# Demand budgets

URL: /workflows/marketing/demand-budgets · Id: `page:workflows/marketing/demand-budgets`

> How a Startup buys demand at $0 by default, and how a founder-funded budget becomes a Play mix with a target and a stop rule.

A [Demand Budget](/business/nouns/demand-budget) is the money a founder puts behind demand each month. A long-tail [Startup](/business/nouns/startup) runs at $0, and a funded budget buys a mix of demand [Plays](/business/nouns/play), each given enough to produce a readable result.

## Zero is the default
Most Startups never fund a budget, and they still find [Customers](/business/nouns/customer). The free Plays need no audience of their own: answer engines, search, directory and registry listings, launch sites and community threads. Borrowed audiences add the rest: referrals, house ads across the [Portfolio](/business/nouns/portfolio), and posts from the [Studio](/business/nouns/studio)’s shared social accounts, which already have followers.

Zero means no media spend, not no cost. Prompt panels, rank tracking and enrichment are metered as platform usage, about $5 to $20 a month. Siblings that track the same queries share most of it.

Paid spend has a strict default: a paid Play opens only when the first transaction pays back its acquisition cost on day one. A funded budget is the one deliberate exception.

## Minimum effective budget
Every paid Play declares a minimum effective budget: the smallest monthly spend that yields a readable signal on its [Metric](/business/nouns/metric) within its latency. Below it, money buys noise.

An ad network optimizing for purchases needs about 50 conversions per ad set each week to stop learning. A purchase test must buy that many.

Each Play also declares a cost curve: how the cost of the next outcome changes as spend rises.

| Curve | Shape | Plays |
| --- | --- | --- |
| Flat | An asset; the next outcome costs almost nothing | [Answer engine optimization](/workflows/plays/aeo), [programmatic SEO](/workflows/plays/programmatic-seo), listings |
| Step | Lumpy units: a mailbox, a slot, an event | [Cold email](/workflows/plays/cold-email), [sponsorships](/workflows/plays/sponsorships), [events](/workflows/plays/events-and-webinars) |
| Linear | A price per outcome | [Customer referrals](/workflows/plays/customer-referrals), [commission partners](/workflows/plays/commission-partners) |
| Concave | The cheapest intent goes first, then each click costs more | [Paid search](/workflows/plays/paid-search), [retargeting](/workflows/plays/retargeting) |
| S-curve | Wasteful below the learning threshold, efficient in the middle, then saturating | [Paid social](/workflows/plays/paid-social), [video ladder](/workflows/plays/video-ladder) |

The [Marketing Team](/agents/marketing) never splits a budget across more paid Plays than their minimums add up to. A $1,000 budget runs paid search or a social test well, never both badly.

## What each tier buys
Each tier includes every Play below it. At $1k, about $850 goes to media, $50 to data and $100 to creative; at $10k, video creative takes about $800.

| Tier | Adds | Default stop rule |
| --- | --- | --- |
| $0 | Answer engines, search, listings, agent registries, launch sites, community answers, shared accounts, referrals, house ads | None: assets compound |
| $100 | Signal-led cold email, retargeting once the list qualifies, an exact-match test on Bing, app store fees | Under 1 percent positive replies after 500 sends |
| $1k | Paid search on Google Search, one test on Facebook and Instagram or Reddit, one sponsorship, five to ten flat-fee creators | [Cost](/business/nouns/cost) per first transaction above its margin for two weeks |
| $10k | Facebook and Instagram at scale, TikTok, YouTube and LinkedIn [Ads](/business/nouns/ad), paid placements on G2 and Capterra, webinars | One placement above 1.5 times the target for two weeks |
| $100k | Every placement at scale, brand-search defense, conferences, holdout tests | Incremental lift below its threshold |

## Funding a budget
A founder funds a budget by recording a [Decision](/business/nouns/decision) with three parts: the amount, a target and a stop rule. The target is a Metric value, such as cost per first [Order](/business/nouns/order) under $40. The stop rule is the reading that ends the spend.

The [Finance Lead](/agents/finance/finance-lead) sets the amount against runway, and the [Marketing Lead](/agents/marketing/marketing-lead) composes the mix. The budget overrides the day-one payback default only for its own window and target.

When the stop rule fires, the budget closes and the Decision records the [Evidence](/business/nouns/evidence). Nothing is deleted, so the next budget starts from what this one learned.

## Spend under Limits and windows
All spend runs under the Startup’s [Limits](/business/nouns/limit). A launch, a raise, a new placement or an audience sync is a one-way act. It waits the ad-spend [delay window](/agents/autonomy-levels/delay-windows), where any watcher can pause it.

A raise above twice the current spend waits one business day, and a pause never waits. Spend states bind every Play at once: a tight state blocks raises, and a hold blocks launches.

## The weekly Play Mix
A [Play Mix](/business/nouns/play-mix) is the set of Plays a Startup runs this week, each with its state, budget and stop rule. [`PlayMix.compile`](/business/nouns/play-mix/compile) builds a new version every week from the Demand Budget, each Play’s ledger and last week’s [Acquisitions](/business/nouns/acquisition).

Each Play keeps a ledger computed from [Events](/business/nouns/event), never typed in: `spend`, `workerCost`, `customersAcquired`, `outsideRevenue`, `margin` and `paybackDays`. Revenue reaches it on [`Invoice.collected`](/business/events/Invoice.collected), which walks back to the Customer’s Acquisition and the Play named there.

The compile runs in a fixed order.

1. Keep the Plays that fit the [Offer](/business/nouns/offer)’s [Motion](/business/nouns/motion) and the Startup’s [Lifecycle](/business/nouns/lifecycle) state.
2. Keep every free Play running.
3. Give each paid Play at least its minimum effective budget, or leave it out.
4. Open a gated Play only when its gate’s Metric crosses its threshold.
5. Give the next dollar to the Play with the shortest marginal [Play payback](/agents/marketing#play-payback).
6. Write the diff: one line per Play, with the reading that moved it.

A running Play holds until its loop window closes with its sample met, or until a guardrail stops it. The compile changes at most two Play states a week, and each randomized allocation logs its propensity.

Every version is kept, so a pivot reads as a diff. One week’s diff reads: "cold email paused at 0.6 percent positive replies; paid search raised to $1,200."

## A worked example
A Startup selling a bookkeeping agent to freelance designers funds $1,000 for 60 days. Its target is cost per first paid month under $45, and its stop rule closes the budget after two straight weeks above $75. Answer-engine [Pages](/business/nouns/page) and directory listings keep running for free.

The Marketing Lead cannot fund search and a social test together, so the Portfolio decides. Sibling Startups selling to freelancers converted better on search, so $850 goes to [Paid search](/workflows/plays/paid-search) on six exact terms. The rest buys data and two landing Pages.

Weeks one and two bring five Customers at $68 each: over target, under the stop line. The [Analyst](/agents/marketing/analyst) cuts the two broadest terms, and weeks three and four bring 14 Customers at $38. The founder renews at $1,000 and the Decision keeps both readings.

## Next
- [Demand plays](/workflows/plays): every Play with its tier and minimum budget.
- [Creator partnerships](/workflows/marketing/creators): borrowing an audience before paying for one.
- [Marketing playbook](/workflows/marketing/playbook): seed, then compound.