Creative Marketing Practices for July 7th
A paid media team launches forty AI-generated ad variations.
The platform identifies several winners. Click-through rate improves. Cost per lead falls. Attributed conversions rise. The campaign report looks good enough that the team increases the budget and adds another round of generated creative.
Three months later, the sales team has a different report.
Lead quality has softened. Several prospects misunderstood what the product did. Branded search absorbed more paid spend. Organic conversions declined as paid conversions rose. The company acquired more names while creating roughly the same number of qualified meetings.
Both reports can be accurate.
The platform can correctly report that its ads appeared before a set of conversions. Sales can correctly report that the business did not gain enough additional demand to justify the apparent improvement. The disagreement comes from two measurement questions being treated as though they were one.
Attribution asks where credit should be assigned.
Incrementality asks what would have happened without the marketing activity.
AI creative makes this distinction more important because marketers can now generate more variations, activate them faster, and let platforms optimize across a larger field of messages. That produces more data, more credited interactions, and more opportunities for a dashboard to look conclusive before the business impact is clear.
Attribution describes a path
Marketing attribution tries to connect a conversion to the touchpoints that preceded it.
A buyer might encounter a creator video, see an AI-generated display ad, ask an assistant about the category, search the company’s name, click a paid search result, return through direct traffic, and request a demo. Different attribution systems can distribute credit across those interactions in different ways.
Last-click attribution may credit paid search.
First-touch attribution may credit the creator video.
A multi-touch model may divide credit across several channels.
A platform-specific model may assign most of the value to the activity visible inside that platform.
Each model tells a story about the path.
None of them automatically proves which activity changed the outcome.
A June 2026 AdExchanger analysis of a proposed W3C advertising measurement standard criticized the tendency to describe attribution as evidence of advertising effectiveness. Its central argument was that observational pathway analysis should not be confused with causal proof of incremental business impact.
That distinction is easy to understand in theory and surprisingly easy to lose inside a campaign report.
Incrementality asks what marketing changed
Incrementality compares an observed outcome against a counterfactual.
The counterfactual is the result that would likely have occurred if the campaign, channel, creative, discount, or marketing intervention had not happened.
Suppose a campaign receives credit for 1,000 conversions. That does not mean it created 1,000 additional conversions. Some buyers may have converted organically. Some may have searched for the brand regardless of the ad. Some may have clicked a paid result because it occupied the easiest path to a company they had already decided to investigate.
Incrementality tries to estimate the difference between:
- What happened with the marketing activity
- What would have happened without it
That difference is the incremental lift.
The distinction matters most when channels overlap heavily with existing demand. A June 25, 2026 research paper titled Attributed, But Not Incremental argues that paid-attributed conversions can systematically overstate incremental growth when paid activity overlaps with organic demand, brand traffic, or other acquisition channels. The researchers propose using incrementality experiments as causal anchors to correct attribution estimates and report that a deployment across several TikTok markets was followed by an approximately 15-percentage-point reduction in measured cannibalization. The result is specific to that system and should not be treated as a universal advertising benchmark, but the measurement problem is broadly relevant.
A conversion can belong in the attribution report while adding little incremental value.
AI creative increases the amount of activity that can receive credit
AI creative systems can generate more copy, images, videos, campaign combinations, audience adaptations, and landing-page variants than most teams could produce manually.
This creates legitimate opportunities. A team can test more messages. It can respond to market changes quickly. It can explore different formats without waiting weeks for every variation. Smaller teams can run a broader creative program.
It also expands the amount of marketing activity competing to claim the same demand.
Consider a buyer who already knows the brand. They see three generated retargeting ads, click one, and convert. The ad platform records a conversion. The campaign appears successful. Incrementality asks whether the buyer needed the ad or whether the system placed itself near a decision that was already underway.
The same issue can appear in branded paid search. An AI assistant recommends the company. The buyer later searches for the brand, clicks the sponsored result above the organic listing, and submits a form. Paid search receives attribution. The original recommendation may remain invisible. The sponsored click may have protected the conversion, accelerated it, or simply charged the company for traffic it was likely to receive anyway.
Surface’s analysis of how AI search ads are moving faster than reporting describes why answer interfaces make these paths harder to interpret. AI-mediated discovery can shape awareness and consideration before the buyer enters a channel the analytics stack recognizes. The final click may be the clearest event in the data and the least important event in the decision.
A lower cost per lead can still produce a worse business result
AI creative systems are often evaluated through platform metrics:
- Impressions
- Click-through rate
- Cost per click
- Conversion rate
- Cost per lead
- Attributed revenue
- Platform ROAS
These metrics are useful for campaign operations. They tell marketers whether the system is finding responses efficiently inside the observable environment.
They can hide what happens after the response.
A creative angle may produce cheap leads because it makes an expansive promise. The promise can attract people who are curious, underqualified, or expecting functionality the product does not provide. The platform sees a form submission. Sales sees a conversation that never should have happened.
This is why teams should connect campaign reporting to lead quality rather than lead volume. Surface recommends looking beyond total leads and cost per lead toward lead-to-conversation rate, lead-to-meeting rate, cost per meeting, source-level conversion, sales acceptance, and pipeline outcomes. Those measures expose whether marketing activity is producing useful business motion rather than inexpensive database growth.
The message that wins the ad auction may lose the sales call.
Incrementality does not replace attribution
Attribution still has an operational job.
Teams need to understand which campaigns, pages, sources, and messages appear in converting journeys. They need to diagnose broken tracking, compare audience segments, optimize placements, and understand how people move through the conversion system.
An inbound marketing automation system also depends on path-level data. Marketers need to know which form, campaign, content asset, and use case brought a lead into the system so they can qualify, route, and follow up appropriately.
Incrementality answers a different level of question.
| Attribution question | Incrementality question |
|---|---|
| Which channel received credit? | Did the channel create additional demand? |
| Which ad appeared before conversion? | Would the conversion have happened without the ad? |
| Which campaign produced the lead? | Did the campaign improve total qualified pipeline? |
| Which creative had the best platform ROAS? | Which creative produced genuine lift after cannibalization? |
| Which page assisted the journey? | Did the page change conversion behavior? |
| Where should daily optimization occur? | Where should the company allocate additional budget? |
Attribution helps operators navigate the system.
Incrementality helps leaders decide whether the system is adding value.
Modern measurement needs several forms of evidence
No single model is likely to explain the entire buyer journey.
Marketing mix modeling can help estimate channel-level contribution over time, but it may be too broad for daily campaign decisions. Multi-touch attribution provides greater granularity, but privacy changes and incomplete tracking can weaken its reliability. A June 2026 research paper on Integrated Marketing Attribution proposes combining MMM-informed priors with channel-specific Bayesian attribution models to derive campaign-level effects from aggregated data. The framework reflects a wider measurement direction: marketers are trying to preserve actionable detail while grounding it in more robust channel-level estimates.
Most B2B teams will not build that exact model. They can still use the principle.
Measurement becomes stronger when teams triangulate several forms of evidence:
Attribution data
Shows the observable path, campaign interactions, credited sources, and conversion events.
Business-quality data
Shows lead fit, conversations, meetings, sales acceptance, opportunities, pipeline, retention, and expansion.
Incrementality experiments
Use holdouts, randomized tests, geo experiments, audience suppression, or matched markets to estimate causal lift.
Marketing mix or modeled contribution
Estimates broader channel impact where individual journeys cannot be observed cleanly.
Brand and demand signals
Includes branded search movement, direct traffic, share of search, message recall, returning visitors, and category consideration.
Qualitative evidence
Includes sales-call language, customer interviews, buyer questions, creator feedback, and recurring objections.
The methods will disagree occasionally.
That is healthy. A measurement system should reveal uncertainty rather than forcing every source into one confident number.
Report observed, inferred, modeled, tested, and unresolved results separately
AI-era reporting needs clearer evidence labels.
Surface’s new brand measurement layer argues that AI visibility is becoming measurable before it becomes cleanly measurable. A buyer can encounter a recommendation, later search the brand, click an ad, and convert without leaving a complete trace of the original influence.
The same evidence framework works for AI creative and paid media.
Observed
The event appears directly in the available data.
Examples:
- The campaign generated 300 attributed leads.
- Cost per lead fell by 18%.
- Forty leads booked meetings.
- AI-referred sessions viewed the pricing page.
- Sales acceptance increased for one message family.
Inferred
Several signals suggest a relationship, but the team has not proven causality.
Examples:
- Branded search increased after a creator campaign.
- Prospects began repeating language from a new creative concept.
- Direct traffic rose during an AI visibility campaign.
- A new message may have improved lead understanding.
Modeled
A statistical model estimates contribution under documented assumptions.
Examples:
- MMM estimates a range of channel-level incremental revenue.
- A calibrated attribution model reallocates credit across campaigns.
- A propensity model estimates expected conversion without exposure.
Tested
A controlled experiment supports a causal conclusion.
Examples:
- A geographic holdout showed incremental qualified pipeline.
- Suppressing retargeting for a control group did not reduce conversions.
- A message test improved lead-to-meeting rate rather than only form completion.
Unresolved
The available evidence cannot support a confident conclusion.
Examples:
- Attribution improved while total pipeline stayed flat.
- Branded search rose alongside several overlapping campaigns.
- Sample size was too small to detect useful lift.
- Tracking loss prevents a defensible estimate.
“Unresolved” is a useful business answer. It keeps a directional signal from hardening into a budget decision before the evidence can support it.
What marketers should test first
Incrementality testing can sound like a project reserved for companies with enormous budgets and specialized data-science teams.
Some tests do require scale. Others can begin with narrow, practical decisions.
Branded paid search tests
Where risk allows, use geographic or audience holdouts to understand how many branded paid conversions are genuinely incremental versus captured organic demand.
Retargeting suppression
Remove a portion of eligible users from retargeting and compare qualified conversion behavior. This can reveal whether the campaign creates lift or repeatedly claims people already returning.
Message-quality tests
Compare creative concepts using lead-to-meeting rate, sales acceptance, and opportunity quality. Avoid declaring a winner solely from click or form conversion.
Creator or expert campaign tests
Use matched audiences, pre/post branded search analysis, survey lift, or controlled paid amplification to estimate whether creator exposure changes awareness or consideration.
Landing-page proof tests
Add customer evidence, implementation detail, or clearer limitations to a test group of pages. Measure qualified conversion and sales understanding alongside page conversion.
Speed and routing tests
Test whether improved follow-up and routing create more value from existing demand before spending more to generate additional leads.
The best first test usually sits near a large budget decision, a disputed channel, or a metric that looks excellent while business results remain unclear.
AI creative needs a quality feedback loop
AI creative production should connect media performance to the rest of the funnel.
The feedback loop should move through five stages:
- Creative system: Which message and format were shown?
- Media response: Who clicked, viewed, or converted?
- Lead operations: Was the lead qualified, enriched, routed, and contacted?
- Sales response: Did the conversation progress? Did expectations match reality?
- Business result: Did the activity create incremental meetings, pipeline, revenue, or retention?
Most creative systems receive strong feedback from stages one and two. They receive much weaker feedback from stages three through five.
That imbalance encourages platforms to optimize for events they can observe, even when those events have a loose relationship to business value.
Marketers should send richer outcome data back where possible and review message performance outside the ad platform. Creative teams need to know which promises produce good customers. Paid teams need to know which cheap leads waste sales time. RevOps needs to know which source and message combinations move into pipeline. Sales needs enough campaign context to explain why expectations are breaking.
The marketing system becomes smarter when the definition of success survives the form submission.
Do not make the report more confident than the evidence
AI will create more marketing material.
It will also create more opportunities for attribution systems to award credit to activity that sat near a conversion. Some of that activity will create genuine lift. Some will capture demand already in motion. Some will protect existing demand. Some will add useful frequency. Some will cannibalize organic channels. Some results will remain unresolved.
The dashboard should preserve those distinctions.
Use attribution to understand the path.
Use lead quality to understand whether the path produced something useful.
Use incrementality to understand what marketing changed.
Use models where experiments are impossible, and state the assumptions.
Use experiments where the decision is important enough to deserve causal evidence.
AI creative can help teams move faster. Better measurement keeps that speed from carrying the budget in the wrong direction.