Revenue predictability is a post-sale design problem.

Most SaaS companies forecast renewals the same way. The CSM closest to the account gives a gut feel. The CS leader aggregates those gut feels. The CRO adds a confidence factor. The board gets a number that nobody fully believes.

The problem isn't the people. It's that the data feeding that forecast was never designed to answer a commercial question.

Where renewal confidence really comes from.

I've sat in enough QBRs across enough geographies to know that renewal confidence almost always comes down to one thing: whether the post-sale motion was built with commercial intent from the start, or assembled reactively as the company grew. The former gives you forecasting infrastructure. The latter gives you informed guesswork.

Revenue predictability isn't a CRM problem or a CSM headcount problem. It's a structural design problem.

Three decisions that determine the forecast.

Revenue predictability sits in the architecture of how your post-sale function captures data, segments accounts, and defines success. Each of those is a design decision that gets made once and then compounds quietly — for or against the number you eventually need to defend. They are rarely made deliberately. In most companies they are inherited — from the tool that was bought first, the segmentation Sales already used, the metric that was easiest to measure in month one. None of that is negligence; the decisions were made before anyone recognised them as decisions. But they don't stay small. Each one quietly sets a ceiling on how much you can know about your own revenue, and you inherit that ceiling whether or not you ever chose it.

How you capture data

Most companies have more customer data than they can use and less than they need. Logins, tickets, feature usage, NPS responses, meeting notes. What's missing isn't volume. It's intent. Nobody decided in advance which signals would indicate commercial risk, so nothing in the stack is pointed at the question the forecast is asking.

The test is simple: can you name the three signals that most reliably precede a churn event in your business, and does your system surface them without someone going to look? If the answer involves a CSM remembering to check a dashboard, you don't have an early warning system. You have a reporting layer.

Designing it properly means working backwards from the decision rather than forwards from the data. Start with the intervention — what would you actually do differently if you knew an account was drifting eight weeks out? Then identify the smallest set of signals that would trigger it, instrument those, and set explicit thresholds. Four or five signals with defined trigger points beat forty tracked fields and a health score nobody trusts.

The other half is capturing what only humans know. Sponsor changes, reorganisations, budget pressure, a champion who has gone quiet — these predict churn better than any usage metric, and they live in conversations rather than in the product. If there's no structured place to record them, they leave when the CSM does.

How you segment accounts

Segmentation looks like an efficiency decision — where to point finite CSM capacity. In forecasting terms it is something more specific: it determines how much you actually know about each account, and therefore how much of your forecast is evidence rather than inference.

Where coverage is undifferentiated, everyone is spread evenly and thinly. The twenty accounts carrying most of your renewal value get read at the same shallow depth as the two hundred that carry very little. Confidence ends up uniform across the book, and uniformly low — because nobody has enough contact with anything to say much beyond an impression.

Deliberate segmentation concentrates depth where the money is. The accounts that dominate the number get real engagement and real knowledge of what is happening inside the customer, which is what lets someone say specifically why an account is rated the way it is and what would have to change.

The more useful consequence is that segmentation tells you which parts of the forecast you are entitled to be confident about. A tech-touch segment forecast from product signals is a statistical estimate: sound in aggregate, unreliable account by account. A high-touch account forecast from a relationship is a judgement, and defensible as one. Both are legitimate. The failure is reporting them with the same apparent certainty, so that a modelled number and an informed one sit in the same column looking equally solid.

That reframes the service-level discipline. Deciding what the long tail doesn't get isn't only about capacity. It's about being explicit that you don't have deep visibility there — and building a forecast that doesn't pretend otherwise.

How you define success

This is the decision that quietly determines everything else, because it sets what the team optimises for when nobody is watching.

Most CS functions are measured on satisfaction — NPS, CSAT, response times, QBRs delivered. Those measure whether customers are content with the relationship. They don't measure whether customers are getting enough value to keep paying, and the two diverge more often than people expect. Accounts churn with healthy NPS scores routinely. The sponsor liked you. The business case never landed.

The better definition is customer-outcome based: did this customer achieve the thing they bought the product to achieve, and can you evidence it? That requires agreeing the outcome at the start, in the customer's terms and their numbers, and revisiting it deliberately rather than at renewal. It's harder than sending a survey. It's also the only version of success that predicts renewal, because it's the same question the customer's CFO will ask.

And it has to carry commercial accountability. If CS is measured on satisfaction while Sales is measured on revenue, you have built a structural gap between the function closest to the customer and the number the business runs on. Give CS a retention and expansion target. Not to turn CSMs into sellers, but because ownership changes which signals register as important — a sponsor going quiet reads as a scheduling problem to a team measured on satisfaction, and as a risk to a team measured on retention. A team accountable for a number forecasts it far more honestly than a team that isn't.

The compounding effect

None of these three is dramatic on its own. Taken together they determine whether your renewal forecast is an instrument or an opinion. Data designed around commercial questions tells you where risk is accumulating. Segmentation designed around coverage depth tells you which parts of the number are evidence and which are estimate. Success defined as customer value, owned commercially, means the people closest to the account are answering the question the board is actually asking.

Get them right and the forecast becomes something you can interrogate: you can point at why an account is rated the way it is, and what would have to change. Get them wrong and no amount of process discipline compensates, because you are aggregating gut feels more carefully.

Fix the structure, and the forecast starts to mean something. Leave it implicit, and the call at the end of every quarter is going to keep sounding the same.