gtm showrunner, this lives in the prep step. Guests book onto your Calendly, and the system prepares you for each taping. The score and the questions are there before you join.
If you want to see how prep and scoring fit into the rest of the system, from invite to booked guest to sales call, it's on how we work.
Frequently asked questions
How does AI lead scoring work?
When someone books a call, an agent reads their booking answers, researches them and their company, and attaches a probability of converting plus a short brief to the calendar event, before you take the meeting.
What should a lead score measure?
Fit with your ideal customer, not how interesting the person would be as a guest. One client's AI was ranking podcast fit instead of buyer fit, which filled his calendar with good guests who were unlikely to buy.
Can I score Calendly bookings automatically?
Yes, if your scheduling tool exposes an API. The scorer reads each new booking and writes the score onto the event. For one client we planned to connect it through his scheduling tool's API.
Why average scores weekly?
A single score describes one person. The weekly average describes your list. If it rises, your targeting is improving; if it falls, fix the list or the copy before adding volume.