Build your own demand forecasting engine
Agents combine your sales history with public retail, price and sentiment data, then build and test a forecast for each category. No new data infrastructure or data science team required.
What runs behind the example
The planner is the part you see. Behind it is a managed data pipeline that agents built and keep running.
Sources
Public data, taken from the agencies that publish it.
- Census retail sales
- BLS consumer prices
- Consumer sentiment
- Your sales data
Ingest and catalog
Agents load each series on a schedule and document every table and column.
- 21 data series
- Documented tables
Model and test
Agents write the forecast models and test each method on past seasons before using it.
- 3 methods tested
- 16 past seasons
Live application
The planner reads from the tested model, and every number traces back to its source.
- Scenario planner
- Audit trail
Refreshed daily. New data runs through every stage, so the planner stays current without anyone rebuilding it.
Using your own data
The example runs on public data. Here is what changes when you bring your own.
- Sales
- In this exampleCensus retail sales, monthly
- With your dataYour sales by product, store or channel
- Prices
- In this exampleBLS consumer price indexes
- With your dataYour prices, promotions and margins
- Demand signals
- In this exampleConsumer sentiment
- With your dataYour web traffic, orders and pipeline
- Decisions
- In this exampleA holiday forecast by category
- With your dataOrders, inventory and markdown plans
- Review
- In this exampleOne person approves the planner
- With your dataYour planners and finance team sign off
Your data stays private to your organization. Nothing is public unless you publish it.
Why you can trust the numbers
Every forecast misses by something. We tested this one on past seasons, so you can see by how much.
1.11%
Average forecast miss
Tested on 16 past holiday seasons, against total holiday retail sales.
Methods tested, by average miss
- Summer sales growthUsed1.11%
- Summer growth plus consumer sentiment1.15%
- Year-to-date growth1.40%
Lower is better. Single categories are harder to call: their average miss runs from 1.8% to 5.2%.
- Tested before it is used
- Every method is run against past seasons first, so you see its track record, not a promise.
- Every number is traceable
- Each figure links to its source and the calculation behind it. Gaps in the data are recorded.
- A person signs off
- Nothing is shared until someone on your team has reviewed and approved it.
Build a forecasting engine for your business
Start with your sales data and the question you need answered.
Connect your data
Upload a sales export or connect your warehouse. The public data is already there.
Ask your question
Describe the forecast you need. Agents build a first version you can use right away.
Review and share
Check the numbers, comment and approve the planner before your team relies on it.