“We built this platform so our customers don’t have to rely on guesswork. Once we had the scale, applying it to our own pipeline was the obvious move, and the results proved it.”
Dror Katz, VP Data & Analytics @ Pecan AI
Meet Pecan (yep, that’s us!)
Pecan AI is a predictive analytics platform that helps teams turn business questions into predictions without requiring data scientists. Companies across industries use Pecan to predict churn, score leads, forecast demand, and identify upsell opportunities.
So when Pecan’s own sales team found themselves struggling with the same problem their customers face every day – too many leads, not enough signal – the answer was obvious. They built a model on Pecan.
The challenge
Pecan’s sales team was handling a growing volume of inbound leads, but not all leads are created equal. With 2,300+ qualified leads in the pipeline, the team needed a way to prioritize. Their existing system used marketing grades based on firmographic data (company size, industry, job title) assigned partly through manual analysis.
It worked, just not well enough.
- Grading was partly manual, introducing inconsistency
- No behavioral or intent data informed who was actually ready to buy
- Reps had no reliable way to distinguish high-potential leads from the rest
Conversion rates from initial contact to Closed Won were lower than they should have been. The result: sales time was being spent on leads that went nowhere, good leads were getting lost in the noise, and the team had no reliable way to find them.
The solution
Why would Pecan use anything else?
The answer was self-evident. Pecan’s own platform exists to solve exactly this problem, turning raw CRM and behavioral data into actionable predictions, without requiring a data science team to build and maintain a custom model.
Building internally also created something valuable beyond the model itself: a real-world proof point. If the platform could deliver meaningful results for Pecan’s own team, a lean, fast-moving sales org with a real pipeline, it could do the same for any customer.
A predictive question sales could act on
The team built a classification model with a clear goal: score every new lead within one day of creation, predicting the likelihood they’d reach a second call or Closed Won. This gave reps a daily, Salesforce-native signal to guide their outreach, not a static grade assigned at intake, but a live prediction refreshed every 24 hours.
What went into the model
The model was trained on historical qualified leads and drew from three data sources: Salesforce, HubSpot, and PostHog site event data. Where previous grades relied only on firmographic and persona information, the Pecan model added a layer the marketing grade had never included: behavioral and intent signals.
Firmographic: Company size, annual revenue, industry, country, year founded
Behavioral & intent: Demo booking confirmation, total site visits, average daily pageviews, pages visited, time from first site visit to lead creation
Persona: Job title, function, seniority
One of the strongest predictive signals was: whether the lead had booked a demo (16.5% feature importance), though the model combines this with a wide range of behavioral and firmographic signals. This is not something the previous manual grading system captured.
How fast it came together
12 days, start to finish – the team went from zero to live in that time. The biggest technical challenge was matching PostHog website event data to CRM contacts accurately, ensuring no duplicates or misattribution before the data entered the model. This was handled through Pecan’s data preparation capabilities, which automatically clean, join, and structure messy data, combined with validation to make sure everything matched correctly and made sense.
How it was used
Predictive scores are now embedded in Pecan’s daily sales and marketing operations.
Every lead created in Salesforce is scored based on its predicted probability of reaching a second call. Leads are then grouped into model grades (A, B, C) based on that score. The grade appears directly on the contact record and is used across both sales and marketing.
Sales prioritization: Reps use the model grade to triage their pipeline each morning. Grade A leads get first attention. The goal is not to ignore B and C leads, but to ensure the highest-potential opportunities are never buried under volume.
LinkedIn campaign targeting: Marketing uses model grades to build segmented LinkedIn ad audiences, concentrating spend on the profiles most likely to convert. The same score that guides sales outreach now guides paid media decisions.Live monitoring dashboard: A dedicated dashboard tracks model performance, grade distribution, and conversion outcomes over time, giving leadership visibility into how the model is performing and where the pipeline is heading, while also surfacing shifts in lead quality that help the team re-evaluate inbound and outbound activities.
The impact
3x conversion rate for top-graded leads
Model Grade A leads reached a second call 32.3% of the time, compared to 15.1% for leads graded A under the previous marketing grading system. Closed Won rates followed the same pattern: 8.3% for Model Grade A versus 2.9% for marketing grade A.
When both signals agreed, Model Grade A and marketing grade A on the same lead, conversion rates climbed further still: 41.9% to second call and 12.9% to Closed Won.
4.3x higher precision
Among the top-scoring leads, the model achieved 44.8% precision, meaning nearly half of all leads flagged as high-potential actually converted. The baseline was 10.5%. That gap is what turns a prioritization tool into a genuine competitive advantage.
Daily signal, not quarterly grade
What was once a quarterly, manually-assigned grade is now a daily, automated prediction. Reps no longer have to guess which leads to prioritize first. The model does the work every morning, in the CRM they already use.
Two teams, one source of truth
Sales and marketing now operate from the same signal. The model grade that drives outreach prioritization is the same one shaping campaign targeting. Alignment that previously required meetings and manual coordination is now built into the workflow.
Key takeaway
Pecan’s sales team replaced guesswork with a model, and the model won. The team had the same problem thousands of B2B teams have: too many leads, not enough signal. Pecan solved it in 12 days using its own platform – and the model outperformed the previous system by 3x.
The lesson isn’t just that predictive analytics works. It’s that the barrier to getting there is lower than most teams think. Pecan built this model in less time than most teams spend debating whether to start.
