पाठशाला Pathshala · ग्राहक Grāhak, The customer · Lesson 27 · Scale
The customer you should fire
A few accounts cost more to serve than they pay, take more of the roadmap than they are worth and are likely to leave anyway. Find them with arithmetic, try to reprice them first, and exit the rest cleanly.
Pathshala, The Founder Library · 11 October 2026 · 7 min read
Every company at scale has a few customers it would be richer without. They are rarely the small ones. They are usually large, heavily discounted, demanding and late to pay, and they are on the logo slide.
This lesson sets out how to find them with arithmetic rather than irritation, how to tell a customer who is expensive from one who is merely loud, why repricing comes before exit, and how to end a relationship without harming your name in a market where buyers talk to each other. A figure costs one account fully.
The curve hidden inside your customer list
Robert Kaplan, who developed activity-based costing at Harvard Business School, found the same shape in company after company. In his account of customer profitability he writes that 15 to 20 per cent of customers generate 100 per cent or more of the profits, the middle 55 per cent roughly break even, and the least profitable 5 per cent incur losses equal to 30 per cent of annual profits. In his example the cumulative profit curve peaks at 130 per cent of what the company actually earned, before the loss-makers drag it back.
The shape exists because revenue is reported by customer and cost mostly is not. Support, engineering and finance are booked as departments, so a customer who uses three times the average support looks exactly like one who uses none. The first job is to put cost back on the customer.
Three costs the revenue line hides
Cost to serve. Support hours, onboarding effort, account management time and delivery costs that scale with the account: hosting, payment fees, field visits. Pull support hours from the ticketing system by account, and ask account managers to estimate their own time honestly for one month. The rough figure is enough; precision is not the point.
Roadmap pull. The engineering spent on work only this customer uses: a custom integration, a bespoke report, an exception to a standard workflow, a version of a feature that the rest of the base will never see. This is the most expensive cost because it is paid twice: once in salaries and again in the feature your best segment did not get. Tag every engineering ticket that serves one customer and total the hours a quarter.
Waiting to be paid. A customer who pays in 150 days on a ₹30 lakh contract is borrowing from you, and you are funding it from equity or debt. If you are a registered micro or small enterprise, the law is on your side: under the MSMED Act, as the government’s MSME Samadhaan portal sets out, a buyer must pay within 45 days of accepting goods or services, late payment carries compound interest at three times the bank rate notified by the Reserve Bank, and the supplier can apply online to the state’s MSE Facilitation Council. Since April 2024 the buyer’s tax deduction for the expense also depends on paying on time, as the lesson on [Section 43B(h)](/library/section-43b-h-paying-msme-vendors-on-time) explains. Checked October 2026.
Then add two judgements the arithmetic cannot make. Churn risk: a customer who has threatened to leave at each of the last two renewals, or whose champion has gone, may leave whatever you do, and every rupee of custom work for them is spent on an account with a short life. Fit: whether they look like the customers in your [ideal customer profile](/library/ideal-customer-profile-on-one-page) or like a different business that bought by accident.
At the starting values a ₹30 lakh contract won at a 30 per cent discount brings ₹21 lakh. Forty support hours a month, 160 engineering hours a quarter and 120 days to pay take it below zero: the account loses about ₹4 lakh a year. Halve the custom engineering and it earns about ₹4 lakh, a thin 19 per cent. Bring the discount down to 10 per cent at renewal as well and it clears 30 per cent. Most accounts that look like candidates for firing are pricing mistakes that compound with each renewal.
Reprice before you exit
The case against firing customers deserves a hearing. Jason Lemkin argues on SaaStr that customers who complain still care, that complaints are often a sign of deep engagement coupled with frustration, and that a segment worth at least 5 per cent of revenue should get a dedicated person to listen to it rather than an exit letter. He adds that not every segment has to be profitable as long as the vast majority are. He is right that loud is not the same as expensive. A demanding customer whose demands point at your roadmap is a research asset.
So the first move is never the exit. It is one of four changes at the next renewal. Reprice: remove the discount that was given to win the logo, or move the account to the plan its usage implies. Charge for the custom work: a separate statement of work at a rate that covers it, or a commitment to build it into the product for everyone. Narrow the service: a support tier with defined hours and response times, priced accordingly. Change the payment terms: advance billing, or a discount for paying within 30 days that costs less than funding the receivable.
Put the change in writing with the arithmetic behind it, in the customer’s terms: what the account uses, what that use costs and what the new price buys. Many customers accept. They often know they were getting more than they paid for. The ones who refuse have told you the relationship depends on you losing money.
Most customers worth firing are a pricing mistake that nobody corrected at renewal. Correct it first; exit only those who refuse.
Exiting cleanly
When a customer will not accept terms that cover its cost, end the relationship in a way you would be content to see described at an industry dinner. Indian B2B markets are small at the top, and the procurement head you exit this year may run procurement at your best prospect next year.
Read the contract first: the notice period, the renewal clause, data return and any service commitments that survive termination. Then decline the renewal rather than terminate mid-term wherever you can. Give notice beyond the minimum: ninety days where the contract says thirty. Provide a full data export in a standard format and confirm in writing when their data has been deleted from your systems. Offer a named alternative if one fits them better. Brief your own team on what to say, which is very little, and say nothing in public. If the customer owes you money, settle the receivable before the final date; once the relationship ends the incentive to pay ends with it.
A worked example: a logistics software company in Gurugram
A company sells route-planning software to distributors at about ₹6 lakh a year per account. Its largest customer pays ₹42 lakh a year after a 40 per cent discount given three years ago to win the logo. Costed fully, the account takes 60 support hours a month, 300 engineering hours a quarter on a custom warehouse integration nobody else uses, and pays in 160 days. On the figure’s rates it loses nearly ₹5 lakh a year.
At renewal the company proposes list price less 15 per cent, the integration moved to a separately priced maintenance contract and quarterly billing in advance. The customer accepts the price and the billing and rejects the maintenance contract. The company declines to renew the integration work, gives six months’ notice of its retirement and offers the customer’s IT team the documentation to run it themselves. The account stays at about ₹60 lakh a year with a true margin near 70 per cent on the same rates, and 300 engineering hours a quarter return to the product the other two hundred customers use.
The annual account review
Once a year, in January, before Indian customers set budgets for the new financial year and before the renewal letters go out. Cost the top fifty accounts by revenue, plus any account with more than its share of support or engineering. Plot them from most to least profitable and draw Kaplan’s curve for your own company. Mark the bottom tenth and write for each: the cause of the loss, the change you will propose at renewal and the date. Decide in advance which accounts you will exit if they refuse, so the renewal conversation is honest. Tag custom engineering every quarter from now on, so next January’s costing takes a day rather than a month.
Report the result to the board as one line: the share of revenue from accounts with a negative true margin, this year against last. The aim is not zero; it is a number that falls.
The figures in the interactive and the worked example are illustrative. Payment rules were checked in October 2026; nothing here is legal advice.
Sources
- Robert S. Kaplan, A Balanced Scorecard Approach to Measure Customer Profitability, HBS Working Knowledge (2005) — 15–20% of customers generate 100% or more of profits; least profitable 5% lose 30% of annual profits; curve peaks at 130%.
- Jason Lemkin, on why not to fire your worst customers, SaaStr (updated 2022) — Customers that complain still care; keep segments worth 5% or more of revenue; not all segments must be profitable.
- Ministry of MSME, MSME Samadhaan: delayed payment monitoring system — Payment due within 45 days of acceptance; compound interest at three times the RBI bank rate; online application to the state MSEFC. Checked October 2026.