Every dollar gets a role, a signal, and a measurement rule.
Marathon Engine runs paid media as one system: four funnel roles with distinct optimization signals, attribution triangulated into a single operating view, and an evidence-led loop that moves budget as the data changes — not a fixed media plan.
Four funnel roles, each with its own optimization signal.
The stages are separated by the behavior we want to create. Each has a distinct audience state, optimization signal, and accountable read.
Build reach + brand opt-ins
Exclude social engagers, website visitors, and purchasers to reach people who have generally never heard of the brand. Optimize for efficient reach and platform-native actions.
Develop measurable intent
Exclude website visitors and purchasers. Keep social engagers eligible, then use a platform-side Marathon custom event to move awareness toward consideration — for example, a session returning from branded organic search.
Convert new customers
Exclude purchasers and optimize for purchase under incremental attribution. Run demand capture broad, or test separate prospecting and website-visitor retargeting campaigns.
Grow known customers
Run a distinct purchase campaign for repeat purchase, with no exclusions by default so delivery can concentrate among the highest-intent existing customers — measured under a conservative recognized-value rule.
Why the upper funnel exists at all.
Purchase campaigns naturally prioritize people already signaling intent. The upper funnel creates future demand among people the purchase algorithm is unlikely to select today.
Create memory before the category is urgent
Reach people who are not currently in-market at a materially lower cost than the last mile of conversion. The goal is familiarity: when the category becomes relevant, the brand is already mentally available.
- Earn the first memory among people purchase campaigns are unlikely to reach.
- Increase the likelihood of a later ad response without depending on it.
- Create the better outcome: branded organic search or direct traffic from people proactively seeking the brand.
Turn awareness into observable consideration
Keep social engagers eligible while excluding purchasers and website visitors. A platform-side Marathon custom event can optimize toward a high-value behavior such as a session returning from branded organic search.
- Move people from awareness into interest or consideration.
- Create a higher-converting audience that has remembered and actively searched for the brand.
- Start with the branded-search-return signal; other event optimizations remain later tests, not the default.
Funnel roles define what we measure — not a fixed spend mix.
The four roles clarify what each dollar is trying to do. The starting mix is set collaboratively inside approved guardrails; live marginal evidence determines allocation.
Top funnel
Net-new reach + brand opt-ins.
Mid funnel
High-value intent.
Bottom funnel
Incremental new customers.
Retention
Incremental repeat purchase.
Triangulating multiple attribution sources into one operating view.
The operating model does not require the team to abandon a trusted source. It makes each source's role explicit, applies an agreed weight or discount, and converts the result into a comparable decision input.
Use the agreed share
Incorporate the attribution view your team already trusts and recognize an agreed portion of its attributed revenue.
Weight last-touch evidence
Use last-touch analytics where it adds signal, with a multiplier or discount that reflects what the team knows about the channel and customer journey.
Recognize by campaign role
Use incremental attribution, standard platform reads, or a conservative recognized-value rule according to the campaign role and the evidence available.
Add modeled value
Layer in Marathon Brand Value, cohort and repeat-purchase evidence, or other modeled inputs where they improve the decision without pretending to be the only truth.
Weights are collaborative — and built to evolve.
You bring the historical context. Marathon translates the agreed view into explicit, inspectable weights, multipliers, and discounts.
Agree on what to recognize
Select the attribution sources and agree on the portion of each source that should count in the operating model.
Make every input explicit
Marathon records the channel, campaign role, source, metric, and agreed weight so the modeled return can be inspected.
Revise as evidence improves
Model trainers, campaign tests, new-customer growth, and contribution evidence can change how much of each source is recognized.
How we turn the framework into evidence-led allocation.
The funnel defines the jobs. This operating loop defines how we launch, learn, and move capital across those jobs.
Assign the role + weight
Set the customer state, optimization signal, attribution inputs, recognized-value rule, spend guardrail, and decision threshold.
Launch the learning matrix
Rebuild the hierarchy, load Testing and Winners, verify exclusions and event quality, then start the model trainers.
Connect platform + business signals
Read reach, event response, purchases, contribution, brand demand, and new-customer value together — using the agreed measurement model.
Move budget by evidence
Promote winners, cap weak cells, update the creative portfolio, and shift spend in controlled steps with human approval.
Start with testing + scaling — then let the business shape the structure.
This is an illustrative initial architecture, not a prewritten account recipe. We begin by protecting learning and proven scale, then add a split only when the brand's economics, catalog, content, promotion plan, or delivery needs make the business read materially better.
Testing + Winners / Scaling
Testing: protect new concepts, hooks, and formats from being crowded out. Scaling: give proven work stable delivery while the next ideas keep learning separately.
Business economics
Separate product, category, buyer persona, or usage occasion only when unit economics, LTV, inventory, or performance differ enough to manage independently.
Higher-SKU / dynamic product ads
Use a separate feed-based cell when a larger catalog and the available creative warrant it; this is an option, not the default for every business.
Content type
Separate influencer, static, brand-owned video, or organic content when spend concentration, fees, or distinct performance require a clearer read.
Promotion isolation
Keep promotional media separate while live only when its spend and impact need to remain independently readable.
Audience, customer state, or delivery control
Add audience, prospect-versus-customer, geographic, or other delivery structure only when it protects signal quality or creates meaningful operating control.
Test options that sharpen allocation decisions.
Where it makes sense, we use focused tests to challenge assumptions and gather additional information about how capital should be allocated.
Adjust initiative spend
Change the spend behind a campaign or initiative and read the effect on incremental new-customer acquisition and contribution — not only platform-reported efficiency.
Consider alternate signals
Test reach, video-view, or traffic campaign types when the question warrants it. These are experiments around the starting framework, not automatic additions to every account.
Estimate marginal incrementality
Increase brand-search spend, read paid and organic brand clicks together, and calculate cost per incremental click so paid clicks are not mistaken for clicks the brand would have earned organically.
Translate the click into a return
Compare cost per incremental click with paid-brand revenue per session after a heavy recognized-value reduction — to form a conservative directional marginal-return curve.
Revisit when the evidence changes
Brand-search dynamics can change by season, promotion, and competitive pressure. If the evidence changes materially, the question can be tested again rather than treating an old answer as permanent.
Want this system running your media?
If your paid media runs on platform-reported ROAS and a fixed media plan, there's usually contribution on the table. Let's find out how much.