A B2B services group with distributed sales operations (name withheld under NDA)
NovuCRM Intelligence Suite — an internal delivery codename, not a commercial product name.
The problem
The client had bought a lead-scoring product that produced a number with no explanation. Sales teams did not trust it, so they ignored it, so it produced no value while continuing to cost money — a common and entirely rational outcome.
Forecasting was spreadsheet-based and each region did it differently, which made the consolidated forecast a negotiation rather than a measurement.
Constraints we had to design within
- Sales adoption was the binding constraint. A technically superior system nobody uses is worth nothing.
- Multiple regions with genuinely different sales motions.
- Existing marketing automation that had to remain in place.
Approach
Score with reasons, not just numbers
The model surfaces the factors driving each score rather than only the score. Salespeople who can see why a lead ranked highly will engage with the ranking; a bare number invites dismissal. Adoption was the real problem, and explainability was the solution to it.
Model prioritises, humans decide
The model orders the queue. It does not disqualify leads or make decisions on its own. This is the same principle we apply in compliance work for a different reason: keep the unexplainable artefact off the path where a decision has to be justified.
One forecast definition, regional inputs
Forecasting logic was unified so the consolidated number means one thing, while allowing regions to reflect genuinely different sales motions in their inputs. The disagreements that remain are now about inputs rather than about definitions, which is a much more productive argument.
Engineering notes
Explainability chosen for adoption, not compliance
The compliance work in this practice keeps models off the decision path because regulators require justification. Here the same architecture was chosen for an entirely different reason: adoption.
A salesperson shown a bare number will ignore it, and a scoring system that is ignored produces no value while continuing to cost money — which is what the client had already bought once. Surfacing the factors behind a score converts it from an assertion into an argument, and an argument can be engaged with.
It is worth noting that the same design answers two unrelated problems. That is usually a sign the constraint is real rather than local.
Unifying the forecast definition without flattening the regions
Regional sales motions differ genuinely, and forcing one process onto all of them produces compliance in the tool and reality in a spreadsheet somewhere else.
The definition of the forecast was unified so the consolidated number means one thing, while inputs remain regional. The arguments that survive are now about inputs rather than about definitions, which is a considerably more productive disagreement to be having.
Keeping the existing marketing automation in place
Replacing adjacent systems that work is a common way to turn a contained project into an unbounded one. The existing automation stayed, with a defined ownership boundary and a single direction of data authority.
Integrating with a system you did not build and cannot change is more constrained engineering than replacing it, and almost always the cheaper decision overall.
Outcome
Lead scoring is used rather than ignored, because it explains itself. Forecast consolidation is a calculation rather than a negotiation.
The model stays in an advisory position where an incorrect output is recoverable.
Figures above are drawn from delivery records held under NDA and are pending independent confirmation. Where a figure cannot be evidenced it will be removed rather than qualified.