"What AI Actually Means for Ecommerce Right Now" - A Pulse Commentary

In the run up to the Pulse eCommerce Summit on the 13th - 14th May 2026, we are launching a series of commentary pieces on topics that will be a focus point at the conference, led by members of the Vervaunt team. Here is our commentary piece on the practical role of AI in ecommerce operations, with input from Barrett Ahern.

AI has become a standard feature of almost every ecommerce platform, tool, and service provider pitch. The challenge for brands is no longer whether to adopt AI, but how to distinguish between capabilities that will meaningfully improve their operations and those that amount to repackaged automation with updated branding. The technology is developing quickly enough that decisions made now will shape what brands can and cannot do over the next two to three years, which makes the quality of those decisions more important than the speed at which they're made.

This article covers where AI is creating genuine operational value for ecommerce brands today, where it introduces risk that requires careful governance, and how brands should be evaluating their readiness to adopt these capabilities in a way that compounds rather than creates technical debt.

Data quality and tech stack readiness as prerequisites

The most consequential AI work most ecommerce brands can do right now is foundational rather than visible. Before any AI-powered tool or capability can deliver on its potential, it depends on the quality of the data feeding it and the flexibility of the systems it connects to.

Product feeds and metadata sit at the centre of how catalogues get surfaced, whether by search engines, shopping platforms, recommendation systems, or increasingly by large language models responding to purchase-intent queries. If that data is inconsistent, poorly structured, or missing key attributes, the outputs of any AI tool built on top of it will reflect those gaps. The brands seeing the most immediate value from AI are using it to address this layer first: standardising data formats so products are comparable and searchable, ensuring tone of voice is consistent across product descriptions, and filling gaps in attribute data that affect discoverability across channels. This work is neither exciting nor novel, but it determines the ceiling for everything that follows.

"AI in fashion merchandising is only as good as the product data underneath it. Before merchants can benefit from automated looks, smart substitutions, or personalised stories, the foundational work of enriching and structuring product intelligence has to be done right. Ultimately, the return from AI is a direct reflection of the data and architecture it is built upon." - Antony Kattukaran, Tagalys 

Tech stack evaluation has also become more important than it was even 18 months ago, and the relevant questions have shifted. The ecommerce platform is the obvious starting point - Shopify has moved quickly with built-in AI capabilities and agentic functionality - but the platform layer is only part of the picture. The questions that matter sit further back in the stack. Can your ERP connect to agentic inventory management systems as they mature? Does your PIM support generative AI for catalogue enhancement? Are you locked into proprietary systems that prevent integration with external models or decision-making frameworks as those capabilities develop?

A proprietary tool that offers a useful automation today may be sufficient for current needs, but if it cannot connect to external models or agentic systems a year from now, the convenience of that tool becomes a constraint. Given the pace of development, optionality in your core systems is likely to be more valuable over the medium term than any individual AI feature available today.

"The real breakthrough for AI in ecommerce isn't a smarter chatbot - it's the interoperability it creates between systems and teams. For most mid-sized brands today, launching a campaign is still a fragmented game of telephone: manual briefs, endless emails, and disconnected tools. What's changing is the ability to connect your marketing platform, design tools, and project management into a single orchestrated workflow. Record a brief, have an LLM structure it against your template, route it through content review and design, and inject a ready-to-review campaign shell back into your platform - turning weeks of coordination into days. This isn't about picking up tools that sound impressive. It's about building connective tissue between the systems your teams already use, amplifying their knowledge, and making their work scale - whether that's a lean internal team or a brand working across multiple agency partners." - James White, Klaviyo

Understanding the distinction between AI and automation

There is a meaningful difference between automation and AI that is frequently blurred in ecommerce marketing, and the distinction has practical consequences for how much oversight a tool requires and how much trust brands should place in its outputs.

Automation is rules-based and predictable. An email flow triggered by cart abandonment, a price adjustment based on inventory thresholds, a workflow that routes orders to the nearest fulfilment centre - these are deterministic systems that do exactly what they are configured to do. AI, in the sense that matters operationally, involves systems that interpret data and make judgments rather than following predefined instructions. A recommendation engine that decides which products to surface to a specific customer, a pricing model that adjusts based on inferred demand segments, or a merchandising system that reorganises a storefront in response to real-time trend signals - these operate differently, with different risk profiles and different failure modes.

Many tools currently marketed as AI-powered sit closer to the automation end of this spectrum, sometimes with a language model layered on top of what is fundamentally a rules engine. That is not necessarily a problem - some of those tools are effective - but brands benefit from understanding what they are actually buying. The level of oversight required, the reliability of outputs, and the risk of unintended consequences all differ depending on where a tool sits on this spectrum, and over-investing in capabilities that are essentially sophisticated automation under a new label is a real and common mistake.

Where AI is creating measurable operational value

Setting aside the marketing claims, there are several areas where AI is producing tangible improvements in ecommerce operations today, and where the evidence base is strong enough to justify investment.

Reactive merchandising and inventory management. This is arguably the most impressive backend application in practical use. The ability to detect real-time demand signals - a product gaining traction on TikTok, a trend accelerating across social platforms, an unexpected spike in search volume for a particular category - and remerchandise a store in response within minutes rather than days represents a genuine step change in operational capability. A product goes viral on a Friday afternoon; historically, it would not be re-positioned prominently until someone on the merchandising team actioned it the following Monday. AI-driven merchandising systems can respond in that window, and the customer sees a store that feels current and relevant without understanding the mechanism behind it.

The same principle extends to inventory management more broadly. Systems that predict demand shifts, adjust reorder points dynamically, and flag potential stockouts before they materialise remove a significant volume of manual forecasting work. The value is less that these systems are smarter than an experienced merchandiser and more that they can process signals continuously across more data sources than any individual or team can monitor.

Customer service at scale. Customer service AI has become the most proven and widely adopted application in ecommerce, and for good reason. The technology has matured to the point where the majority of standard customer interactions can be handled effectively without human intervention, including handling complex queries, negotiating resolutions, and escalating appropriately when the situation requires it. We are approaching a point where a significant majority of customer service interactions across ecommerce will be AI-powered within the next 12 to 18 months. The adoption has been broad because the value proposition is straightforward - cost reduction and consistency at scale - and the failure modes are manageable in ways that customer-facing AI in other contexts often is not.

"The era of AI experimentation is over. For years, brands have "layered on" general-purpose tools and hoped for the best. In 2026, hope isn't a strategy.

The real winners aren't just "using AI", they are deploying agentic systems designed to solve the friction that kills margins: complex returns, cross-border logistics, and the "trust gap" at checkout.

At Reveni, we don’t just embed AI; we give it a job. Our tech lives inside the customer journey, resolving friction in real time. We’ve moved past answering queries to executing solutions, ensuring every customer interaction protects your bottom line while building loyalty.

Don't just adopt AI surgically. Use it to automate the un-automatable." - Kevin Paiser, Reveni 

Qualitative feedback aggregation and analysis. One of the more underappreciated applications, and one where the return on investment can be significant for brands with established customer bases. Ecommerce brands generate substantial volumes of qualitative data through reviews, return reasons, customer service transcripts, and post-purchase surveys, but extracting actionable patterns from that data historically required someone to read through it manually, which meant it was either under-resourced or not done at all.

AI can now aggregate and analyse this feedback at scale, identifying recurring themes in return reasons, surfacing product issues that affect specific segments, and understanding which aspects of the customer experience consistently drive satisfaction or friction. Represent is a brand doing this particularly well - they take post-purchase engagement feedback, reviews, return rationale, and sizing data, aggregate it through AI analysis, and feed the insights back into how products are described on site. The result is more accurate product information, which in turn reduces returns driven by mismatched expectations. They also incorporate sizing-related feedback to generate more accurate size charts, closing a common gap between what customers expect and what they receive.

Trend identification and capitalisation. Beyond responding to individual viral moments, AI systems can identify emerging trends and help brands position relevant products before those trends peak. The "quiet luxury" trend is a useful illustration - a merchandising concept that has existed for decades but became a searchable, categorisable phenomenon through social media. Brands with AI-driven merchandising could identify and respond to that trend automatically, remerchandising their catalogues around it without waiting for a buying or marketing team to notice and act manually. The competitive advantage comes from the speed of response rather than the quality of the insight, which is a pattern that repeats across most operational AI applications.

Areas that require governance before implementation

Not every AI application is straightforward to deploy, and several carry meaningful risk if implemented without adequate oversight and clear governance frameworks.

Generative imagery and brand representation. There has been notable customer pushback against brands using AI-generated product images and model shots without disclosing that they are AI-generated. The concerns operate on two levels. There is the broader question of what AI-generated imagery means for the creative professionals - photographers, models, designers - whose work it replaces. And there is a more directly commercial concern around representation: when brands generate model shots depicting people of different races and genders, customers are increasingly questioning whether that representation is authentic or exploitative, and what it communicates about the brand's values.

The practical advice is not to avoid AI-generated imagery entirely, but to be transparent about its use. The brands that have faced the most significant backlash are those that attempted to present AI-generated content as real. There are also models emerging for ethical implementation - 11 Labs, which works in AI voice generation, licenses real voice artists and pays them a percentage each time their voice model is used, which offers a framework other applications could draw from. The technology has clear production and cost benefits, but deploying it without a considered transparency policy creates reputational risk that can outweigh the efficiency gains.

Dynamic pricing. AI-driven dynamic pricing has genuine potential for margin protection, allowing brands to adjust pricing in response to demand signals without resorting to blanket discounts. However, the risk profile is significant and often underestimated. The concern extends beyond surge pricing in the conventional sense. AI pricing systems can segment customers by inferred characteristics - location, browsing behaviour, purchase history, and other signals - and adjust pricing accordingly, without the segmentation logic being transparent to the brand or the customer. There have been documented cases in financial services where AI-driven pricing effectively resulted in discriminatory outcomes based on customers' perceived wealth or demographics, using opaque categorisations that would not have been applied through traditional marketing segmentation.

Dynamic pricing is a powerful capability, but most brands have not yet built the governance frameworks it requires. Before implementing AI-influenced pricing, brands need clarity on what signals the system is using, whether those signals could produce discriminatory outcomes, and whether the pricing logic can be explained to a customer or regulator who asks. The operational benefit needs to be weighed against the reputational and regulatory risk, and for most brands that assessment has not been done rigorously.

Agentic commerce: understanding the adoption landscape

Agentic commerce - where AI manages the full discovery, selection, payment, and purchase journey on behalf of the customer - receives significant attention as the next phase of ecommerce, but the current adoption data suggests a more measured trajectory than the coverage implies.

In the US and UK, approximately 20% of consumers indicate they are comfortable handing over their details for a fully agentic purchase journey. Adoption is meaningfully higher in markets like China, Korea, Japan, and parts of Europe including Germany, where different consumer attitudes toward technology, privacy, and payment regulation create a more receptive environment. The gap is driven partly by privacy concerns around handing over payment credentials to AI systems, partly by regulatory differences in how payment transactions are governed, and partly by a lack of clarity on what happens when something goes wrong.

The post-purchase question is one that has received less attention than it deserves. If an AI agent purchases a product on a customer's behalf and the customer subsequently needs to return it, the process for doing so is unclear in most implementations. Does the customer contact the AI platform, the merchant directly, or some intermediary? Is the merchant's site trustworthy if the customer has never actually visited it? These are practical friction points in the customer journey that have not been resolved at scale, and until they are, adoption in privacy-conscious markets is likely to remain modest.

For brands, the practical implication is that agentic commerce warrants preparation rather than heavy investment at this stage. Ensuring that product data is structured for machine consumption, that APIs are accessible, and that fulfilment and returns processes can accommodate orders that did not originate on the brand's own website are sensible steps. But for brands whose primary markets are in the US or UK, this is more likely a 2027 or 2028 priority than a 2026 one.

The measurability question for LLM visibility

There is a growing industry around optimising brand visibility in large language model outputs - ensuring that when a consumer asks ChatGPT or Gemini for product recommendations, your brand appears in the response. Before investing in tools or services in this space, brands should understand a fundamental limitation of the technology as it currently operates.

Recent research from Sparktoro, founded by Rand Fishkin of Moz, has demonstrated that LLM responses are statistically worse than random in terms of repeatability. If you ask a large language model the same product recommendation question over a thousand times, you would struggle to get the same answer twice. The responses vary based on phrasing, timing, model version, and factors that are neither transparent nor controllable by the brand or the optimisation provider.

This has a direct and significant implication for any tool or service claiming to track and improve "LLM mention rates" or "AI search visibility" as a measurable, improvable metric. If the underlying outputs are not repeatable, measuring your position within them is not meaningful in the way that search rankings or share of voice in traditional media are meaningful. That does not mean LLM visibility will not matter over time - the trajectory of consumer behaviour suggests it will become increasingly relevant - but the current generation of measurement tools in this space should be approached with appropriate scepticism.

The more productive focus for most brands is on the fundamentals that drive discoverability regardless of channel: clean, well-structured product data, authoritative brand content, and genuine signals of brand relevance. These are the inputs that LLMs draw from, and they are the same inputs that improve performance across search, shopping platforms, and social discovery. Investing in those foundations will serve brands better than attempting to optimise specifically for outputs that are inherently variable and unpredictable.

Custom AI and the emerging frontier

Most ecommerce brands are still in the phase of evaluating and adopting off-the-shelf AI tools. A smaller number are beginning to build custom AI systems tailored to their specific operations, and this is where meaningful competitive differentiation is most likely to emerge over the next two to three years.

The backend is where custom implementations are appearing first, for practical reasons. The customer-facing risk is lower, the iteration cycle is faster, and the guardrails around data management are more controllable. Brands using flexible systems - Airtable-based ERPs, composable commerce architectures - are already implementing custom AI functionality for internal operations. The applications are varied, from automated inventory decision-making to custom reporting and analysis workflows, and while adoption is still early, the direction is clear. Brands that can build AI systems trained on their own operational data, optimised for their specific business context, will have advantages that generic off-the-shelf tools cannot replicate.

For fashion and lifestyle brands specifically, the highest-potential area for differentiated AI investment is probably visual commerce: expanding on AR try-on capabilities, 3D product visualisation, and AI-enhanced product representation that helps customers understand what they are buying before they commit. The business case operates on two fronts simultaneously. Conversion improves because the shopping experience is richer and more informative. Return rates reduce because the gap between customer expectation and product reality narrows. In an environment where tariffs and logistics costs are compressing margins across the industry, return rate reduction has a direct and measurable impact on profitability that can be as significant as any conversion rate improvement.

The technology to deliver these experiences exists, and has for some time. What has held back widespread adoption is implementation cost and complexity, both of which are declining as the tools mature and the supporting infrastructure becomes more accessible.

What to reassess before committing to AI investment

If AI capability is part of your ecommerce strategy over the next 12 to 36 months, these are the areas worth addressing before committing capital:

On data and infrastructure:

• Audit product data quality across feeds, metadata, and catalogue attributes. If this data is inconsistent or incomplete, it is the highest-priority investment regardless of which AI tools you are considering.
• Evaluate your core systems - ERP, PIM, ecommerce platform - for extensibility. Can they connect to external AI models and agentic frameworks, or are you locked into proprietary ecosystems that will limit your options as the technology develops
• Understand where your current tools sit on the spectrum between rules-based automation and genuine AI decision-making, and apply oversight proportional to the risk profile of each.

On implementation priorities:

• Start with backend applications where the risk is lower and the feedback loop is faster. Inventory management, merchandising automation, and data analysis are proven use cases with manageable failure modes and clear ROI.
• If you have not deployed AI-powered customer service, this is likely the highest-return starting point given the maturity of available solutions.
• Invest in qualitative feedback aggregation across reviews, return reasons, and customer service transcripts. The insight sitting in this data is underutilised by most brands, and AI makes it accessible at a scale that was not previously practical.

• If you are using or considering AI-generated imagery, establish a transparency policy before you face pressure to create one. Proactive disclosure builds trust; reactive disclosure after backlash does not.
• If dynamic pricing is on your roadmap, build governance frameworks before implementation. Document what signals the system uses, test for discriminatory outcomes, and ensure the pricing logic can withstand scrutiny from customers, press, or regulators.
• Approach LLM visibility and mention-tracking tools with appropriate scepticism until the measurement methodology matures. Invest in data quality and brand authority instead, which will serve you across all discovery channels.

• Identify where AI can reduce friction in the purchase journey - through better search, more relevant recommendations, or more accurate size and fit guidance - without introducing new friction points around transparency or trust.
• If agentic commerce is relevant to your market profile, audit whether your product data, APIs, and post-purchase processes are ready for orders that do not originate on your own website.
• Evaluate the business case for AI-driven return rate reduction through better product representation. In a margin-compressed environment, this may deliver more bottom-line impact than equivalent investment in conversion optimisation.

On competitive positioning:

• Recognise that AI is levelling the operational playing field. As baseline capabilities become table stakes, competitive differentiation will increasingly depend on the things AI cannot replicate: brand distinctiveness, creative quality, strategic judgment, and the quality of the customer experience.

• The pace advantage that AI creates is real, and it compounds over time. Brands that use AI to compress research and iteration cycles will accumulate learning faster than those that do not, but the goal should be faster iteration with human oversight rather than automated decision-making without appropriate guardrails.

Many of the themes explored here - from building the right data foundations and evaluating where AI creates real operational value, to managing governance risks and preparing for emerging models like agentic commerce - will be unpacked in far more depth at the Pulse eCommerce Summit on the 13th and 14th May 2026. Across two days, we’ll bring together senior eCommerce leaders, operators and specialists to share real-world experiences, practical frameworks and honest lessons from scaling brands internationally in a far more complex global landscape. If international growth is on your roadmap for 2026 and beyond, register now to secure your place.