The Creative Bottleneck in Ecommerce: Why Shopify brands need better workflow, not just better ads

A conversation with Colyn Montgomery from Tempo on AI, Meta ads, measurement, and what Shopify brands should actually automate next

For years, ecommerce teams have treated measurement as the hardest problem in growth.

That made sense.

If your purchase data is incomplete, your reports do not line up, and Meta or Google is optimizing against a partial picture of reality, every decision downstream gets harder. Teams end up debating dashboards instead of improving performance.

But something is shifting.

As more Shopify brands invest in cleaner infrastructure, server-side tracking, better identity resolution, and more reliable purchase signals, the constraint starts to move. The question becomes less about whether you can measure what happened, and more about whether you can generate enough useful creative variation to do something with the signal.

That was one of the most interesting threads in Edward Upton’s conversation with Colyn Montgomery, founder of Tempo.

Colyn’s view is that creative workflow is becoming the bigger challenge for many brands, especially as AI makes it easier to generate production-ready assets, test more angles, and move faster inside Meta. Not because measurement has stopped mattering. It still sits underneath everything. But because once the foundations are good enough, the bottleneck moves higher up the stack.

And that has consequences for brands, agencies, and the way Shopify teams think about AI.

The real promise of AI in ads is not infinite volume

When people talk about AI creative, the conversation often drifts toward extremes.

Either AI is going to flood the internet with endless variations of the same ad, or it is going to replace the whole creative function.

The Tempo conversation lands somewhere more useful.

Most brands are not producing too much creative today. They are producing too little. Testing is slow, expensive, and operationally heavy. Teams have to decide which products get the hero treatment, which angles get explored, and which audience segments are worth building around. A lot gets left on the cutting room floor long before performance data gets a say.

AI changes that.

It lowers the cost of getting from an idea to an asset that is close enough to evaluate, improve, and put in market. That matters because it gives teams more real options to work with. You can test more messages, more product angles, more audience-specific concepts, and more catalog depth than a manual process would usually allow.

That does not mean the right answer is infinite volume.

Colyn made an important point here: Meta does not need twenty near-identical versions of the same ad. If the differences are cosmetic, the system is likely to treat them as effectively the same. The opportunity is not more noise. It is more distinct concepts.

That is a useful reframe for Shopify brands.

The goal is not to flood the ad account with endless creative. The goal is to use data and AI together to generate better tests.

The hard part is deciding what is worth testing

This is where the conversation gets more interesting than the usual “AI can make ads faster” take.

Tempo’s approach is built around context.

That means connecting creative workflows to the systems where the useful signals already live: Meta performance data, Shopify product and sales data, catalog changes, inventory, and lifecycle tools like Klaviyo. The idea is that creative gets more valuable when it is grounded in what the brand is actually selling, what is already performing, what has changed, and what kind of audience or message is underexplored.

That matters because creative testing is not just a production problem. It is a decision problem.

Which variables should stay constant? Which ones should move? Are you exploring a new persona, a new story, a new product angle, or just another variation of something Meta has already seen ten times?

AI helps most when it can widen the set of plausible starting points without losing the commercial context.

That is also why the most practical use of AI in creative still feels collaborative rather than fully autonomous. As Ed pointed out in the conversation, one of the most useful things AI does today is give you five different angles, five different hooks, or five different directions to react to. The human still combines, edits, rejects, and sharpens. But they are doing that work with a much broader surface area of ideas.

That is a better mental model for ecommerce teams than “push a button, get a winning ad.”

Better data still matters because creative systems are downstream from it

None of this makes measurement irrelevant.

If anything, it makes the quality of the underlying signal more important.

Tempo relies on Meta’s own numbers and Shopify data as the working source of truth for performance. That is a strategic choice. If Meta is the platform allocating spend and deciding what to scale, then the most valuable number to improve is the one being fed back into Meta’s optimization system.

That lines up with Littledata’s view of the world.

For most Shopify brands, the first job is not building a prettier attribution dashboard. It is making sure the platform doing the optimizing has access to the cleanest possible purchase signal. If the purchase event is missing orders, duplicating conversions, or carrying inconsistent values, the algorithm is learning from bad inputs.

That is why measurement hygiene still matters so much in the mid-market.

Large brands can layer on lift studies, MMM (Marketing Mix Modelling), holdout testing, and more advanced attribution approaches. But a lot of Shopify brands do not yet have the scale, the geographic spread, or the internal resources to make those methods practical. They get more leverage from getting the basics right: server-side purchase tracking, durable identity, better match quality, and consistent data flowing into Meta and Google.

Creative systems sit on top of that foundation.

If the signal is weak, you can generate all the ads you want and still struggle to learn what is working.

The more interesting AI challenge is context, not just content

One of the smartest parts of the conversation was not even about ads. It was about data structure.

Colyn raised a problem that a lot of teams are only starting to appreciate: having data is not the same as having data in a form that an LLM can use intelligently. Raw events, disconnected attributes, and messy exports do not automatically become useful just because you put AI on top of them.

You need context.

Edward framed this in a way that feels very close to where ecommerce tooling is heading. The role of a data layer is not only to send events from Shopify to platforms. Increasingly, it is also to act as an intermediary between raw event streams and usable output. That might mean stitching journeys together, structuring customer context, or making sure the system understands what a purchase, a product view, or a market actually means.

This matters far beyond reporting.

If AI is going to help brands generate better creative, answer better questions, personalize better flows, or identify better tests, it needs more than access to data. It needs data that has already been made legible.

That is where a lot of the next wave of ecommerce infrastructure will get built.

Agencies are not disappearing. Their role is changing

Another useful theme in the discussion was what this means for agencies.

There is a lot of talk right now about software-enabled agencies, AI-native services, and the blending of SaaS with execution. Edward pushed back on the idea that agencies will simply become software companies. He has seen firsthand how different those business models are.

That distinction matters.

The more likely outcome is closer collaboration, not category collapse.

Brands still need partners who can stay on top of a market that is changing weekly. The models improve, the workflows shift, and what failed six months ago may work surprisingly well today. Most in-house teams do not have the bandwidth to retest every tool, prompt structure, and workflow every week. A specialist partner often does.

That is where agencies can create real value.

Less time spent reconciling broken reporting. More time spent deciding what to test, where to push harder, which angles deserve investment, and how to combine human judgment with machine leverage.

The strategic edge becomes less about manually producing every asset and more about designing better systems around data, experimentation, and decision-making.

Where this is heading for Shopify brands

Over the next couple of years, more brands will use AI inside creative workflows. That feels inevitable.

The bigger question is which brands will use it well.

The winners are unlikely to be the ones producing the most content for the sake of it. They will be the ones using AI to test sharper ideas, support more of the catalog, reach narrower segments profitably, and move faster without losing coherence.

For some brands, that might mean using AI to generate more distinct concepts around a hero product.

For others, it may mean finally giving long-tail products, seasonal collections, or niche audience segments enough creative support to find out whether they can work.

And for a smaller set of brands, especially those with stronger creative identities, the edge may come from deliberately resisting full automation and using AI more selectively in ideation and workflow support.

Read: Edward Upton says “The real question for brand leaders isn’t whether AI matters. It’s where it changes outcomes.”

Either way, the direction is clear.

As measurement infrastructure matures, the pressure shifts toward execution. Not just making more ads, but building a system that knows what to make, why it matters, and how to learn from the result.

That is a much more interesting challenge than volume alone.

Because the future of ecommerce creative will not be won by whoever can generate the most assets.

It will be won by the teams that can turn context into better decisions, and better decisions into tests worth running.