I know after more than three decades managing imaging teams at companies including Gap, Amazon, and Wolt, a subsidiary of DoorDash, from film through to digital and now AI, one pattern has remained consistent. The companies that win are not the ones with access to the best tools, they are the ones that agnostically and surgically use them to their advantage.
The same is now true in ecommerce, wherein the cost of producing product visuals is falling rapidly, and AI tools are increasingly available across the market. This means access to technology is no longer a differentiator, with the competitive advantage shifting into how that technology is embedded into production systems. As a result, the constraint has moved away from the ability to create images and towards the ability to deploy them effectively at scale.
As production costs decline, content volume rises, bringing more sellers into the market and increasing the number of listings competing for attention; visibility becomes harder to secure, while conversion becomes more difficult to sustain. At the same time, consumer expectations remain high, with product images continuing to play a defining role in purchase decisions, widening the gap between the volume of content and the quality required to drive trust.
The Shift Is Operational, Not Technical
Much of the discussion around AI continues to frame it as a creative or technological shift, when in reality, it is more of an operational influence. The operational argument holds, but it is important to distinguish the role of AI models trained specifically on e-commerce product imaging. At Photoroom, over one billion images perform materially differently from general-purpose models on the dimensions that matter most to commerce: background accuracy, object fidelity, and consistency across catalogues. We are agnostic, using all AI models along with our own to get the best fit for the use case, but specialist AI outperforms general AI in this context, and that gap is visibly apparent.
In many cases, what appears to be a limitation of AI is actually a limitation of infrastructure, with underlying systems and workflows often failing to support effective deployment at scale. Fragmented workflows prevent organisations from realising the value of the tools they already have, because AI only delivers value when it is embedded into how production actually happens, rather than applied as a bolt-on layer on top of existing processes.
This gap is already visible and widening fast, as adoption struggles to keep pace with the speed at which digital commerce is scaling, and expectations are being set. While a majority of organisations remain in experimentation or pilot phases with AI, consumer expectations are already established, with product visuals remaining one of the most important factors in a purchase decision. The issue is not future capability, it is the current execution.
Where Scale Exposes Structural Gaps
At scale, the challenge has never been creating images, but maintaining consistency, quality and speed across complex workflows. From my personal experience running multiple studios globally with millions of images processed monthly, this is where systems begin to fail. Small defects do not remain isolated, as they compound across products, categories and markets.
This is particularly visible as ecommerce continues to expand across marketplaces and platforms, bringing more sellers into increasingly competitive and standardised digital environments. Smaller sellers are now operating in the same environments as large enterprises, but without the same production infrastructure. The ability to produce and deploy images quickly is becoming a key differentiator, regardless of size.
What Effective Production Looks Like in Practice
When production is properly structured, the impact is immediate, with improvements visible across speed, consistency and overall output quality from the outset. Retailers such as Decathlon have demonstrated how embedding structured workflows can reduce production timelines while improving consistency, allowing products to go live faster and respond more effectively to demand. The same discipline at scale is reflected in marketplace environments, where platforms like Wolt, operating across more than thirty countries, report an error rate of less than 0.1% across images processed each month, underlining how standardised production directly translates into accuracy, reliability and trust.
More broadly, the cost of image production is no longer the primary constraint, as advances in technology and workflow efficiency have reduced barriers to creation across the market. The greater commercial risk lies in delay and loss of customer trust. When production cycles cannot keep up with demand, opportunities are missed, and the impact shows up in conversion, revenue and inventory performance.
From Standalone Tools to Embedded Infrastructure
The distinction between using AI as a tool and embedding it as infrastructure is now critical, as more organisations move beyond isolated use cases toward system-wide implementation. When applied one image at a time, it produces outputs without changing the system. When integrated into workflows, it enables a one-to-many model, a flywheel where a single input can generate multiple outputs, variations and formats without adding complexity while retaining brand consistency.
This is what allows production to scale, as structured workflows and integrated systems create the conditions for consistent, repeatable output at volume. It also changes the role of creative teams, moving focus away from repetitive execution and towards setting standards, better creative direction and high visual quality at scale. This is what visual commerce infrastructure means in practice, not a tool applied to images, but a system embedded into how commerce operates.
Trust as the New Constraint
As content volume increases, trust becomes the limiting factor, with consumers navigating a growing number of listings, formats and sellers in increasingly compressed decision windows. Customers are comparing products across multiple sellers and platforms in seconds, and in that environment, inconsistency in a merchant’s visuals quickly signals risk and reduces confidence.
In commerce, the goal was never a beautiful image, it was an image that sells, and while that principle remains unchanged, the challenge now lies in delivering it consistently at scale. That conviction hasn’t changed the ability to deliver it consistently, but scale has. When images are inaccurate or inconsistent, hesitation increases, conversion decreases, and return rates often follow. Poor images do not just reduce conversion, they increase returns, which remains one of the fastest ways to erode margin in ecommerce. The inverse is equally true: GoodBuy Gear, a recommerce platform, recorded a 23% uplift in conversion after embedding structured visual production into its listings workflow.
In practice, effective visual stimulation is built on three fundamentals: fidelity, realism and photographic integrity. Products must be represented truthfully, placed naturally within their environment and adhere to the physical rules of light and perspective. When those conditions are not met, the impact is commercial rather than cosmetic.
Rethinking Competitive Advantage
Taken together, these dynamics point to a broader shift within ecommerce, as advances in tools and accessibility continue to lower the barrier to content creation across the market. The ability to create content is becoming commoditised, while competitive advantage is increasingly defined by the ability to control, standardise and deploy that content effectively at scale.
This requires systems, not tools, as businesses look to move beyond isolated solutions toward integrated, repeatable production environments that can operate at scale. Platforms such as Photoroom are increasingly being embedded into production workflows to remove friction and enable consistent output, but the broader trend extends beyond any single provider. It reflects a structural shift in how ecommerce operates.
The greater risk for many organisations is not a lack of access to AI, but a failure to adapt their operating model. From what we have seen across previous technology shifts, once the gap opens between those who have embedded these systems and those who have not, it compounds quickly, with those slow to commit losing traction.
In a market where content is abundant, and where the barriers to creation have been significantly reduced, performance is no longer determined by access. Performance is determined by execution.
