FMCG
Additional working capital identified in the first quarter
Reduction in Days Sales Outstanding
About
A mid-sized FMCG manufacturer running tight working capital against 60–90 day distributor payment terms. We replaced a weekly manual cash flow forecast with a machine-learning model that predicts payment timing account by account, giving treasury three weeks of advance warning instead of zero.
Industry
FMCG
Company size
500 – 1,000 employees
Founded
2003
The Company
A tight working capital model with a forecasting problem at its centre
The business manufactures across two states and distributes through 15 regional markets. The business procures raw materials upfront against distributor and modern-trade payment terms that frequently extend to 60 or 90 days, so accurate cash flow forecasting is a core operational requirement, not a nice-to-have.
Leadership needed advance notice of when inflows would actually land, where shortfalls were forming, and what treasury needed to do before a shortfall became a crisis.
The challenge
A manual process that was consistently too optimistic
Two analysts spent three days a week pulling receivables data from the ERP and building a 4-week cash flow forecast in Excel. Payments the forecast showed landing in week three routinely slipped to week five or six, either because distributors were themselves cash-constrained or because invoice disputes created delays the model had no way to anticipate.
These surprises triggered reactive short-term borrowing at unfavourable rates and strained supplier relationships when payables had to be extended without warning.
The Solution
A rolling cash flow model built on payment behaviour, not payment terms
We built a TensorFlow model on Google Cloud, with BigQuery as the warehouse and Vertex AI managing training and inference, drawing on 24 months of actual payment timing by distributor, current receivables aging by dispute status, a proprietary credit-risk score, and upcoming order and shipment data. A rolling 90-day forecast regenerates daily, with automated alerts when projected balances fall below threshold at any point in the horizon.
Alongside the forecast we deployed a collections prioritisation tool that ranks outstanding receivables by expected delay risk, so the collections team spends its time on the accounts most likely to slip rather than following a uniform aging schedule.
The Results
Rs. 3.2 Cr recovered and a treasury function transformed
Rs. 3.2 Cr in additional working capital was identified in the first three months, through faster collections on high-risk accounts and reduced reactive borrowing. Days Sales Outstanding fell 19% as risk-weighted collections replaced uniform aging-based outreach.
Rolling 30-day forecast accuracy reached 94% within two months, against an estimated sub-60% accuracy for the previous manual process. Two supplier payment disputes were avoided in the first quarter because the early-warning system flagged a projected shortfall three weeks out, in time to have the conversation proactively.
KEEP READING
Explore more studies on AI, analytics, and enterprise intelligence.
