What each channel really adds, and where the next euro goes

How the model works

The model explains a daily KPI, revenue, contribution before ad spend or new customers, with up to two years of spend per ad platform. It accounts for trend, weekday, yearly season, discounts, Black Friday to Cyber Monday and your own promotion calendar. Each channel has a carryover (adstock) and a saturation curve. It is fitted in ROASaurus itself and retrained every week; the last 14 to 28 days are held out to measure how well it predicts.

Per channel you see incremental sales, iROAS, the response curve with its band, the marginal ROAS of the next 10% of spend, the daily spend at which the curve reaches 80% of its maximum, and carryover as a half-life in days. The page also shows the share of sales driven by ads and the fit against actual sales.

Budget optimiser

Enter a budget for a period and, if you want, a minimum and maximum per channel. The optimiser splits it to get the most sales, or the most profit after ad spend, and compares the result with today's mix at the same budget. It also shows the outcome at 50% to 200% of the budget, so you see what each extra step returns, and flags allocations far beyond what a channel ever spent.

Incrementality tests

• Geo-lift tests: the designer picks test countries or regions that a synthetic control of the other regions predicts well, and a power analysis tells you the smallest lift you can detect in 14 to 56 days. Afterwards you get the lift with a 90% interval, the p-value and the iROAS.
• On/off tests: a channel paused or switched on, everywhere or in some countries.
• Platform lift studies: Meta, Google and TikTok conversion lift results, entered by hand.

Tests calibrate the model: a precise test pulls the channel's curve towards its result, a vague one barely moves it. The calibration page puts the model next to any attribution model, so you can scale a campaign's attributed revenue by its channel's incrementality factor.

Spend curves and daily budget advice

Each campaign's own history gives a spend curve: every day's spend against the orders credited to it. At today's budget ROASaurus reads off the marginal CPA, marginal CAC, marginal ROAS and POAS, and marginal LTV POAS, which adds what new customers buy again within your payback target. Every morning each campaign gets scale, hold or cut, with a suggested daily budget that moves at most one step (20% by default) at a time.

Who it is for

Brands that spend on several channels and want to know which spend is incremental, not only which ads get the last click. The same numbers are in Ask AI and the MCP server, the weekly action plan and Autopilot.

Limits

• The model needs at least 120 days of history with varying spend. With less it explains what is missing instead of guessing.
• It learns only from how spend changed over time: channels that always moved together cannot be fully told apart, and spend far beyond what was seen is an extrapolation.
• The intervals approximate a Bayesian fit; they are not a full one.