From AI Hype to AI ROI: What I Really Discussed on Stage

Notes from my panel at the Fashion Network x The Ecommerce Club event

On stage discussing AI ROI in ecommerce with fellow panellists

Last week I had the pleasure of being invited by Fashion Network and The Ecommerce Club to join a panel on one of the most debated questions in our industry right now: where is AI actually driving measurable growth in ecommerce, and where is it just noise?

Beyond the panel itself, it was a genuinely energising event. I met founders, marketers and fashion designers from across ecommerce and fashion, and the conversations in the room were refreshingly honest. Less "AI will change everything" and more "here is what actually worked, and here is what quietly failed."

Our chair, Davina Lines, structured the discussion around four questions. Here is my take on each of them, expanded with the data behind my answers.

1. Where has AI genuinely delivered bottom line impact, and where hasn't it?

My answer on stage was simple: the biggest bottom line impact so far has come from routine tasks, because automation is now easier than it has ever been.

Automation is not new. You could have automated your workflows five or ten years ago. But it was costly, time consuming, and it required specialist technical expertise for almost every step. That barrier has collapsed. You might still want technical advice for complex builds, but today a marketing team can turn content creation into a workflow, run data analysis and build dashboards, and conduct market research in hours rather than weeks.

And perhaps the single biggest impact of all: code writing and software creation. Tasks that once required a development team and a budget line are now accessible to anyone willing to describe what they want clearly. That is a genuine shift in who gets to build.

The industry data backs this up, and it also shows the other side of the coin. Around 88% of retailers have adopted AI in some form, but only about 7% have scaled it. That 82 point gap between doing AI and generating measurable earnings from AI is, for me, the defining statistic of this moment (McKinsey, Stord).

Where it has worked, the numbers are striking. AI driven recommendations contribute an estimated 25 to 35% of ecommerce revenue. Shoppers who engage with an AI assistant convert at roughly 12.3% versus 3.1% for those who don't. Walmart's AI demand forecasting saved $2.3 billion in inventory costs in its first year.

Where it hasn't worked is just as instructive. MIT's widely cited study found that 95% of enterprise generative AI pilots deliver no measurable P&L impact, and crucially, not because the models are bad, but because of poor integration and misaligned priorities. The same study found that most GenAI budgets flow into sales and marketing tools, while the highest measured ROI sits in unglamorous back office automation.

Klarna is the cautionary tale worth remembering. Its AI assistant genuinely handled two thirds of customer inquiries, but after cutting its workforce aggressively, satisfaction fell and the company went back to hiring humans. The CEO admitted cost had become too dominant a factor in the decision. The ROI case built purely on removing headcount turned out to be fake ROI, because it ignored quality and lifetime value.

The pattern across every failure is the same. AI fails where it is deployed as a substitute for people, for data quality, or for judgment. It succeeds where it is deployed as a multiplier on something that already works.

2. Which AI tools or use cases are building board level confidence?

Boards trust what they can tie to a number on the P&L. Right now, three things are doing that.

First, personalisation and AI assisted search. Visual search users convert more with meaningfully larger basket sizes, and AI leaders show higher revenue growth than laggards. Those are the comparisons that move a board from curiosity to budget approval.

Second, AI as a traffic source. Traffic to US retail sites from generative AI sources grew roughly incredibly year over year, and shoppers arriving from AI platforms are far more likely to purchase. When a channel goes from a rounding error to something the board asks about by name, confidence follows.

Third, agentic commerce. Amazon's Rufus assistant alone is estimated to drive $12 billion in incremental annual sales, and AI assisted sessions convert at roughly three times the rate of ordinary browsing. The agentic segment in retail is already valued at over $60 billion and projected to more than triple by 2031.

The honest caveat I shared on stage: confidence should not mean complacency. Payback typically takes two to four years, and only a small minority of firms see returns inside twelve months. Boards that expect quarter one miracles will kill good projects prematurely.

3. Who should own AI inside an ecommerce organisation?

This was the most contested question, and rightly so.

The default answer has been technical ownership. In most organisations the CTO or CIO leads AI strategy, with the CMO leading in only around a third of cases. But the performance data tells a different story. BCG found that the companies deriving the most bottom line value from AI are 50% more likely to have shared business and technology ownership, with clear decision rights on each side, rather than a single owner.

My view: ownership should follow accountability for outcomes, not job titles. A model I find useful splits it three ways. Technology owns whether the system survives production, meaning data quality, security and reliability. The commercial side owns whether it actually improves the customer experience and the P&L. Governance owns the boundaries of what AI is allowed to do.

The failure mode every operator in the room recognised: five AI initiatives running at once, and nobody accountable for the business result.

There is also a strong argument that marketing's role will grow, simply because marketing sits closest to the customer and the data. For those of us teaching the next generation of marketers, that is both an opportunity and a responsibility.

4. Looking ahead to 2026 and beyond, what should ecommerce teams stop doing with AI?

Big tech would tell you differently like use more tokens adn spend more money on AI usage etc. but:

• Stop measuring AI by activity instead of outcome. Usage statistics and pilot counts are not a business case. Margin, revenue and cost are.

• Stop treating AI as a fix for broken fundamentals. AI cannot repair a messy product catalogue, fragmented systems or weak merchandising. It just makes the mess visible faster.

• Stop deploying customer facing generative AI without guardrails. Air Canada was held legally liable for a policy its chatbot invented. Companies own what their AI says.

• Stop ignoring trust. Fewer than half of consumers currently trust AI generated shopping results. Pushing AI storefronts ahead of that trust curve risks brand damage rather than conversion gains.

• Stop confusing headcount reduction with ROI. Klarna's reversal showed that savings which destroy customer experience are not savings at all.

My closing line on stage, and I'll stand by it here: 2026 is not about doing more AI. It is about doing AI narrowly and well. The gap between adoption and scale is the story of our industry right now, and closing it will come from discipline, not from more pilots.

Thank you to Fashion Network and The Ecommerce Club for the invitation, to Davina Lines for chairing a genuinely substantive discussion, and to my fellow panellists and everyone who came to talk afterwards.

I'd love to hear from you: where has AI genuinely moved the needle in your business, and where has it quietly disappointed? Share your experience in the comments.

If you found this useful, subscribe to the newsletter for more evidence based takes on marketing, consumer behaviour and AI.