Work That Held Up in Production.
Three engagements from the team's earlier consulting work — the operating history behind the scans. Names withheld and identifying details changed to protect confidentiality; deduction-scan case studies will publish here as current engagements complete.
Rebuilding the Delivery Math for a Rural Distribution Network
lower overhead
- Delivery cycles improved network-wide
- Held up across multiple high-complexity markets
- Recognized at CEO level
The problem
A large CPG company ran a rural network of thousands of micro-entrepreneurs, field distributors carrying product the last mile into villages and small towns. The routes had grown by habit, not math. Territories followed relationships and precedent. The map looked complete; the delivery cycles said otherwise.
What we did
- Audited the existing distribution logic before proposing any changes
- Rebuilt routing with nearest-neighbor optimization, same distributors, same infrastructure, new math
- Replaced hand-drawn territories with a framework the team could update themselves
- Built a separate multi-distributor model for a South India territory conflict with overlapping coverage
- Wrote the documentation for the operations team, not for consultants
No new platform. The network ran on the same infrastructure it always had, finally supported by the right math.
MMM and SKU Optimization Across Global CPG Portfolios
- Budget moved from underperforming channels
- SKU cannibalization surfaced from aggregate noise
- Budget talks shifted from precedent to models
The problem
Across large portfolios in oral care, confectionery, and alcoholic beverages, the same structural gap: budgets had accumulated around history. Nobody could say which channels drove sales, or which SKUs earned their shelf space. When budgets tightened, the only benchmark was last year's plan.
What we did
- Audited each client's data-capture workflows first, one version of the truth before any model code
- Built Marketing Mix Models in Python and SPSS linking media spend to sales
- Layered SKU-level recommendations: which products drove category growth, which cannibalized it
- Worked brand-level across flagship chocolate and biscuit lines in India
- Delivered to the people setting the budget, no statistician required in the room
3–5% promotional uplift post-reallocation. On seven-figure budgets, that compounds fast.
From Legacy Reporting to Decisions the Team Can Make Today
- Reporting cycles: weeks to near real time
- Teams answer their own questions now
- Nothing existing was torn out
The problem
Manufacturing and healthcare organizations with enormous commercial datasets, and almost none of it usable in time. Reports sat in legacy BI stacks and siloed databases, taking two to three weeks to generate. By the time an answer arrived, the question had moved on.
What we did
- Migrated legacy BI to Azure and GCP, keeping the core business logic intact
- Preserved what worked, the goal was scale, not replacement
- Deployed natural-language query tools so business users could skip the report queue
- Ran change management alongside the build: teams trained as the system went up
We modernized what didn't work and kept everything that did. The team's knowledge of their own data stayed intact.
Client names are withheld throughout and identifying details changed. Evaluating a similar problem? The best conversation happens on a call.