The first generation of ad creative automation tools did something specific and useful: they reduced the human time required to produce creative variants for paid advertising. A tool like Pencil, which we backed in 2023, took the brief-to-asset production cycle from days to hours for a performance marketing team running significant paid social volume. That was enough to drive real adoption and real revenue. The category established itself.
The next generation of ad creative automation will need to do something harder. The efficiency gains from first-generation tools are now table stakes. The differentiation question for the next two years is not "can your tool produce creative faster?" It is "can your tool produce creative that performs better?" That is a structurally different technical problem, and the approaches that will win are not obvious.
Why efficiency gains are no longer sufficient differentiation
The performance marketing teams that adopted first-generation automation tools fastest have now been using them for 12 to 24 months. The workflow efficiency gains are real but they have been absorbed. The teams have restructured around the new throughput — more tests per week, more channel coverage, more variant rotation. The incremental efficiency gain from switching to a competing tool is no longer compelling enough to drive churn.
What these teams are now asking for is performance improvement. They want tools that tell them what creative will perform before they spend money testing it. They want tools that learn from their campaign performance data and use that learning to improve the next generation of creative. They want tools that can predict winner variants from a batch rather than requiring every variant to be deployed before knowing which one works.
These are harder problems. They require an analytics and prediction layer on top of the generation layer that most first-generation tools did not build. The ones that are building it now are doing so at an accelerated pace because the market signal is clear.
The three architectures competing for the performance-optimization prize
We see three distinct architectural approaches to ad creative performance optimization in this next wave, and we have views on their relative merits.
The first is what we call the generation-evaluation loop: a tight cycle between creative generation, performance prediction (using a model trained on campaign performance data), and iterative generation guided by that prediction. The appeal is that the prediction model gets better the more campaign data it processes, creating a compounding advantage for platforms that accumulate more performance-labeled data. The challenge is that building a reliable prediction model requires very large volumes of labeled performance data — specific creative attributes correlated with specific performance outcomes across channels and verticals — that most platforms are still accumulating rather than having fully accumulated.
The second is the multimodal brief-to-asset approach: a system that takes a brand brief as input — including product information, audience definition, channel context, and performance objectives — and generates a fully specified creative package rather than individual assets. The package includes copy, visual composition guidance, and format specifications calibrated for the specified channel. The appeal is that it addresses the creative coordination problem — the fact that high-performing ads typically require visual and copy elements that are designed together, not assembled from separately generated components. The challenge is that this approach requires more sophisticated multimodal modeling and is harder to integrate into existing production workflows.
The third is the performance database approach: a tool that provides access to aggregated performance intelligence — what creative patterns are currently working in specific categories, channels, and audience segments — and uses that intelligence to guide generation toward patterns that are likely to perform. This is essentially a market intelligence layer that informs creative decisions. The appeal is that it provides value even when a team's own campaign history is limited. The challenge is maintaining the freshness and specificity of the performance database, which requires ongoing data acquisition at a scale that is technically and commercially demanding.
What we look for in the next wave of investments
We backed AdCreative.ai in 2025 as part of our interest in the generative ad platform architecture. When we evaluate new companies in this space, the questions we focus on are different from what we asked three years ago.
Can the platform demonstrate performance improvement rather than just efficiency improvement? We want to see evidence that teams using the tool are seeing better campaign performance outcomes — not just faster creative production. This requires longitudinal data, which most pitching companies do not yet have, but the ones that do are compelling.
Is the compounding data advantage real and proprietary? A performance prediction model trained on the platform's aggregate campaign performance data is only as valuable as the quality and breadth of that data. Platforms that have been accumulating performance-labeled data across a wide range of verticals, channels, and geographies are building something that is genuinely hard to replicate. Platforms that are still in the early stages of data accumulation are making a bet on their ability to get there.
Does the product embed deeply enough into the workflow to capture the feedback loop? A tool that generates creative assets and then disconnects from the deployment and performance measurement process cannot close the generation-evaluation loop. The platforms that are building native integration with ad platforms — direct performance data ingestion from the platforms where the ads actually run — are building the infrastructure for the compounding loop. The ones that require manual data export and re-import are creating a friction point that will prevent the loop from being tight enough to be useful.
What the category looks like in 2028
We think the ad creative automation category will consolidate significantly over the next two years. The first-generation efficiency tools will face pressure from below — improved native AI capabilities in the ad platforms themselves — and from above — integrated marketing operating systems that include creative as one component of a broader optimization layer. The platforms that will survive and grow are the ones that have accumulated genuine performance data advantages and have built the closed-loop architecture that translates that advantage into better outputs over time.
That consolidation is not a negative indicator for the category — it is a sign of maturity. Mature categories consolidate around the platforms that have earned structural advantages through genuine capability differentiation. The teams that built the efficiency layer first are well-positioned to build the performance layer next, if they have the data to do it. The teams that are starting from the performance layer are making a bet that they can accumulate the data faster than incumbents can build the prediction capability. Both bets are live.