A scenario
Concentration is a ratio in a board pack until the month it is an event. This scenario is that month, with the revenue going, the reference going with it, and none of the cost leaving on its own.
Run this on your own numbersAll scenarios
Ships as a starter scenario: one lever, lose_customer set to the top account, landing in month three.
Mechanics
The lever is small. What follows it is not, and most of it runs through the cost line rather than the revenue line.
Before the lever fires, every named account is living its own year. Each has a renewal clock drawn per replication between 40 and 130 percent of its contract term, and a logistic that weighs price, relative quality, its own satisfaction and switching cost when the clock runs out. In some replications your largest customer has already gone before month three arrives.
It picks the highest revenue named account that has not already left, marks it gone and books an event at the top severity. If that account had already churned in this replication, the lever takes the next largest instead, which is correct behaviour and worth knowing when you read the event list.
Operating cost is headcount times cost per head times the wage index. Headcount does not change when a customer leaves. On the invented Harborline twin, the largest account is 18 percent of 14.24 million dollars a year, about 215,000 dollars a month, carrying roughly 67,000 dollars a month of gross profit against an operating profit of about 74,000. The model does not assume you shed the capacity that served them, because most companies do not, at least not in the first quarter.
Satisfaction drops two points and a morale shock of 0.05 is booked. The shock is subtracted from morale in the month it lands and then keeps 55 percent of its size each month until it falls under 0.01, so it is gone inside two or three months. Small, but it arrives in the same month as everything else.
Less revenue over the same headcount means lower utilisation, which raises the quality target, which raises satisfaction. That is a real effect and it is also the model being generous: it assumes the freed capacity goes into serving everybody else well rather than into people standing around.
Attainment is what got booked against quota times reps. The lost account does not change quota. Under 0.72 the miss counter runs, rep attrition rises by 1.6 times the shortfall below 0.85, and a rep who leaves is backfilled at hiring cost and starts a fresh ramp. That is the second hole, and it opens a quarter after the first.
If the loss pushes operating profit negative, the down counter runs and the investor is told. If runway falls below a threshold set by the finance lead's own risk appetite and patience, between 4 and 20 percent of headcount goes along with 30 percent of marketing, and a morale shock of 0.12 plus the cut percentage lands on a company that has already had a bad quarter.
The ecosystem
This is the scenario where the interesting agents are inside the building.
Objectives: keep getting what they signed for at weight 0.4, renew at a price they can defend at 0.35. In this scenario the decision is imposed rather than simulated, which is the point. You are asking what it costs, not whether it happens.
They carry on with their own clocks. A concentrated book means several large renewals sit inside the same twelve months, and the run will show you which ones land near the hole.
Objectives: hit quota at 0.55, keep the accounts they already have at 0.25, avoid a quarter that ends their year at 0.2. Losing a flagship raises the third one, and the people who can leave are usually the ones you want to keep.
Objectives: protect margin at 0.45, protect cash at 0.4. Acts on runway, on a six month cooldown, and takes marketing down 30 percent at the same time as headcount.
Objectives: growth that compounds at 0.45, a path to cash generation at 0.35, no surprises at the remainder. Patience is a number of months in the ledger and a covenant floor is a number in cash.
Objective: work somewhere that still looks stable, weight 0.3. Morale drives attrition through an exponential, and attrition is backfilled at 70 percent by default, at full hiring cost.
The ledger
| Assumption | What it does here | Why to check it first |
|---|---|---|
| Largest customer share of revenue | Sets the size of the hole if no customer list was uploaded | With a customer list the model uses the actual account. Without one it takes this share off every segment, which is a cruder thing |
| Average months left on contract | Sets every named account's renewal clock, drawn at 40 to 130 percent of it | Decides whether the other large renewals land inside your horizon or just outside it |
| Gross margin | Turns the lost revenue into lost profit | The whole answer scales with it, and a blended margin hides the fact that big accounts are often the thin ones |
| Monthly cost per head | The cost that does not leave with the customer | If this is a default, the size of the remaining cost base is a default too |
| Revenue a head can serve per month | Drives the utilisation fall and the quality rise that follows it | The optimistic part of this scenario runs through here |
| Monthly quota per rep and sales attrition | Turn a revenue hole into a sales team problem one quarter later | Most plans for losing a customer forget this line entirely |
| Months investors will fund a dip, cash floor before a covenant bites | Decide whether this becomes a forecast or a phone call | Worth setting from the actual agreement rather than from memory |
Top five share of revenue is computed into the ledger and shown in the graph, but the simulation itself does not read it. It is there to tell you whether to run this scenario twice.
Honestly
The mechanics of the hole are right. The mechanics of what a real company does about the hole are thin, and they are thin in the direction that makes the answer look worse than it will be.
On the invented Harborline record there is a second problem the model cannot see: two months of cash are tied up in work in progress on that customer's programme. There is no working capital in this engine. Receivables, payables and inventory do not exist, so cash moves with operating profit and one off items only.
The earliest month you could know
The lever falls back to the largest customer share assumption and takes that fraction off every segment in the month it fires. It says so in the event. It is a reasonable approximation of the size of the hole and a poor approximation of the shape, because a named account has a renewal date, a satisfaction score and a salesperson attached to it, and a percentage does not.
It should not, for a given seed. Across replications it differs because the other named accounts have renewal clocks drawn inside each run, so in some years the loss lands next to another renewal and in some years it does not. That spread is the answer, not noise around it.
Copy the scenario and add a headcount change at the month you would actually make it, which is usually two or three months after the loss rather than the same week. Then compare the two side by side. The difference between them is the cost of being slow, which is a number worth having before the event rather than after.
Yes. The lever takes a customer name, matched against the customer list. Running it on the second or third largest is often more useful, because the largest one is already the one everybody watches.
Next
The graph shows how much of you sits with one account. This scenario shows what the month after looks like, and the tripwires show the earliest point the real world could have told you.