Forecast-Line Ask

A company cuts 15% of engineering, and an architect still needs eighteen months of platform investment. The ask that arrives as "we need six engineers" is declined before it is finished, because headcount was the thing just removed. The cut did not make the investment less necessary; it removed the only currency the request was denominated in.

A forecast-line ask is the alternative: find a number the finance function already forecasts — revenue at risk on a named event, infrastructure cost per order, gross margin points, cost per active user — and express the request as a change to that number, on a date.

The mechanism is that an executive evaluates a request by inserting it into a model they already hold. A request in units that are not in the model has to be translated by the listener, under time pressure, and the translation usually fails in the direction of a decline.

Why it matters

Architecture asks compete against every other request for the same budget, and the competition is decided by comparability. Two asks denominated in the same unit can be ranked. An ask in engineer-months against an ask in revenue cannot, so the one in revenue wins by default.

The second reason is durability. A number taken from the forecast comes with a date and an owner already attached, so the conversation six months later is about whether the number moved rather than about whether the work was worthwhile.

Implementation patterns

• Ask four questions before writing anything: which event in the next eighteen months carries the largest single-day revenue exposure; is the cut structural or a freeze with a reopening date; what is infrastructure cost per unit today and is it rising faster than volume; which number does the executive already report upward.
• Lead with the event, not the system. "Our largest onsale next year is in March and carries X of the quarter" puts the ask inside their model in one sentence.
• State the failure in commercial language. A cancelled sale, a missed regulatory date, a refund wave. Availability percentages do not describe any of these.
• Attach reversibility and a kill number: twelve weeks, one squad, no new headcount, and the month-three measurement that stops the work.
• Shrink the ask to the next decision rather than the end state, so the amount being granted matches the confidence available.

Industry example

Ticketmaster's November 2022 Eras Tour onsale is the useful anchor, read as a communication case rather than a capacity one. The company reported about 3.5 billion system requests, four times its previous peak, with about 3.5 million people registered through Verified Fan and only a limited number issued access codes while the remainder were waitlisted; more than two million tickets sold in a single day; and the next day's public sale was cancelled. The incident drew a letter from a United States senator.

None of that outcome is expressible as an availability figure. It is revenue, brand and regulatory exposure concentrated into one date — exactly the shape a forecast-line ask is built for. The same facts also reframe the request itself: capacity for 4× peak is bought once and idle for 364 days, while an admission mechanism that lets the business choose how much demand to let in is cheap and useful every day. Verified Fan was already the shape of that mechanism.

Failure scenarios

• The number is invented. A revenue-at-risk figure the finance team cannot reproduce ends the conversation and damages the next one.
• The ask is denominated in a number nobody forecasts, such as p99 latency, so the translation is left to the listener.
• The event framing is borrowed. A business with steady traffic and no concentrated demand events gets an argument built on someone else's outage, and it reads as such.
• Reversibility is claimed and not true, and the kill number is quietly dropped at month three, which costs the architect the mechanism for every future ask.

Trade-offs

Denominating an ask in the forecast means accepting the forecast's own frame, including its horizon and its assumptions. A platform investment whose return lands in year three is hard to express in a model that looks twelve months ahead, and compressing it into that window either understates the value or overstates the near-term effect. Pulling a number from finance also means finance now owns part of the argument, which is usually good and occasionally inconvenient.

When not to use it

When the decision is not about money. A request for a change in how a team is organised, a decision-rights question, or an engineering standard does not belong in a revenue frame, and forcing it there invites a cost-benefit debate about something that was never a spending question. Also avoid it when no credible number exists: an honest "I cannot size this yet, and here is the two-week piece of work that would" is stronger than a denominated guess.

Interview question

Q: Your CFO has just signed off a 15% reduction in engineering and you need eighteen months of platform investment. You have one meeting. What do you bring, what do you deliberately leave out, and what is the sentence you want them to repeat afterwards?

What a strong answer covers: the clarifying questions about concentrated revenue events and current unit cost; leading with the event rather than the system; the failure expressed commercially rather than as an availability figure; reframing capacity into demand admission; reversibility with a stated kill number and no new headcount; leaving out technology names, architecture diagrams and availability targets; and a single repeatable sentence such as the cheapest version of this being the right to turn demand away in a way the brand survives.

Quick check

Quiz: After a 15% engineering cut, which currency is unavailable for an investment case and what replaces it? — Headcount; replace it with a number already in the forecast, such as revenue at risk on a named event or cost per unit of volume.

Flashcard: Why does "we need six engineers" fail as an ask after a reduction? — The reduction was the answer to that question, and the request is denominated in the one unit the decision-maker has just removed from the model.