A decision-intelligence panel for B2B construction-materials trading — fusing a CRM, VoIP call center, segmentation and revenue intelligence into one action-first, explainable, RTL workspace.
SivanLand is the internal panel for a company that buys and sells construction materials and steel in a fast, relationship-driven market, running on a “sell first, source later” model — deals are agreed on the phone, payment is collected, and goods are procured afterwards. The panel fuses a daily to-do queue, a VoIP call center, customer & supplier records, a sales funnel, revenue forecasting, customer segmentation and cancellation analytics into one coherent workspace: not a dashboard you stare at, but one that hands you the next decision, simple enough for a non-technical operator to use on day one.
The company didn't need “more software” — it already had an ERP and plenty of data. The pain was that the data and the daily work had drifted apart:
The reframe: the software asks people to feed it without ever helping them back, so they stop feeding it — and the intelligence dies with the data. Every time the system asks for something, it must give something back immediately.
I talked through daily routines with sales and procurement staff, shadowed how an operator handles a live deal from first call to payment, reviewed the raw events the ERP already produces, and studied why operators abandon mainstream CRMs. Each insight became a design rule:
The panel is grouped into three plain-language areas — Daily, Customer Segmentation, Records & Reports — deliberately stripped of jargon. An early version used words like “Intelligence Hub” and “Studio”; against non-technical readers those names caused hesitation, so “Segment Studio” became “Build Segment”, “Customer 360” became “Customers”, and “Revenue Intelligence” became “Revenue & Forecast”. A menu label should say what a screen is, not how clever it is.
The daily flow — Today. The operator starts here: a queue pre-sorted into Critical / High / Medium, each card with a one-line reason to act, snooze or override.
The VoIP call flow — the centerpiece. The operator clicks Call; the system does the rest. During the call the voice is recorded, and the moment it ends, the system transcribes the conversation, writes a short summary, and auto-fills the customer's fields (phone, interest, best time to call). The operator only taps an outcome — Interested, Needs follow-up, Bought, Cancel order — and the whole interaction drops into the customer's timeline as a recorded, summarised event. A job that used to be a five-minute form becomes: click, talk, tap.
Cancellation flow. A customer can only cancel before payment; after payment the option is disabled with a plain explanation, so the rule is obvious rather than a dead end.
From segment to action. Segmentation isn't a dead report — picking rules recomputes live metrics, and a segment can be sent straight to Today as an action card for the whole group.


This strip appears under every AI suggestion in the product: a plain-language reason, a confidence bar and percentage, and a one-tap override the operator can use to say “this is wrong” and pick why. Every override teaches the system — the human is never overruled by a black box, and the black box gets less black over time.
Leading with actions and making every suggestion explainable changed the product from “a CRM you maintain” into “an assistant you trust.” Folding call capture, records, segmentation and cancellations into one flow removed the gaps where work used to leak.
Run moderated usability sessions on the call → outcome → timeline flow, test whether operators read confidence scores the way they're intended, and wire the VoIP call to a real telephony layer and speech-to-text/summarisation model.