Sign in

Blog · How-to guides

Sales intelligence from the data you already hold: five sources, one roll-up

What sales intelligence means when it is built from finance, CRM, activity, product and benchmark data a company already has: the five sources, how they join on the client, the roll-up that reconciles them, and the questions a salesperson can then ask that no dashboard anticipated.

The short answerSales intelligence is the joined view of a salesperson's clients across finance, CRM, activity, product and benchmark data, rolled up and reconciled, with an assistant that answers the next question from the same figures. The five sources join on the client identifier; the roll-up is asserted against reported revenue; and every screen and every answer calls the same functions, so a chart and a sentence cannot disagree.

Sales intelligence is a phrase that means many things in vendor decks and one thing in practice: the salesperson can see their clients whole and ask questions about them. The data to do that exists in every company, in five systems that have never been joined. This guide sets out the five sources, the join, the roll-up, and what changes when the salesperson can ask.

The five sources

Source What it holds Joins on
Finance Revenue by client, product and period Client identifier
CRM Client ownership, contacts, pipeline Client identifier
Activity Calls, meetings, emails with dates Client identifier
Products The product hierarchy and what each client holds Product code
Benchmarks The client's estimated wallet by product Client identifier

Every join is on the client identifier, which is why the identifier mapping between systems is the first and most valuable piece of work.

The roll-up

Once joined, the rows roll up through one hierarchy:

desk = Σ regions = Σ salespeople = Σ clients = Σ products

The identity is asserted against the revenue finance reports for the period. Every figure downstream inherits that reconciliation. A client under two salespeople, a product coded differently in finance and in the CRM, or a duplicated export each break the identity at one level, and the report says which.

What the salesperson sees

Their own book: clients ranked by revenue, penetration by product against wallet, opportunities valued, clients at risk from falling activity or revenue, and coverage against plan. Every tile is a function on the reconciled rows.

What the salesperson can ask

The assistant reads the question, picks the function, runs it on the salesperson's rows, and explains the result with the function named.

  • "Find my biggest opportunities." The opportunity detector, on their clients, ranked by the gap between held and wallet.
  • "Which clients traded rates but not credit this quarter." A filter on the client-by-product view.
  • "Who have I not spoken to in ninety days." The coverage function, with the contact log.
  • "Why is my Asia North revenue down." The revenue calculator by client, ranked by change.

None of these were built as a tab. All of them use functions that already existed, because the functions are the product and the assistant is the interface.

A worked example

A salesperson covering eighteen clients in Japan, FX products, quarter to date.

Client Revenue Wallet, est. Penetration Last contact Signal
2291 £2.1m £2.6m 81% 12 days
0876 £0.6m £2.2m 27% 30 days Opportunity: forwards
1140 £0.4m £0.5m 80% 96 days At risk: activity
1187 £0.3m £1.9m 16% 8 days Opportunity: options

The assistant's answer to "find my biggest opportunities" is client 0876, strong options flow and 27 percent penetration against a £2.2m wallet, followed by 1187. Its answer to "who is at risk" is 1140, ninety-six days since contact. Both came from the same rows the tiles used.

Where it goes wrong

Identifiers that do not join. The whole thing rests on the client identifier mapping. Count the unjoined rows and fix them first.

Wallet estimates without a basis. A benchmark figure with no recorded source becomes a fact in a slide. Keep the basis on the row.

The model doing the sums. If the assistant computes figures itself, the numbers have no provenance and the tiles and the answers drift apart. The model chooses and explains; the functions compute.

Scope in the prompt. A salesperson must only see their own book. That is enforced in the data layer before any row is read, not by asking the model nicely.

The product this describes

This is the shape Covirage was built in: five sources joined on the client, a reconciled roll-up, a tool registry serving every screen and the assistant, and a deterministic floor when the model is unavailable. The sales intelligence page describes it, and you can upload a sample and see the roll-up on your own rows.

Questions people ask

Is this a CRM replacement?

No. The CRM stays the system of record for activity and pipeline. Sales intelligence reads it alongside finance and product data and answers the questions the CRM's reports were never built for: penetration against wallet, coverage against plan, the client whose activity fell before its revenue did.

What is the minimum data to start?

Revenue by client and product, and a coverage list of who owns which client. That gives penetration, ranking and the roll-up. Activity adds coverage and risk; a benchmark file adds wallet share.

How is this different from a dashboard?

A dashboard answers the questions someone anticipated when they built it. Sales intelligence lets the salesperson ask the next one, and the answer comes from the same functions the dashboard used, with the rows shown.