See festive peaks coming. Know why the forecast moved.
Sanvaya brings statistical and machine-learning forecasting, promotions, India's festival calendar and stockout analysis into one planning workspace, built for mid-sized manufacturers and brands that have outgrown spreadsheets.
Every step of demand planning, in the order your team runs it
Sanvaya means coordination. Each step hands its output to the next, so the number your planners agree on traces back to the history and drivers behind it.
- 01
Demand patterns
Sales history, cleaned actuals, stockouts, excess orders, holiday and promotion lift.
- 02
Statistical forecast
Segmentation, outlier capping and several models competing on a holdout.
- 03
Machine learning
LightGBM with promotions, prices, weather and festivals as drivers.
- 04
Disaggregation
Down to product, customer, location and channel, respecting launch and end dates.
- 05
Consensus
Statistical and ML side by side, sales overrides, one agreed number.
- 06
Post-game analysis
Accuracy, bias and value added by each step, by lag and cycle.
Find what distorted your history before you forecast
Lost sales and one-off bulk orders quietly shape every forecast built on them. Sanvaya checks every series against its normal level and shows you where to correct.
- Stockouts with the units that were lost
- Excess orders no holiday or promotion explains
- Holiday and promotion lift, present or not, per series
- Corrected demand, ready to feed the forecast
An ML forecast your planners can question
Every machine-learning forecast sits next to the statistical one, with the units each driver adds or removes. When the number jumps, you can see whether it was a promotion, a festival or the weather.
- Sell-in and sell-out promotions with original and discounted prices
- Festivals and holidays by country, region, location or customer
- Weather by region and day
- Same split for both: ML uses the statistical profiles, so they compare cleanly
One agreed number, and proof of what each step added
Sales teams enter overrides, planners set the consensus, and every change is logged. Forecast value add then measures whether each step, from a naive forecast to the consensus, actually improved accuracy.
- Roles for admins, planners, sales and viewers
- Accuracy and bias by lag, reference date and level
- Scenarios to test what-ifs without touching the plan
The whole planning cycle in just over two minutes
From sales history and stockout analysis to ML drivers, consensus and accuracy tracking. Narrated, using demo data.
Ask your plan a question in plain English
Two assistants answer from your own planning data, with the table behind every answer. They only read data, and they work without sending anything outside your servers. Connecting Claude adds free-form reasoning on top.
Planning Assistant
Stockouts, excess orders, holiday and promotion effects, and the largest statistical forecast errors.
Forecast Assistant
Coming forecast, statistical vs ML, growth against last year, and what drives the ML forecast.
See Sanvaya on your own data before you decide anything
We are taking on a small number of companies for a free benchmark. You get an item-by-item comparison of your current forecast against Sanvaya's, and you keep the results whatever you decide.
Share your sales history
Two to three years of sales by product, customer, location and channel, anonymised if you prefer. A spreadsheet export is enough.
We run the full cycle
Exploratory analysis, statistical and ML forecasts, with your promotions and festival calendar where you have them.
Compare, item by item
Accuracy and bias against your current forecast, the stockouts we found in your history, and where ML helps.
Controls a planning team needs from day one
Roles
Admin, Planner, Sales and Viewer, each limited to what they may change.
Audit trail
Every sign-in, change, refusal and assistant question, with who and when.
Working versions
Each user plans on their own version or scenario without disturbing others.
AI on your terms
Assistants run offline by default. With Claude connected, only summarised results are sent.
Frequently asked
Do we need to replace our ERP?
No. Sanvaya plans from your sales history and master data, which most teams can export from their ERP or reporting system. Plans go back out the same way.
What data do we need for the benchmark?
Sales quantities by week or month for each product, customer, location and channel, ideally two to three years. Promotions, holidays and prices improve the machine-learning forecast but are optional.
Weekly or monthly planning?
Both. You choose the time grain and the level you forecast at, for example product by region, and Sanvaya splits the plan down to the lowest level.
Does the AI see our data?
The assistants work without any external AI. If you choose to connect Claude, it receives only the summarised results of each question, never your raw tables.
We already use statistical forecasting. Why add machine learning?
Statistical models read the history alone. Machine learning also learns from promotions, prices, festivals and weather, and Sanvaya shows its forecast and its drivers next to the statistical one so planners can choose per item.
Request your free forecast benchmark
Tell us a little about your planning today and we will get back to you by email. You can also write to us directly.
hello@sanvaya.in