A CRM report on your biggest customers usually gets made like this: someone exports a table, matches it against invoices from accounting, sorts buyers by revenue and two days later brings in a figure that is already out of date. The question itself is simple. What share of revenue do your five largest buyers carry, and is it going up or down?
The number everyone senses and nobody has
The owner of a smaller company usually knows it by feel. Two big buyers, a few mid-sized ones, a long tail of small ones. But you cannot decide by feel whether you can afford to turn down the discount your largest customer has just asked for. For that you need a number. And that number sits in two systems that do not talk to each other.
The CRM holds companies, contacts, deals and the sales team's notes. Accounting holds issued invoices, credit notes and payments. A revenue concentration report needs both: who the customer is and what they actually paid. Joining them by hand means matching company IDs, fixing typos in names, rolling branches up into one group and making sure a December credit note does not land in November. It gets done once a quarter, if at all.
“I reckon the Nováks account for about a third. Maybe more. I'll send it over by Friday.”
— A management meeting, abridged
What it means for Claude to read the data directly
Instead of an export, we build a bridge. One small MCP server connects to your CRM, another to your accounting. Claude receives the question through them and looks up exactly what it needs: the list of buyers with their company IDs, invoicing for the chosen period, credit notes. Nothing is copied into yet another spreadsheet and nothing is stored in someone else's cache. The data stays where it is.
The bridge carries your identity. When the managing director asks, Claude sees what the director sees. When a sales rep who only has their own accounts in the CRM asks, the answer covers only those accounts. The answer arrives as one sentence, with a table beneath it and links to the invoices it came from, so every figure can be checked.
Concretely: Raynet and Pohoda
Take a common Czech setup: customers and deals in Raynet CRM, invoicing in Pohoda. Both systems stay as they are and nobody has to learn anything new. The only addition is the bridge between them and Claude. For a company with a few hundred buyers and a few thousand invoices a year, that is a volume Claude works through in tens of seconds (an illustrative estimate).
- Matches companies in the CRM to buyers in Pohoda by company ID and flags the ones that will not match.
- Rolls branches and subsidiaries up into one group, if that is how you keep them in the CRM.
- Totals invoicing for the chosen period net of credit notes and ranks buyers by share of revenue.
- Works out how much the top five and top ten names carry and compares it with the same period last year.
- Marks customers whose share is rising or falling fast and attaches the latest sales note from the CRM.
Illustratively: a 40-person manufacturer in the Vysočina region asks on a Monday morning, “What share of last year's revenue came from our five largest buyers, and how did that change against the year before?” The answer arrives before the coffee is finished. In the same thread, the owner can ask what the year would have looked like without the biggest one.
What Claude will not do for you, and why that is good
Claude will tell you that a large slice of revenue rests on five names. It will not tell you what to do about it. Whether relying on one retail chain is a fair price for steady orders, or a risk worth spreading out, is a commercial judgement, and it stays with you. Claude also does not send offers on your behalf, change prices in the CRM or edit anything in accounting. By default the bridge is read-only.
That boundary is what makes the report trustworthy. Every figure traces back to a specific invoice you can open. When Claude hits a buyer with no company ID, or two firms with the same name, it does not pretend to be certain. It says so.
What it would take
Not a year-long project. It starts with a short conversation about where your customers live and where your invoices live. Then we build two small MCP servers, connect them to your logins, and run the whole thing on your own cloud: one platform, one tenant, one audit trail. Once the bridge is in place, it answers the next questions too. How promptly buyers pay, margin by customer, who has stopped ordering.
What is left
Claude already reads and calculates well. The model is not what stands between you and the exact figure. The gap between Claude and the data already sitting in your CRM and accounting is. We close that gap.
If you want to know how much of your revenue hangs on five names, write to us. A short call is enough to see which systems you run and how quickly the bridge could be up.