Sales CRM and BI
AW Cycles distributes bikes and components across Italy: 7 sales agents, 594 customers, 293 products, 71,000 sales lines over three years, 48 offers and 243 contact notes. The screens are the ones every sales team knows. What changes is that they understand what people write.
Loading…
Under the hood
- Every day of the calendar, customer, product, offer, note and agent has a fingerprint of its meaning, computed once: 2,639 fingerprints, about 2 MB. Sales lines are a 1 MB file read directly in the page; totals over 71,000 lines take a few milliseconds.
- Periods: each day has a short description of its commercial context (“Friday before Black Friday, Christmas shopping starts”). A request is compared with all 1,454 days and the days that stand out for that request form the period, in any language. Years, month names and “weekend” written in the request are applied exactly.
- Notes: similar notes are found whoever wrote them; themes are discovered by grouping notes that mean the same thing.
- Offers: suggestions mix closeness in meaning (same discipline, other category) with what customers bought together; a lost offer’s notes tell the reason (price, delivery times, supplier) and the revision follows it. Prices come from the catalogue.
- Semantic treemap: each item goes to the single label that fits it best, if above the threshold; the rest goes to Others.
- The data is synthetic (an adaptation of a public sample database): notes were generated from templates, so similar notes are very close to each other. Values stay in English; the interface and the requests work in any language.