Section 1
The client who scored you an 8 and left six weeks later
Take a 60-seat construction firm paying 4,800 a month, about 20 percent of your recurring revenue. Here is what happened over eighteen months, in the order it happened. A server migration that was scoped for a weekend ran into Tuesday. Two invoices carried project hours nobody had explained in advance, and got paid anyway. The technician who knew their line-of-business application left your firm and the replacement rescheduled a site visit twice. Their long-serving office manager retired and a new controller came in from a company that used a different provider. Backups were reported as green in the review deck while a restore test was quietly six months overdue. Then their insurer sent a questionnaire asking for written confirmation of multi-factor authentication coverage, and your team took nine days to answer it. On day nine the controller called two other MSPs. On day seventeen they signed with one. Total elapsed time from the first visible sign to a signed contract elsewhere: about two weeks. Total elapsed time from the first grievance: eighteen months. Now look back at that list and try to find the firing offence. There is not one. Not a single item on it would survive being read out as a reason to terminate a four-year relationship. That is exactly why owners come out of these situations convinced something irrational happened, or that the new controller had a friend in the business. Sometimes she did. But the grievance ledger was already nearly full before she arrived, and she was simply the first person in the building with no reason to keep tolerating the balance on it. Your last satisfaction survey, sent four months earlier, came back an 8.
Section 2
Churn is a threshold, not a slope
There is a family of models in social science, associated with Thomas Schelling and Mark Granovetter, built to explain exactly this shape of event. The idea is simple. Each person carries a private threshold, meaning the amount of pressure at which their behaviour flips from one state to the other. Below that number, nothing visible happens. At the number, everything happens at once. The models were originally written about neighbourhoods and crowds, but the mechanism transfers cleanly to a service contract. Applied to managed services it says this. Every client holds an internal count of unresolved irritations, unexplained charges, missed commitments and small credibility losses. They also hold a private number, usually unstated even to themselves, at which the accumulated count makes leaving preferable to staying. Below that number their observable behaviour is identical to a happy client, because complaining is costly, awkward and rarely productive, and because they have already decided that this is simply what having an IT provider is like. Above the number, behaviour changes completely and quickly. Two consequences follow, and both are uncomfortable. The first is that nobody half-switches an MSP. There is no equivalent of reducing your order, shortening a contract or trying a smaller engagement. The account is either fully yours or fully gone, which is why the transition from apparently stable to terminated is measured in days rather than quarters. The second is that the count and the threshold are separate quantities, and you can only ever guess at either one. A client with nine grievances and a threshold of twenty is safe. A client with four grievances and a threshold of five is about to leave. From the outside, on a survey, on a call, in a review meeting, those two clients look the same. They will often say the same words. This is the specific reason experienced operators keep getting surprised by the departures they should have seen and keep wasting attention on the accounts that were never going anywhere.
Section 3
Why the QBR and the satisfaction survey cannot see it
Client satisfaction instruments measure the slope. They ask how things have been going, which returns a reading of accumulated sentiment. They do not and cannot measure the threshold, because the threshold is a property of the client's circumstances rather than their feelings: how easy switching looks to them right now, whether a credible alternative is in view, whether anyone in the building would be blamed for the move, whether a contract date makes the move cheap this quarter. That is the whole problem in one line. You are measuring the numerator and the decision depends on the ratio. Three further things degrade the reading. The person completing your survey is very often the person who originally chose you, which means an honest low score is a personal admission, so the scores drift upward for reasons that have nothing to do with your service. A quarterly review is a scheduled, prepared, semi-social meeting, and clients bring their meeting manners to it. And a client who has already crossed the threshold has every incentive to be pleasant, because they need their documentation, their licence records, their admin credentials and a clean final month. Politeness peaks just before departure. Several owners have told the same story of an unusually warm review meeting followed within weeks by a termination notice, and there is nothing mysterious in it. None of this means you should stop running reviews. Reviews do useful work: they document, they surface roadmap items, they justify price. It means you should stop treating review sentiment as a retention instrument. It is not measuring the thing that decides the outcome, and treating it as though it does buys you false confidence in precisely the accounts where confidence is most expensive.
Section 4
What holds a client in place, and what flips them
The threshold model explains the timing. It does not explain the force holding the client still beforehand, and for that you need a second lens. Before the threshold, the client is governed by status-quo bias, the well-documented tendency to prefer the current arrangement simply because it is current, and by an honest reading of switching cost. Moving providers means credential resets across every seat, agent removals and reinstalls, documentation you may or may not hand over cleanly, a fortnight of unfamiliar faces, and the very real chance that something breaks in the middle of month-end. Against that, a provider who is merely irritating is cheap to keep. The rational move is to absorb the irritation, which is exactly what they do, for years. After the threshold, the frame inverts. The client stops asking what they might lose by moving and starts asking what they are risking by staying: the next unreported failed restore, the insurance claim denied because a control was never actually in place, the outage during a tender deadline. Loss aversion, which had been anchoring them to you, detaches and re-anchors on the other side. Staying becomes the risky choice. Once that flip happens the same person who tolerated eighteen months of drift will now move in eleven days and treat urgency as prudence. This is why the standard rescue attempts fail. A discount addresses price, and price was never the count. A promise to do better addresses the future, and the client has already reclassified your promises as a source of risk. It is also why the buyer's side of this is worth reading in its own right. The mechanics of how a switching decision actually gets made, and what makes it feel survivable to the person who would be blamed, are set out at https://bizgrowthaxel.com/blog/how-to-make-switching-msps-feel-survivable/. Read it as a description of what is happening in your departing client's head, because the MSP that wins them is the one that answers those fears explicitly.
Section 5
The signals that sit next to the threshold
If you cannot detect churn by asking how things are going, what can you detect? Threshold-adjacent behaviour: events that either add several units to the count at once or lower the threshold itself. These are observable, and most of them already pass through your systems without anyone reading them as risk. Ticket tone. Not ticket volume, tone. Watch for the shift from "when you get a chance" to "please confirm you have received this", for one issue per ticket instead of a batched weekly email, for a second name added to the CC line, for the first time someone restates the history of a problem in writing. That last one is a client building a file. A new decision-maker. A new controller, operations manager, practice manager or second-generation owner arrives with no relationship equity, no memory of the crisis you handled well in 2023, and frequently a provider they already trust from their last job. Their threshold starts near zero and they inherit the full count. This is the single highest-value signal on the list and it is usually sitting in a signature block nobody read. A renewal or auto-renew date inside ninety days. Switching costs drop at the boundary, which lowers the threshold without changing a thing about your service. An insurance renewal or a security questionnaire. These force the client to enumerate controls in writing, which converts vague unease into a specific documented gap, and they arrive with an external deadline attached. One visible failure. Visibility matters more than severity here. An outage the whole office watched adds far more to the count than a serious problem only you knew about. Requests for documentation, asset inventories, licence records or admin credentials. Sometimes innocent. Frequently the first operational step of a move already decided. Treat a documentation request as a stronger churn signal than any survey answer you have ever received.
Section 6
Building a threshold register, and what it costs you
The action here is small and unglamorous. Once a week, twenty minutes, one sheet listing every client with the following columns: share of your recurring revenue, decision-maker changes in the last twelve months, visible failures in the last ninety days, contract or renewal date, insurance or compliance date, a tone flag from the service desk, and any documentation or credential request. Score it however you like. The scoring matters less than the fact that somebody is looking at threshold proximity rather than sentiment. When an account lights up, the intervention is not a survey and not a discount. It is a specific repair plus an explicit reset. Name the two or three things you know went badly, including the ones they never complained about, say what changed structurally so they will not recur, and ask directly whether anything else is sitting on the ledger. Naming your own failures before they do reduces the count in a way no goodwill credit does, because the count is made of unacknowledged items, not unpaid ones. Now the costs, because there are real ones. This process generates false positives, and you will spend hours on accounts that were never leaving. Budget for that rather than abandoning the register the first time it is wrong. It depends on technicians flagging tone, and technicians dislike reporting anything that could be read as a complaint about their own handling of a client. If you do not make flagging consequence-free, the column stays empty and the register is worthless. And it will not save every account. Some clients cross the threshold on an event you were never going to see. The realistic gain is recovering a meaningful share of the departures that were visible in hindsight, not eliminating churn. Retention and acquisition are the same mechanism viewed from opposite ends. Every threshold crossing in someone else's client base is a buyer entering the market for about two weeks, which is the entire reason MSP marketing has to be built for recall at the moment of failure rather than for gradual persuasion. That side of the argument sits at https://bizgrowthaxel.com/msp/.
Section 7
When every threshold moves at once, and what none of this sees
The model above treats each client's threshold as independent. Usually that holds. Occasionally it does not, and the failure mode is severe. A significant breach at a known business in your metro, an insurer imposing a new control requirement across a whole class of policies, a regulator changing a compliance baseline, or a large regional competitor arriving with a funded migration offer: any of these lowers the threshold for every client at the same time. The count does not change. The bar drops. A book of business that looked stable for years can then produce three or four departures in the same quarter, and the owner reasonably concludes that something has gone wrong with the firm when what has actually happened is that the environment moved underneath a stable service. This matters most because insurance and compliance cycles cluster by season, so correlated churn tends to arrive in the same weeks. Two practical responses. Track the shared dates on your register, not just the per-client ones, so a cluster is visible before it fires. And treat concentration honestly: concentration risk is not only one client at 30 percent of revenue, it is also thirty clients who all renew their cyber policy in March. The honest limit of everything above: none of it sees individual relationship dynamics. A client can cross their threshold for reasons that have no relationship to accumulated grievance at all. The most common version is one of your own people leaving. If a trusted engineer built the relationship, knew the client's systems by memory and answered the phone by name, their resignation can move that client's threshold on its own, immediately, with a clean grievance ledger and a genuinely excellent service record. No register with a tone column will catch that, because nothing about the client changed. The workable adjustment is to add one row to the register for your own staffing: when an engineer resigns, list the accounts they owned and treat every one of them as threshold-adjacent for the next quarter. That is a patch on a blind spot, not a fix for it. Treat the whole framework as a guide to where attention is worth spending, not as a prediction about any individual account.
Section 8
The fitness test
You are ready to run churn as a threshold problem if most of the following describe you. You have more than fifteen recurring clients, which is enough for pattern-watching to beat personal memory. Your ticket system records who submitted what and when, so tone and pattern changes are actually retrievable rather than reconstructed. You know every client's contract date without looking it up, or you are willing to build the list this week. You can absorb twenty minutes a week of owner-level attention that produces nothing visible for months. And you are prepared to open conversations by naming your own failures first, before a client raises them, which is the part most owners quietly refuse to do. You are not ready if any of these are true. You have fewer than ten clients and speak to all of them weekly, in which case a register adds process without adding information you do not already hold. Your service delivery has an unfixed structural problem, such as chronic understaffing at the service desk, because a register will simply produce an accurate list of accounts you cannot save and the money is better spent on the delivery gap. You would use the signals to pre-emptively discount, which converts a retention system into a margin leak. Or your firm is under fifteen months old with most clients still in their first year, where the dominant pattern is early-life fit failure rather than accumulated grievance, and those are diagnosed differently. One last note on judgement. In any single quarter, local conditions, one competitor's hiring and one badly timed outage will dominate the outcome, so do not read a good or bad quarter as evidence about the system. Judge this on the trend across four to six quarters, and on whether the departures you did have showed up on the register beforehand. That second measure is the one that tells you whether you are watching the right thing.