Fractal Rock

Marketing Performance Modeling

Know where the next advertising dollar goes.

In almost every budget we model, a fifth of the spend is already returning nothing — and a fifth to two-fifths is doing most of the work. Insights ranks every placement by the revenue it actually produced and moves your budget across that line. No extra spend. No agency incentive.

TV Radio Digital Social OOH POS
Same annual budget
$32M
Across nine markets
Modelled revenue gain
+$318M
From reallocation alone
Time to answer
41 ms
Native C++ optimisation engine

Eight years, one named client, on the record

Florida Lottery — five straight years of growth, then a real reason it stopped.

Total ticket sales are public record: Florida Lottery grew every year for five years straight, from $6.7B in fiscal 2018 to a peak of $9.8B in fiscal 2023, while the advertising budget behind them held flat. The first year alone added $450M — and growth didn't taper from there. It ran 3% to 21% a year, every year, through five full cycles.

The first decline of the period landed in fiscal 2024, the same year Hard Rock Bet relaunched legal sports betting across Florida. Sales are down two years running since. Not because the model ran out of room — because a new product now competes for the same discretionary dollar.

Our implementation partner, Clarity i2 (formerly Bottomline Analytics), attributes a 24.8% average annual improvement in marketing ROI to the reallocation itself — the part of this growth the method can actually claim. In one fiscal year, 16% of the ad budget was found to be driving 59% of ad-driven revenue across 474 individual placements. That's the shape the next section describes — measured on a real advertising plan, at national scale.

Florida Lottery total ticket sales, fiscal 2017-18 through 2024-25 A bar chart of Florida Lottery's total ticket sales by fiscal year, from $6.7 billion in FY2017-18 to a peak of $9.8 billion in FY2022-23, growing every year in between. FY2020-21 shows an outsized 20.9% jump, attributed to record Mega Millions and Powerball jackpots plus likely pandemic-era spending. Sales then declined in FY2023-24 and FY2024-25, coinciding with the statewide relaunch of legal sports betting in December 2023 — the first decline of the eight-year period. $6.7B FY17-18 $7.2B +6.7% FY18-19 $7.5B +5.0% FY19-20 $9.1B +20.9% FY20-21 $9.3B +2.7% FY21-22 $9.8B +5.1% FY22-23 $9.4B −4.1% FY23-24 $9.1B −3.2% FY24-25 Record jackpots + likely pandemic-era spending Statewide sports betting launches Hard Rock Bet, Dec 2023 — mid FY2023-24
Florida Lottery total ticket sales by fiscal year. Real figures, not illustrative — source: Florida OPPAGA, Review of the Florida Lottery, annual reports 2018–2025.
$6.7B → $9.8B
Total ticket sales, FY2017-18 to the FY2022-23 peak
~$32M
Advertising budget, held essentially flat the same years
#1
In total U.S. lottery sales, every year since FY2020-21 — including through the recent dip
+24.8%
Average annual marketing-ROI improvement, 2018–2024 — per Clarity i2

Sales, national rank and sports-betting timeline: Florida OPPAGA, "Review of the Florida Lottery," annual reports 20182025, each verified individually. Marketing-ROI and reallocation figures: Clarity i2, published Florida Lottery case study.

Why reallocation works at all

Every advertising plan has this shape.

Spending below a channel's response threshold buys almost nothing. Spending past its saturation point buys less and less for every dollar you add. The money that matters is in between — and where that line sits moves every time your mix, your competitors or the market does.

Insights finds that line for every channel and market you run, every planning cycle, and reallocates inside it.

The advertising response curve An illustrative s-curve of incremental revenue against advertising spend. Below a response threshold the curve is almost flat and the spend returns nothing measurable. Between the response threshold and the saturation point it climbs steeply — this is where a fifth to two-fifths of the budget does most of the work. Past saturation it flattens again and each additional dollar earns less than the one before it. Response threshold Saturation point The last dollars in still cost full price. Moving them left is the entire job. Advertising spend, low to high → Incremental revenue, low to high → No measurable return about a fifth of spend, in almost every plan Where reallocation lands a fifth to two-fifths, doing most of the work Diminishing returns the last dollars earn the least
Illustrative shape, not a specific client's curve — but every plan we have modelled has one, and the thresholds move every time the mix, the competition or the market does.
Executives seated around a boardroom table during a presentation.
This meeting happens every quarter. It is the most expensive hour on the calendar, and it usually ends without an agreed answer.

The real pain point this solves

The argument between marketing and finance ends here.

Marketing says the campaign worked. Finance says prove it. The CEO hears two different stories about the same quarter, and the CSO is left building a strategy on an answer nobody fully trusts.

Insights ends it with one true snapshot — not one number. It separates what advertising actually drove from everything else that moved revenue, splits that across channels, then down to every individual placement, and prices each one the same way: dollars returned per dollar spent.

So the CMO can point at a campaign running across four channels and show it returned more than it cost. Rank every campaign against every other. Say whether television beat radio last quarter, and by how much. Same accounting for all of it, so nobody in the room is arguing about the method any more.

What it will not do is tell you how a campaign nobody has run yet will perform. That is still the advertiser's art, and we would rather say so than sell you a forecast we cannot stand behind. Run it, and it gets measured on the same terms as everything else.

CEO

The headline without the caveats — what advertising returned, and which bets earned it.

CFO

A figure the model is structurally incapable of inflating — it won’t recommend a spend level it has never seen work.

CMO

Every campaign ranked by what it returned, in the units finance already trusts.

CSO

A plan built on evidence nobody upstream is going to relitigate next quarter.

What it is worth

Three more things change the quarter you start.

Drastically higher EBITDA

A 3–5% optimisation swing on an eight-figure ad budget drops straight to net earnings. You are not buying more media; you are stopping the waste.

Stronger corporate valuation

Predictable, mathematically governed customer acquisition compounds enterprise value. Marketing stops being a black-box gamble.

An unbiased referee

Agencies earn more when you advertise more. We earn nothing from your media plan — we only audit it.

How it runs

Three steps, every planning cycle. Not a loop — a spiral.

The same three steps run every cycle. What changes is where they start. Run them once and you have a plan; run them every cycle and it stops being a loop back to the same spot — Implement → Observe → Model → Optimise → Strategise → Repeat, each turn compounding on what the last one earned.

01

Refresh the raw data

Two years of weekly impressions and spend, by market, language, creative, campaign and daypart — matched against sales.

02

Model the causal relationships

Non-linear causal models learn how each placement moves revenue, and how placements move each other.

03

Reallocate to what performs

Set your guardrails. The engine reallocates within them and returns the plan in milliseconds.

How the engine runs, and why each turn starts higher A first pass rises through three steps — refresh, model, reallocate — and hands off into a repeating cycle of observe, model, optimise and strategise. The cycle is drawn as a revolution that does not close: it comes back around above the point where it entered, and the gap between those two levels is what the cycle earned. A dashed arrow shows the next turn beginning from that higher baseline. Revenue baseline → Time → First pass · days, not months Every planning cycle, from here on Higher baseline 01 Refresh 02 Model 03 Reallocate Implement Observe Model Optimise Strategise
Scroll the diagram sideways to follow the cycle → The turn never closes. It comes back around above where it started, and implements again from there — same four stages, higher baseline. That gap is the whole argument.

Whether it will work for you

We gate on data quality, not company size.

We work with companies spending enough, across multiple channels, for the math to find a signal — and small enough to still move fast on what it finds. What actually decides whether it works for you is the shape of your history, not your budget.

This works precisely because it doesn't need what's disappearing. Third-party cookies are failing, browsers are killing them outright, and privacy law keeps tightening. Insights was never built on tracking an individual — only on what a market actually bought. The less you can track, the more that matters.

Historical depth

Two or more years of continuous weekly spend records, so seasonal baselines are anchored.

Channel breadth

Active spending across three or more channels at once, so cross-channel effects can be separated.

Structured inputs

Clean record-level regional or product-line data, free of systemic tracking gaps.

What this looks like

Two years. Three channels that behave nothing like each other.

A wall of television screens with a hand holding a remote control.
Television A slow build. The spot keeps paying for weeks after it stops airing.
A blank advertising poster panel on a busy pedestrian street.
Out of home A burst when panels go up in a market, then a long, quiet tail.
A city at night overlaid with social media engagement counts.
Social Immediate and spiky — and the platform can change the rules mid-flight.

104 weeks of weekly spend and sales is enough history for the model to hold these three apart — the slow build, the burst, the platform shift — instead of blurring all three into one line and calling the average an answer.

Bring two years of weekly spend across three channels. We will show you what it earned.

A strategic briefing takes forty minutes.

Request a briefing