What a computational neuroscientist at Rithum taught me about AI, the economics of small improvements, and why the next generation of commerce might depend on the most unglamorous work in retail.
There is a particular kind of ecommerce conversation that makes me nervous.
It usually begins with somebody explaining how AI is going to revolutionise everything, progresses to a slide showing an exponential curve, and ends without anybody explaining how a retailer is supposed to make more money.
I have sat through enough of them to develop a mild allergy.
Which is why my conversation with Gregor Kiddie was such a welcome change.
Gregor is Senior Engineering Manager for Strategic Intelligence at Rithum. He’s been with the business for more than sixteen years, stretching back to the ChannelAdvisor days. Before that, he was a computational neuroscientist, developing mathematical models of how brains grow. Some of his research supported Alzheimer’s work in Amsterdam; some has subsequently been referenced by Neuralink.
That’s quite a journey to end up discussing product descriptions and returns rates with me.
But it makes a peculiar sort of sense.
We used to grow by turning up
Gregor said something early in our conversation that has stuck with me.
Ten years ago, you could practically breathe in the direction of a new ecommerce channel and enjoy double-digit growth.
He’s exaggerating, of course. But not by much.
The expansion of marketplaces created extraordinary opportunities. More channels, more customers, more transactions. For a long time, distribution itself was a meaningful competitive advantage.
That world is changing.
The channels are crowded. The operational costs are substantial. Customers move between platforms with little concern for our carefully constructed channel strategies.
The growth isn’t simply waiting for us to arrive anymore.
So where do we find it?
Gregor’s answer was not a new marketplace, an enormous AI transformation programme or another expensive dashboard.
It was a collection of small improvements.
The one-percent problem
We talked about Dave Brailsford and British Cycling: the famous idea that improvements too small to matter individually can become extraordinarily powerful when combined.
Better sleep. Better hygiene. Better recovery.
Not particularly glamorous.
And rather less exciting than announcing a revolutionary new cycling technology.
Gregor sees a parallel in commerce.
Take returns.
A retailer may have thousands of products across multiple categories, countries and marketplaces. Somewhere in that catalogue are products with confusing descriptions, inadequate sizing information, misleading titles and subtle mismatches between what the customer expects and what arrives.
Each problem may be small enough to escape attention.
Collectively, they’re eating the margin.
A human analyst can investigate the biggest problems. But asking somebody to examine every listing, every return pattern and every market-specific variation is an enormous undertaking.
Machine learning can look for patterns across a much larger body of information and identify opportunities that would otherwise go unnoticed.
A sizing correction here. A better product description there. A recommendation not to list a particular product in a particular market.
Gregor described how these small changes can accumulate into substantial improvements.
And that is a much more interesting proposition than asking a chatbot to write fifty product descriptions before lunch.
Because the objective isn’t more content.
The objective is fewer mistakes, better buying decisions and more profitable orders.
A bikini in France is a surprisingly useful business lesson
One example from the conversation deserves attention.
Gregor described a merchant expanding from Zalando Germany into Zalando France.
The obvious approach would be to take a successful German catalogue, translate it, upload it and start selling.
But Rithum’s analysis could identify products that were commercially attractive in Germany yet potentially unprofitable in France because of expected returns.
That changes the question entirely.
Instead of asking, Can we list this product in France?, the merchant can ask, Should we?
And the important part isn’t simply receiving a yes or no.
It’s receiving the reasoning.
An expected return rate, an estimated financial outcome, an explanation of why one product is worth expanding and another might not be.
I have spent a considerable part of my career helping brands expand into marketplaces and new territories.
The industry has become very good at discussing the mechanics of getting products onto a channel.
We have been considerably less enthusiastic about asking whether those products should be there in the first place.
Perhaps that’s where the next competitive advantage lives.
Your customers aren’t necessarily telling you the truth
This was another fascinating part of the discussion.
Gregor is sceptical of subjective returns data.
When a customer selects a reason for returning a product, that reason may reflect what is easiest to choose, what they think will secure a refund, or what they believe the retailer wants to hear.
It isn’t necessarily a reliable account of what happened.
Consider somebody ordering five sizes of the same garment, trying them all on and returning four.
The return form may say incorrect size.
Technically, perhaps.
But the real commercial problem is more complicated.
Gregor’s argument is that retailers should be careful about relying on those explanations when objective transaction and performance data can reveal different patterns.
This has implications well beyond returns.
We spend fortunes collecting customer feedback, reviews, service interactions and surveys. All of them can be valuable. But we also need to distinguish between what customers say and what their behaviour demonstrates.
The two are not always the same.
AI shopping is going to expose the quality of your product information
Eventually, our conversation moved into agentic commerce.
And here the implications become even more interesting.
For years, brands could get away with relatively basic product information.
A product identifier, a price, an available quantity and a halfway respectable description might have been enough to get a listing moving.
That isn’t going to be sufficient when customers start asking increasingly specific questions of AI shopping services.
What are the best trainers for somebody with wide feet?
Which products suit a particular use case?
What should I wear with a kilt?
That last example is Gregor’s own rather wonderful test prompt.
A conversational shopping system needs enough meaningful information to understand what a product actually is, who it’s suitable for and how it compares with alternatives.
That means accurate identifiers, attributes, dimensions, sizing, materials, descriptions and the context around their use.
In other words, the data work that ecommerce teams have been postponing for years.
Only now the consequences of postponing it may become much more visible.
Gregor described work exploring how AI shopping services select and recommend products, and how merchants might improve the information influencing those results.
He also made an important distinction: systems such as Gemini and ChatGPT do not necessarily behave the same way. The mechanisms change, and the ability to influence recommendations differs.
There is no magic formula here.
Which makes me even more suspicious of anyone already selling one.
The future may arrive in a spreadsheet
Perhaps my favourite observation from Gregor was about how enterprise customers actually want to use AI.
Not through another chatbot.
Not necessarily through an entirely new interface.
Through the processes they already rely on.
Spreadsheets. EDI. Reports. Existing integrations.
Gregor’s point was straightforward: if a retailer already receives useful operational reports through established workflows, adding better intelligence to those reports may create far more value than asking the team to adopt a new tool.
It is a deeply unfashionable idea.
Which probably means it’s worth listening to.
We often confuse innovation with visible change.
But for the person trying to run a retail business, the best innovation might be the one that quietly improves a decision without making Tuesday afternoon any more complicated.
Trust is not a product feature
Gregor also made an important distinction between established machine-learning methods and generative AI.
Predictive models built around defined inputs and repeatable processes are different from systems that can produce varying answers to similar questions.
For businesses making consequential commercial decisions, that difference matters.
Rithum has years of data and established customer relationships behind it. Gregor’s view is that introducing AI capabilities needs to preserve that trust rather than undermine it through change for its own sake.
It’s a useful counterweight to an industry that frequently rewards the loudest announcement.
And it left me with a thought.
Maybe the next phase of ecommerce won’t be defined by who adopts AI first.
Maybe it will be defined by who understands their own business well enough to use it intelligently.
Who knows where the margin is disappearing.
Who understands the relationship between product information and customer expectations.
Who knows which markets are worth entering — and which products should stay at home.
Who can tell the difference between a persuasive customer explanation and a genuine operational pattern.
Those are not new questions.
But we may finally have better ways to answer them.
And for an industry struggling to find profitable growth, that could be rather more valuable than another revolution.
Watch or listen to the conversation
Gregor Kiddie joins me on The Ostrich Report to discuss his journey from computational neuroscience to ecommerce intelligence, the economics of marginal gains, returns, product data, AI shopping recommendations and why enterprises might prefer better spreadsheets to another chatbot.
The Ostrich Report is presented in partnership with Rithum. My thanks to Gregor, Sammy and the Rithum team for supporting conversations that get beyond the headlines and into the mechanics of commerce.
Explore Rithum:
https://www.rithum.com/
Explore The V Spot:
https://www.thevspotnews.com/
Subscribe for more conversations about marketplaces, retail technology and the commercial realities behind the ecommerce headlines.