Case studies / Applied AI
Applied AI · Consumer services
Real-time lead scoring that puts every lead in front of the right rep
A single credit-filter heuristic was the only lead qualifier — good leads buried alongside bad ones, scarce sales capacity wasted on low-probability leads. We scored every lead in real time, before it ever reached a rep.
Outcomes
Impact in production.
Scored in real time, before each lead reaches a sales rep.
Delivered to production by a three-person team, then expanded company-wide.
Lift over the previous credit-filter baseline in the pilot division, which led to a full company-wide rollout.
The conversion lift is measured in the pilot division; the rollout then extended across all divisions.
The challenge
One heuristic, doing the job of a decision
A single credit-filter heuristic was the only lead qualifier in use — good leads buried alongside bad ones, and scarce sales capacity spent chasing low-probability leads instead of the ones likely to convert.
Before
- One heuristic as the only qualifier
- Good and bad leads mixed together
- Sales capacity spent on low-probability leads
- No visibility into model drift or accuracy over time
After
- Every lead scored in real time before reaching a rep
- Sales queue prioritised by predicted conversion
- Human validation gate on every score
- Drift monitoring built in from the start
How we approached it
Build a learning system, not a model.
A learning system, not a one-off model
A one-off model decays and fails in production. We scoped retraining, a human validation gate, and drift monitoring as first-class requirements from the start — and built knowledge transfer in from day one, so the client could own the lifecycle.
Two clean paths
A scheduled batch training path, with a mandatory human validation gate before any model goes live; and a real-time scoring path from intake through enrichment to a live score — with the two kept cleanly separated.
Built for ownership and safe rollout
We chose low-barrier, cloud-native tooling so the client’s team can retrain without deep ML expertise, delivered explainability and lift dashboards for the business, and rolled new models out through structured A/B testing rather than direct replacement.
The cloud foundation, built to de-risk the start
We stood up the entire Azure foundation — subscriptions, networking, security, service provisioning, and the ML workspace — in our own environment first, proved the platform end to end, then ported it into the client’s cloud. The client didn’t need their estate ready, or to grant deep access on day one, for the build to move.
Under the hood
The production stack.
The human validation gate here wasn't a formality.
It's the same discipline behind why AI-assisted engineering needs review that can actually keep pace.
Next step
Sales capacity going to leads that were never going to convert?
We build scoring systems that put every lead in front of the right rep — with a human still in the loop where it counts.