Customer profitability analysis: joining ERP, CRM and support data step by step

How do you run a customer profitability analysis?

Calculate each customer's revenue, subtract the cost to serve it, and rank the result. Revenue and direct cost come from the ERP or accounting system, cost-to-serve drivers come from CRM activity and support tickets, and the join key is a clean customer ID. The work is mostly data preparation, and doing it shows a fractional CFO how many systems and years a client really has.

This guide is for fractional and outsourced CFOs doing the analysis for B2B clients. It covers the model, the join, the traps, and how to note the number of systems and years as a fit signal. All customer-level data stays inside the company; nothing about individual customers is shared with anyone else.

What inputs does the analysis need?

Steps to build it

1. Create a customer master. Match IDs across ERP, CRM and help desk. Merge duplicates and parent-child accounts.
2. Choose the period. Use at least twelve months; use more where contracts or seasons are long.
3. Allocate direct cost. Use job or project costing where it exists; otherwise allocate by drivers such as units or hours.
4. Allocate cost to serve. Use tickets handled, support hours, account manager time and meeting counts as drivers.
5. Compute contribution margin by customer. Revenue minus direct and cost-to-serve allocations.
6. Rank and band. Group customers into profitable, marginal and unprofitable bands; look at size, product mix and tenure.
7. Test the finding. Share with the account owner before concluding anything; context matters.

An Excel or spreadsheet version works for under a few hundred customers. Beyond that, put the model in the reporting tool the client already uses.

How should you present the results?

Lead with a one-page summary: the share of customers in each band, the contribution margin of the top and bottom deciles, and three actions the owner can take, such as repricing, minimum order terms or a support tier. Put the allocation method and data period in a footnote so the reader can challenge it. Offer a second view by segment and tenure, because averages hide the accounts that are expensive only in their first year. Agree with the owner which customer names appear in any board material before it circulates.

What are the common mistakes?

What does the join reveal about the client's records?

When you join ERP, CRM and help desk, you learn how many systems exist, how many years each covers, and whether older versions were archived. Companies where ticket histories link to accounts, accounts link to deals and deals link to invoices hold a connected record of how work was done and what it earned. That connection is what gives records value to AI buyers; for the support side see licensing customer support data for AI.

Record the observations only. You do not need to read, export or describe tickets, emails or invoices.

• The company had 50+ full-time employees at peak (contractors excluded).
• At least three connected systems carry several years of data.
• Older systems were archived, not deleted.
• The company created the records and customer contracts do not bar licensing.
• An authorized sponsor (owner, CEO, CFO or authorized representative) exists.

Run the same list through the company fit checker. The who qualifies page states the full baseline.

What privacy and contract issues come with customer-level data?

Customer-level data has commercial and personal elements. Customer contracts may restrict how a supplier uses information about the customer relationship, and support tickets can contain personal data from the customer's staff.

The company and SourceX settle these in the rights review. A partner only makes the introduction.

How to raise the introduction

Wait until the analysis is delivered and discussed, then ask permission.

If the owner agrees, share your referral link or submit basic company details through the referral form. The company works directly with SourceX. Before the owner commits, who can sign a data license is a useful question to settle. If the client is moving to a new platform, read about what happens to legacy history in an AI-native ERP and preserve exports first. The forecasting guide and churn analysis guide cover related work.

How do rewards work?

Partners earn 25% of the eligible platform fees SourceX actually collects from the referred company's licensing deals, capped at $100,000 cumulative per referred company. The reward is payable only after the buyer pays and SourceX receives its fee; no reward is guaranteed. It is never deducted from what the company receives. Your engagement letters and any professional rules on referral fees and disclosure come first; the program terms set out the rest. For the broader context see the fractional CFO overview and the CAS growth guide.

When to skip it

• The customer base is mostly consumers with no licensing basis for their personal data.
• The client acts mainly as a processor on behalf of customers, so records belong to them.
• Systems are new and history is under two years.
• The data was already licensed for AI training.

Next step

After the next profitability project, write down the systems joined and their year ranges. If the picture is deep and the owner is open, register as a partner and make the introduction.