Tara Meyer
September 10, 2026
AI has made it possible to produce more ad creatives than most user acquisition teams could make a few years ago. One concept can be generated into multiple hooks, formats, voiceovers, localized versions, and AI-assisted variations within a day and overall, there’s a growing trend of increasing creative testing, ads, and harder decisions.
But as the quantity of ads increases, there’s another side to the story. Ways to test creatives also has to scale, especially if meaningful results are desired. In a recent episode of Tenjin’s ROI 101 podcast, Roman from Tenjin spoke with Mark, VP of Growth at Bidmatrix, about the evolving strategies to address the increase in volume and the amount of data teams have to interpret. Mark’s recommendations is simple:
“The same ad must have the same identity.”
He reckons that using a stable creative ID is the heart of a scalable creative testing process. When it comes to the naming conventions of many creatives, it’s important that they remain inclusive and share enough context for teams, measurement platforms, and AI assistants, so everyone can understand what was tested, how it performed, and what should happen next. There’s a lot that goes into a name.
The Move Towards Consistent Creative Data
Creative naming conventions are as simple and straightforward as it sounds. For example, when a team is running a small number of ads, a filename like puzzle_ad_v3_final.mp4 might be enough for someone on the team to recognize what it is. However, at larger volumes, this name structure says very little.
What changed in version three? Was the hook different? Was AI used to create part of the ad? Did the opening show real gameplay? Was there a fictional concept? What was different about the format? Was there a new creative concept introduced? These are the questions that matter once a team wants to understand why one ad performed differently from another. As Mark explains:
“To monitor the fatigue pattern, the same ad must have the same identity.”
And creative identity shouldn’t stop at the original creative file. Mark argues the same creative should stay recognizable throughout the workflow:
“The same ad should have the same identity in the file name, the ad platform, Tenjin or another measurement system, and the team’s report.”
For mobile growth teams, this creates a common reference point between creative production and campaign measurement. Instead of manually working out whether video_final_4 in a shared folder is the same ad as CR_238_US_A in a campaign report, the team has one stable identifier that follows the creative through the whole process.
What is a Stable Creative ID?
A stable creative ID is a consistent identifier that helps a team recognize the same ad across every tool used to create, run, measure, and analyze it. In theory, it is similar to an app’s Bundle ID. However, a stable creative ID carries structured information about the variables being tested.
Mark provides an example with these building blocks: creative ID, AI tool, hook type, promise match, format, aspect ratio, and version, which produces a structure like this:
CR48_AI_Midjourney_HK_FailFix_PROM_RealGameplay_UGC_9x16_V2
It looks detailed because it’s meant to capture information that may matter later during analysis. Mark’s advice, though, is not to make the system unnecessarily complex:
“For small teams, it doesn’t need to be even that complicated. You can start with just those parameters, with variables that matter most.”
The ID should reflect the questions your team wants to answer. If you’re primarily testing hooks, format, and how accurately an ad reflects the product, those may be the only attributes worth adding alongside the core creative ID. As your testing process matures, the structure can grow with it.
How to Explain Creative Performance
The value of a stable ID has less to do with length and more to do with the properties included. Mark uses several attributes in his example, but one is particularly relevant to mobile creative testing: the promise match.
The promise match is a way of distinguishing between creative concepts such as real gameplay, meta gameplay, fictional elements, or ideas where there may be a wider gap between what the ad promises and what the user actually experiences after installing.
It helps balance out top of the funnel performance analysis, which rarely tells the whole story. An ad may generate strong CTR or IPM simply because the hook grabs attention, and if the users it acquires then retain poorly or fail to create value, the team needs enough context to investigate why. Promise match adds that context. Without it, the data reads as:
CR48 → strong CTR → weak retention
With promise match built into the ID, it reads as:
CR48 → Fail/Fix Hook → Fictional Promise → strong CTR → weak retention
That doesn’t automatically prove the promise caused the retention result, but it does give the team more structured information to work with when deciding what to test next. As Mark explains:
“If we put this promise and differentiate that promise inside the creative ID, it will make it a lot easier for an AI agent to read the data, understand it, highlight possible winners, explain the reasons, and apply those reasons to future creative production.”
The point is to give people and AI enough consistent data to identify patterns worth investigating.
Creative IDs Aren’t For Humans Anymore
Traditionally, a creative name had two jobs: help a person recognize the asset, and keep reporting organized. But, things have changed and now there’s a third job: give machines structured context they can use when analyzing performance. Adding structured context made for primiarily for AI rather than human and manual reading can create friction within teams. Mark acknowledges it directly:
“From the start, it was hard getting UA to enforce this within our team…but, that’s the AI job, and the AI reads it very precisely and very carefully.”
As a best practice, teams should be creating a structural hierarchy for creative IDs that’s deliberately hard to understand and interpret, even if AI is able to hold more and process more information across hundreds and thousands of iterations.
Connect Creative Testing to Performance
Designing a structural hierarchy and nomenclature for your creatives is only part of the strategy. The next phase is analyzing creative performance by its layers:
“CTR and IPM could tell us whether the ad got attention and installs. Retention tells us whether the people were actually interested in the app. ROAS and LTV tell us whether the campaign made any business sense.”
For an MMP like Tenjin, this is where creative identity becomes especially useful. The goal isn’t simply to know that a creative generated an install. It’s to connect acquisition with what those users did afterward: whether they retained, completed key in-app events, started a trial, generated IAP, ad, or subscription revenue, and ultimately produced ROAS. A stable creative ID gives teams one more consistent dimension for connecting what was shown in the ad with what happened after the install.
Creative Fatigue, Tackling Decreasing CTR
The same approach becomes useful once a creative starts performing well, because winning an initial test doesn’t mean an ad keeps delivering the same value as spend increases. Mark uses an example from a lifestyle app: at roughly $1,000 a day in spend, a creative generated around 100 trials. When the team raised the budget to about $5,000 a day, trials increased, but not proportionally, producing roughly 200 trials rather than five hundred, with gains flattening further as spend kept climbing.
Mark’s point is that creative fatigue or saturation shouldn’t be reduced to one top-of-funnel metric:
“People often describe fatigue as a falling click rate when they look at CTR, but that is not the only version of fatigue.”
A creative can hold its appeal while becoming less efficient, which is why creative testing needs to continue after the initial winner is found. Teams can compare additional spend against incremental trials, conversions, revenue, or another relevant outcome, and watch for the point where more budget stops producing proportional value. The stable ID carries that continuity through the process, so the team always knows exactly which creative and variation it’s evaluating.
How MCP Fits In
Once creative data is consistently structured and connected to performance data, another possibility opens up: letting AI assistants query it directly. This is where Model Context Protocol, or MCP, enters the conversation. During the episode, Roman describes teams using Tenjin’s MCP for reporting and creative-analysis tasks, including asking which creative sets are showing signs of fatigue. Mark says Bidmatrix is already using this type of workflow:
“We work with MMPs a lot. We work with Tenjin, and we see that [MCP] facilitates a lot of the operation team’s routine.”
He goes further:
“I think every modern measurement tool should build a bridge between their data and the AI agents that could be processing those data on the advertiser side.”
The important part is the combination of three things:
- Structured creative data
- Downstream measurement
- AI access to that data
Without a clean, stable creative identity, the AI has less context to work with. If there’s a lack of downstream measurement, it may optimize around attention rather than user quality. If ways to query the measurement platform are missing, then teams end up moving data between dashboards, spreadsheets, and AI tools.
An MCP helps connect that last step, so a team could ask questions such as:
- Which creative variations are showing signs of fatigue?
- Which hook types have the strongest retention?
- Which ads improved IPM but underperformed on ROAS?
- Which promise types are associated with higher-value users?
- Which creatives should we review before increasing spend?
AI can help surface the answers, but the team still owns the decision.
How to Improve Creative Testing
Mark introduces a practical four-week framework for using AI in the creative testing process. It starts with measurement, not generation.
Week 1: Clean the Measurement
Mark recommends starting by stabilizing creative IDs, reviewing in-app events, simplifying reporting. This helps make sure the metrics used to evaluate performance are reliable. For mobile UA teams, this also means confirming the MMP is capturing the downstream events needed to judge user quality.
“Week one is about clean measurement. Stabilize your IDs, clear your in-app events, simplify views and styles, make sure that the metrics are accountable.”
Before generating more creatives, ensure you can measure its performance.
Week 2: Test Clear Creative Hypotheses
The second week moves into creative production. Rather than creating hundreds of minor variations, start with a smaller number of meaningful ideas or hypotheses. Mark recommends using AI to explore and develop those ideas, but the team still chooses what’s worth testing.
Week 3: Evaluate
The third week introduces two checks: does the ad attract people, and are the people it attracts relevant and valuable? Mark explains:
“An ad should pass both gates before receiving any major budget.”
The exact metrics depend on the app. For one team, that may mean IPM and day-one retention. For another, it might mean CTR and new trial starts, or install volume and an early monetization event. The principle matters more than the specific formula: don’t make a scaling decision on appeal alone.
Week 4: Review Fatigue and Feedback Loops
The final week is about iteration: review which creatives are saturating, which ideas should stop, and which promising concepts deserve another version, then use those findings to inform the next creative brief. Using Stable IDs makes that feedback loop easier because the team can connect each result with the attributes that were actually tested.
Creative Testing Success Start with Structure
Generative AI gives growth teams more creative options than ever. It doesn’t mean every variation needs to be tested, and it doesn’t mean teams need a complicated data architecture before they can begin. What they need is consistency: give each creative a stable identity, carry that identity through the ad platform, MMP, and internal reporting, add the attributes that matter to your testing strategy, and connect all of it with downstream acquisition and monetization data.
As Mark summarizes:
“With clean labels [and] clear quality checks, teams can improve the whole process with AI.”
AI workflows help teams analyze more data, identify patterns, monitor fatigue, and surface opportunities faster. But the principles and foundation should always come first. Building better AI-assisted creative testing strategies starts with gathering the right knowledge: what you tested, who it acquired, what those users did next, and how results implicate the next round of creative variations.