Supply Chain Assistant deployed across 3,200 SKUs and 12 markets for one of India's largest FMCG operators — with demand planners in the loop, not replaced.
Client
Hindustan Unilever
Sector
Retail & Consumer Goods
Duration
14 months
Team
11 (AI engineers, supply-chain SMEs, SAP consultants, product engineers)
Hindustan Unilever is India's largest FMCG company by revenue and one of the largest globally by volume, with a portfolio spanning home care, beauty, personal care, foods, and refreshment. The company's supply chain moves product across a country whose retail estate ranges from urban hypermarkets to single-shopkeeper kirana stores. Demand-planning at this scale is a discipline where every percentage point of forecast accuracy is a nine-figure-rupee number.
Hindustan Unilever's demand planners manage SKU proliferation, promotional intensity, and monsoon-driven demand shifts across a country where the retail estate spans hypermarkets and single-shopkeeper kirana stores in equal measure. The existing planning stack was accurate on the slow-moving core and unreliable on the long tail — which was where growth was.
01
The assistant doesn't publish forecasts. It proposes them. Planners see the drivers, edit the number, and the model learns from the edit. The assistant is a colleague, not a replacement — a positioning decision that landed with the planner community before any pixel of UI was designed.
02
Every forecast comes with the drivers that moved it — promotions, weather, macro, factory constraints. Planners see why the number changed and can push back on the reasoning without leaving the tool.
03
When leadership asks 'why is Mumbai soft?' the assistant answers in a paragraph, with the number and the drivers, in-context — not as an ad-hoc analyst request that takes three days.
04
Started with two markets and three product families. Expanded market-by-market as planners requested onboarding — not as a top-down decree.
01
Months 1–2
Six weeks of shadowing planners across two pilot markets — Mumbai and Chennai. Attended two S&OP cycles as observers. Mapped the actual planner workflow (as opposed to the documented one). Positioned the assistant as a colleague, not a replacement, before any UI decisions were made.
02
Months 3–6
Two markets, three product families, planner cockpit and Slack surface in production. Retrained forecasting models on innovation SKUs. First measured MAPE reduction and first planner-community feedback loop.
03
Months 7–10
Rollout by market on request — not by decree. Four more markets came online in this window. Feature roadmap driven by the planner community; leadership asked for a between-review chat surface, which landed as the Teams assistant.
04
Months 11–14
All 3,200 SKUs indexed. Audit-trail hardening for planner overrides — every edit carries the reason, the driver, and the reviewer signature. Passed HUL's internal audit review at the end of month 14.
Supply Chain Assistant plumbed into HUL's SAP IBP and Kinaxis stack, with a Slack/Teams assistant surface for leadership and a planner cockpit for daily work. Forecasts, exceptions, and ad-hoc question answering across 3,200 SKUs and 12 markets.
01
The assistant reads from and writes to IBP. The forecast that gets shipped to the plants is the IBP number. The assistant's role is to propose, explain, and learn — not to replace the source of record.
02
Leadership already lives in Slack and Teams; making them switch tools would have failed. The assistant answers questions where the questions get asked, in-context, with the same authoritative number IBP would return.
03
Explanations are generated as part of the forecast, not as a post-hoc rationalisation. Every driver comes with a magnitude and a confidence; planners can push back on the reasoning without leaving the tool.
04
Model training, retraining, and eval run on Databricks. Model artefacts are versioned; each model release is gated by an eval against a golden set of forecasts curated by the planner community.
The rollout strategy was the reason the programme worked. We started with two markets and three product families — Mumbai, Chennai, and the personal-care core. Planners in those markets were the ones who brought in Bangalore and Kolkata; the four subsequent markets came online because the community pulled the rollout, not because leadership pushed it. Audit-trail hardening was done in parallel with rollout, not after, so by the time the assistant was in every market, it had already passed internal audit review.
38%
Reduction in forecast MAPE
Weighted average across all SKUs and markets, measured against the pre-change baseline over four quarters. Innovation SKUs and promotional weeks saw the largest gains.
3,200
SKUs under continuous forecast
Including the innovation SKUs the legacy stack couldn't handle — with planner-in-the-loop override discipline preserved and the audit trail captured on every edit.
12
Markets covered
Rolled out market-by-market on planner request. Adoption drove the schedule; the schedule didn't drive adoption. Two markets are onboarding this quarter.
The counter-intuitive lesson: positioning mattered more than the model. When we called the assistant 'planner-in-the-loop' from month one, the planner community treated it as a tool. When earlier automation projects at HUL had positioned themselves as 'replacing' the planner, they were resisted at every stage. The technical work would have been the same either way; the framing was the difference between adoption and non-adoption.
Two additional markets are onboarding this quarter. Roadmap includes an expansion into promotion-planning assistance, a scenario-comparison surface for the sales-and-operations executive review, and a feed into the S&OP scenario workflow. The audit trail is being extended into a firm-wide 'decision provenance' pattern that other planning functions are adopting.
“The bit I didn't expect was how much the between-review conversation changed. Leadership stopped asking us to pull a number and started asking us what to do about the number.”
Book a 30-minute discovery call. A senior practitioner from the same practice that shipped this engagement will scope yours.
or email contact@headify.in
