How ecommerce brands are using AI to compete on margins

Margin pressure is the defining commercial problem for ecommerce right now. Acquisition costs are up. Returns are eating fulfilment budgets. Promotional cycles have trained customers to wait for discounts. And the brands that built their models on growth-at-all-costs are finding that scale without profitability is just a bigger version of the same problem.

The instinct, when margins compress, is to cut. Cut headcount, cut spend, cut product range. That instinct is usually wrong - or at least incomplete. The brands that are actually defending and expanding margins are not doing it through austerity. They are doing it through better decisions, made faster, at a level of granularity that manual analysis cannot reach.

AI is the mechanism. But not AI in the abstract - specific applications, applied to specific problems in the margin stack. This article covers where those applications are delivering real commercial results, what good deployment looks like and how to assess whether your current setup is leaving money on the table.

The margin stack problem most brands misdiagnose

When ecommerce margins are squeezed, most leadership teams look at the obvious line items: cost of goods, shipping rates, return rates, paid media efficiency. These are real levers. But they treat each one as a separate operational problem rather than a connected system.

Consider a mid-size fashion retailer turning over £200m. Their marketing team is optimising return on ad spend in isolation. Their merchandising team is managing markdown cadences based on last season's data. Their logistics team is negotiating carrier rates. None of these teams is working from a shared model of what a unit of revenue actually costs to acquire, fulfil and retain. Each team is locally rational. The aggregate result is margin erosion that nobody owns.

This is the structural problem AI solves when it is implemented correctly: it creates a connected view of the margin stack, not just better point solutions. Before deploying any AI capability, the right question is not "which tool should we buy?" It is "which decisions are we making badly because we lack a connected, real-time picture of our commercial position?"

Dynamic pricing and promotion: where the gains are largest

Static pricing in ecommerce is a margin subsidy you are giving to every customer who would have paid more. It also means you are discounting for customers who do not need a discount to convert.

AI-driven pricing and promotion optimisation changes both equations. Instead of a blanket 20% sale event, machine learning models can score each customer's price sensitivity, predict their conversion probability at each price point and serve a personalised offer - or no offer at all - accordingly.

A consumer electronics brand in the £400m revenue range used this approach to restructure its promotional calendar. Rather than four major sale events per year anchored to retail conventions, they moved to continuous promotional optimisation across their customer base. Customers with high purchase intent and low price sensitivity saw no discount. Lapsed customers with high churn risk received targeted re-engagement offers. Average order value increased. Promotional margin drag decreased by a material amount in the first two quarters.

The underlying requirement is not a sophisticated AI platform - it is clean, connected data: transaction history, browsing behaviour, email engagement and return history at the customer level. Most brands of this size have this data. Most are not using it at this level of granularity.

Inventory and demand forecasting: the hidden margin killer

Excess inventory and stockouts are two sides of the same forecasting failure, and both destroy margin. Overstock drives markdown. Stockout drives lost revenue and, in some categories, permanent customer defection.

Traditional demand forecasting relies on historical sales patterns with manual adjustments for seasonality and promotions. This works reasonably well in stable conditions. It breaks down under demand volatility - and ecommerce demand is structurally volatile, driven by algorithm changes, viral moments, competitor actions and macro shocks that historical data cannot anticipate.

AI-based forecasting models incorporate a broader signal set: social trend data, search volume, competitor pricing, weather, promotional calendars and real-time sell-through rates. For a homeware brand selling through its own DTC channel and several wholesale partners, this kind of model reduced overstock levels by around 18% over a 12-month period while simultaneously improving in-stock rates on top-performing lines. That is both a margin gain and a revenue protection outcome.

The decision criteria for investing here are straightforward. If your average markdown rate exceeds 15% of gross revenue, or if you are regularly writing down end-of-season inventory, demand forecasting is almost certainly your highest-ROI AI investment. The margin recovery potential is direct and measurable.

Reducing return rates through better pre-purchase intelligence

Returns are the line item that ecommerce brands consistently underestimate in their true unit economics. Net Promoter Score tracks customer satisfaction. P&L tracks net revenue. Neither captures the full cost of a return: fulfilment, processing, repackaging, potential write-down and the customer acquisition cost that produced a transaction that generated no margin.

For apparel brands, return rates of 30–40% are not uncommon. At that level, returns are not an operational inconvenience - they are a structural threat to profitability.

AI can attack this problem at the point of purchase rather than after the fact. Fit prediction models, informed by body measurement data, purchase history and returns history, can reduce apparel return rates by 10–15 percentage points for customers who engage with them. Size recommendation tools trained on return patterns surface the right option before a customer buys two sizes and sends one back.

Beyond fit, AI-generated product content - more accurate descriptions, richer imagery, video - reduces purchase ambiguity for higher-consideration categories like furniture, beauty and consumer electronics. A £1bn home furnishings retailer that invested in AI-enriched product content saw a measurable reduction in "not as described" returns, which had been running at nearly a quarter of its total return volume.

The key principle: every point of return rate reduction flows almost directly to gross margin. It is one of the highest-leverage interventions available.

Building the commercial intelligence layer

The individual applications above - pricing optimisation, demand forecasting, return reduction - deliver value independently. They deliver substantially more value when they feed into a single commercial intelligence layer that leadership teams can actually interrogate.

Most ecommerce brands at the £500m–£1bn scale have a reporting environment that tells them what happened last week. What they need is a system that tells them what is happening now, why, and what the forward-looking margin implications are across their product portfolio, customer segments and channels.

This is what a well-configured BI and AI layer looks like in practice. Natural language querying against live commercial data. Automated anomaly detection that surfaces margin-dilutive patterns - a product category where return rates have spiked, a customer cohort where reorder rates are declining, a promotional mechanic that is cannibalising full-price sales. Scenario modelling that lets a head of ecommerce test the margin impact of a pricing decision before it goes live.

The technology to build this exists. The barrier is usually not capability - it is the absence of a clean, integrated data foundation underneath it. Brands that have invested in data infrastructure tend to find that the AI applications on top of it deliver returns quickly. Brands that have not find that every AI initiative stalls at the data layer.

What to do next

Margin is not recovered in one move. But it is recovered faster than most leadership teams expect when the right interventions are sequenced correctly and built on a coherent data foundation.

The brands that will pull away from competitors over the next two to three years are not necessarily those with the largest AI budgets. They are those that have correctly identified which decisions in their margin stack are currently made badly, and have deployed AI against those specific decisions rather than pursuing capability for its own sake.

If you are a head of ecommerce or growth leader and your margin position is under pressure, the right first step is a structured diagnostic of your current commercial data environment - not a vendor selection exercise, not a platform procurement. Understand the problem before you buy the solution.

Rodan runs focused diagnostic engagements designed exactly for this starting point. In a short, structured engagement, we identify where your margin stack is leaking, which AI applications would close those gaps and what your current data infrastructure can actually support. The output is a prioritised roadmap, not a slide deck of possibilities.

If that conversation is relevant to where you are, book a diagnostic with Rodan.