Distribution
Which customers are buying less?
A customer decline usually hides in a monthly sales report until it is already a pattern. FactSmith compares recent order volume and value against each customer's own history, flags the ones trending down, and lets you ask why -- by product, territory or rep -- in the same conversation, on your own warehouse.
Why this matters
By the time a declining account shows up in a standard report, the moment to act on it has often already passed. Getting there sooner today usually means someone pulling sales history and reconciling it by hand.
What data is needed
Sales or order history by customer over time, at minimum -- ideally with product, territory and rep attached so the answer can be broken down further.
The evaluation copy runs against PostgreSQL. SQL Server is on the adapter path for a customer warehouse; that path is not what the package proves end to end today.
How the analysis works
Each customer's recent activity is compared against their own trailing baseline, not a company-wide average, so a naturally smaller account is not flagged next to a genuinely churning one. Results rank by rand value at risk as well as percentage, so a big account does not hide behind several small ones.
What can go wrong
Seasonal customers can look like decline if compared to the wrong period -- the comparison needs to use a comparable prior period, not just the immediately preceding one. A customer correctly reassigned between reps can also look like a drop if the history is not tracked at the customer level.
Example
A distributor with a few hundred active accounts asks which customers ordered noticeably less this quarter than their own trailing average. Ask returns the list ranked by rand value at risk, with each account's last order date.
How FactSmith approaches it
Ask is what is available now. Easy Data -- turning messy sources into that dimensional foundation -- is in early development.
FactSmith calls this the Fact Layer: a data integration (ETL) step, a knowledge base, and a business ontology, working together underneath the answer. On screen it stays simple -- ask a question, get a trusted answer -- with the layer-by-layer detail available to anyone who wants to see how the answer was built.
Permissions are enforced beneath the AI, not by it -- see Trust and security.
Related questions
Slow-moving inventory · Margin leakage · For distributors · How it works