How our Artificial Intelligence process works
From the first call to a live, scoring model inside your CRM, here is every step we take and roughly how long each one lasts.
The six steps
Most engagements go live within three weeks. Enterprise projects with complex integrations may take four to five.
Discovery call (day 1)
A 30-minute video call where we learn about your sales cycle, CRM setup and current lead sources. We ask specific questions: how many leads come in per month, what percentage close, and which data fields you already track. No slides, no pitch deck. We just listen and take notes.
Data audit (days 2–4)
You grant us read-only access to your CRM. We export a sample of historical deals, check for missing fields, inconsistent formats and duplicate records. Then we produce a short report: what is usable, what needs cleaning, and how large the training set will be. If fewer than 200 closed-won deals exist, we recommend enrichment sources to fill the gap.
Model training (days 5–10)
Our data engineers build a gradient-boosted scoring model using your CRM history. We test multiple feature sets: industry, company size, source channel, engagement signals, time-to-first-response. The model that produces the highest area-under-curve on a held-out test set becomes the candidate. We share accuracy metrics with you in a plain-language summary, not a 40-page PDF.
Validation workshop (day 11)
A one-hour session with your sales manager. We show the model's predictions on 50 recent leads and ask: "Does this ranking match your gut feeling?" If the model disagrees with experienced reps on more than 15% of leads, we retune. This step catches edge cases that data alone misses, like leads from a partner channel that always convert regardless of score.
CRM integration (days 12–15)
We deploy the model as a lightweight API that writes a score field directly into your CRM. New leads receive a score within seconds of creation. We also set up a dashboard view so managers can sort pipelines by predicted close probability. No new software to install; everything runs in the cloud and connects through your CRM's native API.
Go-live and first retrain (day 16 onward)
The model is live. We monitor scoring accuracy weekly for the first month. At the end of month one, we retrain on fresh conversion data and compare performance. You receive a one-page report showing how many high-scored leads actually closed. From here, retraining continues on the schedule your plan includes.
What you need to provide
We do the heavy lifting, but a few things depend on you.
Read-only CRM access is the big one. We need at least six months of deal history, ideally twelve. The more closed-lost records you have, the better the model learns what a bad lead looks like.
A single point of contact on your side who can answer questions about your sales process. This person joins the discovery call and the validation workshop. Total time commitment: about two hours across the whole project.
If you use a custom CRM or an uncommon tool, let us know during the discovery call. We have built connectors for Zoho, Close, Freshsales and several bespoke systems. If yours is new to us, we will scope the integration separately.
After go-live: ongoing support
Deploying the model is not the finish line. Here is what happens month after month.
Monthly performance reviews
We compare predicted scores against actual outcomes. If the model's accuracy drifts below the threshold we agreed during validation, we retrain immediately rather than waiting for the scheduled cycle.
Feature updates
When your CRM adds a new field or you start tracking a new engagement signal (webinar attendance, for example), we can incorporate it into the model at the next retrain. No extra charge on Growth and Enterprise plans.
Quarterly strategy calls
Growth and Enterprise clients meet with our lead data scientist each quarter. We review conversion trends, suggest pipeline optimisations and plan any model expansions. These calls typically last 45 minutes and are recorded so your team can share them internally.