A scenario
A build window you pay for and a demand lift you hope for. This is the only scenario on the list modelled as a chance rather than a plan, because it is the only one where the thing you are deciding is whether to take a bet.
Run this on your own numbersAll scenarios
Ships as a scenario: one lever, launch_product, starting in month one, with a four month build and a 55 percent chance of landing.
Mechanics
Five numbers: build months, monthly build cost, quality lift, demand lift, and the chance it lands. The last one is the one that makes this scenario different from every other page here.
In each replication the engine draws once, at the moment the lever fires, against the hit chance. That draw decides whether the product lands. Nothing that happens during the build changes it. Across 200 replications at a 55 percent chance, about 110 land and about 90 do not, and the band you get back is mostly the distance between those two futures.
The monthly build cost defaults to 8 percent of one month of revenue, about 95,000 dollars on the invented Harborline twin. It sits on the cost line for the whole build window and comes off in the month the product lands, whether or not it worked.
The engine holds the build cost as the larger of itself and whatever is already on the extra cost line, rather than adding the two together. So if the same scenario also enters a new market, the two monthly costs do not stack while the build runs. If you are combining levers, check the cost line in the series rather than assuming.
The demand shock index is multiplied by one plus the demand lift, 8 percent by default, and the quality target gains the quality lift, 0.12 by default, permanently from that month. An event records it. Quality then moves 22 percent of the way to the new target each month, so the improvement arrives over a quarter rather than on launch day.
The build cost comes off and nothing else happens. The engine writes a separate event saying so in plain words. Those replications are the ones worth reading, because they are the honest version of the case that was made for the project.
The demand shock index multiplies the new logos calculation. It does not reduce churn on your existing base and it does not raise revenue per account. The quality lift does reach your existing base, through relative quality in the churn deviation and in every named account renewal. So the two halves of a launch arrive in different places.
More new business is more revenue over the same headcount, so utilisation rises. Above 1.12 the strain counter starts, quality falls toward a lower target and partly gives back what the launch gained. A successful launch with no hiring behind it is one of the few ways this engine will show you a good decision going wrong.
The ecosystem
A launch is mostly an internal event until the month it lands, which is why the agent list is short and the timing matters more than usual.
Objectives: get the work done without drowning at 0.4, be paid fairly at 0.3, work somewhere that still looks stable at 0.3. A department mapped to product or engineering carries 0.7 on quality, which is why cutting one while building the other is a scenario worth running.
Objective: get the outcome they bought at weight 0.35. They feel the quality lift through the ratio against competitor quality, and only from the month it lands.
Objective: keep getting what they signed for at weight 0.4. A quality lift that lands before a big renewal is worth more than one that lands after it, and the engine will show you which side of the line you are on.
Objective: take share when you give them an opening at 0.45. They do not respond to your launch. Their quality only moves if a scenario moves it, which means the competitive answer to your product has to be added as a separate lever.
Objectives: deliver what was sold at 0.5, keep people from burning out at 0.2. Hires into strain after two months above 112 percent utilisation if runway allows, which is the mechanism that rescues a successful launch.
Objectives: growth that compounds at 0.45, a path to cash generation at 0.35. Four months of build cost against flat revenue is exactly the pattern that starts the down counter.
The five numbers
| Input | Default if you leave it | What it actually decides |
|---|---|---|
| Chance it lands | Fifty five percent | Drawn once per replication at the start of the build. This number is the whole shape of the distribution you get back |
| Build months | Four | How long you pay before you find out, and whether the answer lands inside your horizon at all |
| Monthly build cost | Eight percent of a month of revenue | The certain part of the decision. Set it from the actual plan |
| Demand lift | Eight percent | Multiplies new business only. Not churn, not revenue per account |
| Quality lift | 0.12 | Reaches your existing base through relative quality, permanently, from the landing month |
| Revenue a head can serve per month | From the ledger | Decides whether a successful launch is deliverable or becomes a strain problem |
| New customers per month | From the ledger | The base the demand lift multiplies. A small base makes an eight percent lift almost invisible |
The hit chance is the one number nobody can read off a document. Put it in wide, look at the band, and then argue about it with the run in front of you rather than the other way round.
Honestly
A launch is the decision with the most judgement in it and the least evidence, and the model reflects that honestly rather than hiding it behind machinery.
Two futures in one distribution
Start from your own history. Of the last ten things you shipped, how many moved a number you can point at. That fraction is a better prior than any industry figure, and it is usually lower than the room expects.
Then run it at that number, at twenty points above and at twenty below. If the decision is the same at all three, the hit chance is not the thing to argue about.
Because half the runs are a company that built something that worked and half are a company that spent four months of build cost for nothing. The median sits between two futures that do not resemble each other. Read the p10 and the p90 separately, and read the event list, which says in plain words which of the two happened in that path.
Not in one lever. The nearest approximation is two scenarios: one that builds and stops, and one that builds and continues, compared against the same baseline. The difference between them is what the option to stop is worth, which is usually the number the stage gate argument is really about.
No. Once it lands it is permanent within the horizon, and so is the quality lift. That is optimistic for anything with a novelty effect, and it is the first thing to adjust if you are modelling a product with a short shelf life.
Next
Nine out of twenty replications, at the default chance, are the case nobody wrote a slide for. They are in the run, with the months and the cash position attached.