What a Customer Success Function Costs to Build
The first thing you build isn't a playbook. It's a decision about who doesn't get your people's hours.
The first thing you build isn't a playbook. It's a decision about who doesn't get your people's hours.
You've been told to build customer success. There's no team, no playbook, no tooling, and a customer list that someone else has been holding together by answering emails at night. The pressure is to show something in the first month.
So you do the thing that looks like progress. You write a playbook. You buy a platform. You post a job description for a customer success manager, copied from a company forty times your size, and you start interviewing.
I've done all three, and I'd gently push back on each one. Every one of those is an artifact of a function rather than the function. You can have all three inside six weeks and still not have built anything, and you won't find out until the first renewal you should have saved goes the other way.
Before any of it, there's a question with an uncomfortable answer, and the answer has to be written down.
Which customers are you willing to lose?
Not as a provocation. As a document, one page, dated, with names or segments on it. A customer success function is an allocation of a fixed number of human hours across a customer list that's always longer than the hours. That's true on day one, and it stays true at every size. You're not deciding how to make customers successful. You're deciding who gets your people's hours and who doesn't.
If you never make the allocation on purpose, it still gets made. It gets made by whoever escalates loudest, one Tuesday at a time.
That's the failure I'd warn anyone about first, because it doesn't look like a failure while it's happening. It looks like responsiveness. Six months in you've got a team that's fully occupied, a customer list where the accounts getting the most attention are the ones with the angriest contacts, and no way to say which of those two facts caused the other.
The reason the allocation can't wait is that the numbers aren't close. They're never close, and doing them once on paper ends the argument faster than any amount of discussion.
Invented numbers, real shape. A company with $9M in ARR across 180 accounts, and two people on customer success, including the person building it.
A working year is about 1,400 hours per person once you take out everything that isn't customer work.1Fourteen hundred assumes holidays, illness, internal meetings, hiring, and the fact that nobody spends a full working day on customer work. Use your own number, but use one, because the round figure of two thousand hours will make the model look affordable in a way it isn't. So there are roughly 2,800 hours to spend.
Spread evenly, every account gets fifteen hours a year. That's under twenty minutes a week, which is enough to be present and not enough to be useful anywhere. It's the worst available outcome, and it's what happens by default.
Allocated instead:
The remaining 325 hours go to building the thing, hiring into it, and the escalations you didn't plan for. They aren't spare.
The 115 are the decision. Everything else in the model is arithmetic once you've made it.
Once the allocation exists, the sequence isn't really a matter of preference. Each step is only specifiable because the one before it exists.
The platform goes last. Most people put it first, because it's the only step on this list you can buy.
The record deserves more than a line. The most valuable thing this work produces is a customer's objectives in their own words, and it's almost always lost, because it lives in a deck in somebody's drive. Whatever you build, build the place that outlives the person who took the note. Sponsors leave, and a function that restarts every time a sponsor leaves isn't a function.
Everything in the second column follows from the first column existing. None of it could have been specified in advance, which is why buying it in month one buys the wrong thing.
The decisions I'd defend hardest are the ones where the answer was no. Each of them was expensive to hold, and none of them looked good at the time.
No operations hire in the first year. It's the most tempting requisition on the list, because the manual work is visible and irritating and a person would make it go away. But doing it by hand is how you learn which parts of it actually matter. Hire it away too early and you automate a process nobody has pressure tested, and you pay a salary to preserve it.
No health score until there were outcomes to score against. A health score built before you have a year of renewal outcomes is a set of weightings someone made up, dressed as a measurement, and it's worse than nothing because people believe it. The version that works is built backwards, from the accounts that actually churned, and you can't build it backwards until some have.
Support wasn't staffed to what customers asked for. This is the one people flinch at, and I understand why. Customers asked for phone support, and they asked repeatedly. Phone support at that size meant a headcount whose whole year of cost had to come from somewhere, and the somewhere was onboarding.
So the trade got made in the open. A dollar into onboarding returned more retention than a dollar into support response time, because the accounts that left weren't leaving over hold times. They were leaving having never got the thing running properly. Support got a faster written channel and a real documentation investment instead, and the response-time complaints stayed.
They stayed for two years. I'd make the same call again, and the reason I can say that with any confidence is that we watched what the churned accounts had in common, and it was never the phone.
Some customers weren't kept. The bottom of that 115 cost more in support and discount pressure than they returned, and the honest thing was to stop pretending they were a growth segment and let them go. That isn't a customer success decision. It's a decision about where the company spends its money, and it needs to be made by the people who own that question and agreed to it in the room. When one of those accounts leaves noisily, the agreement is the only thing standing between the decision and the person who made it.
The title of this isn't rhetorical, so here's the bill.
About nine months before the function looks like one from outside. The first two quarters produce a document, a motion, and a record, and none of them photograph well. Anyone evaluating progress on artifacts will conclude in month three that nothing is happening. That has to be agreed in advance with whoever is evaluating you, or it gets reversed at exactly the moment it starts to work.
A permanent group of customers who are getting less than they want. Not a transitional state. The 115 are structural. Saying so honestly costs you a certain kind of goodwill that the alternative, quietly under-serving everyone, doesn't cost until much later, all at once.
Your own hours, in the wrong shape. The person building it runs the motion themselves for the first thirty accounts. Not to be heroic. Because a motion you haven't personally run is a motion you can't hire against, can't teach, and can't tell is broken.
The escalations you deliberately didn't staff for. Some of them turn into churn. The model isn't that they won't happen. It's that they cost less than the alternative in aggregate, and you have to be able to hold that position while looking at a specific account it didn't work for.
The argument above assumes the hours are genuinely scarce, and that's the assumption everything rests on, more than anything about customer success specifically.
There's a version of this that goes differently. If the interpretation work keeps getting cheaper at the rate it currently is, the fifteen-hour tier stops being fifteen hours of a person and starts being one hour of a person on top of work that was already done, and the arithmetic that forces the allocation loosens a lot. I've written about where I think that stops, and I don't think it reaches all the way, because the part that decides whether a finding changes anything was never the part that took the time. But it moves the line, and someone building this function two years from now should run the numbers again rather than inherit mine.
What doesn't change is the order. The allocation still comes before the motion, the motion still comes before the hire, and the tooling still comes last, because each one is what makes the next one specifiable. The cheaper the tooling gets, the more expensive it becomes to buy it first.
If you try one thing from this, write the one page. Take your actual customer list, your actual number of people, multiply out the hours, and see what falls out at the bottom. The list is usually longer than anyone has said in a meeting, and the argument about what to do with it is a much better argument to have on paper in month one than in a renewal call in month eleven.
If you're in the middle of building one of these and want a second pair of eyes on the page, reach out. Happy to help.