·
4 minute read
How to separate real-world impact from industry noise
AI has become one of the most talked-about technologies in retail — from product recommendations to generative content, from automated support to real-time pricing engines.
But how much of it actually drives value? And how much isn’t delivering measurable results yet?
As a company focused on operational impact, we’ve taken a closer look at how AI is being used across ecommerce today — what’s working, what’s not, and where the true potential lies. We gathered input from across our team to understand where AI is already delivering value in real operations and where it’s still falling short.
But how much of it actually drives value? And how much isn’t delivering measurable results yet?
As a company focused on operational impact, we’ve taken a closer look at how AI is being used across ecommerce today — what’s working, what’s not, and where the true potential lies. We gathered input from across our team to understand where AI is already delivering value in real operations and where it’s still falling short.
Personalisation and search are leading the way
Personalization at scale remains one of the clearest — and most effective — use cases for AI in ecommerce.
“AI helps personalize recommendations, content, and campaigns in real time. Done well, this lifts AOV, conversion rates, and retention without adding manual effort.”
— Paul Maguire, Business Development Manager
Search is another critical area. Smarter algorithms help users find what they’re looking for — even when queries are vague or contain errors.
Voice-powered product search, in particular, simplifies the user journey by removing the need for precise input and enabling more natural discovery.
“Interfaces are often overloaded. If AI can reduce interaction to a couple of clicks — or a voice command — that’s real value.”
— Aleksei Sinkov, Frontend Engineer
“For the user, AI should make it easier to find what they need — what fits, where, and how,” says Marat Bolatov, CEO of Cloud Retail. “And for the business, it’s about reducing the cost and complexity of getting that product in front of the right person. The real value is when AI shortens the path to the customer.”
Operations and forecasting: progress behind the scenes
While frontend tools get more visibility, some of the most valuable AI applications are happening in the background — in inventory planning, supply chain management, and replenishment.
“Over the past five years, I’ve seen real progress in how AI is used in supply chain and operations — particularly in forecasting, planning, and inventory. Tools that help predict demand, optimize stock levels, and automate replenishment are now becoming essential, especially as brands look to minimize waste and avoid costly stockouts.”
— James Wall, Brand & Communications Director
For operators dealing with large or seasonal assortments, these tools aren’t just helpful — they’re becoming the standard.
Smarter frontends, built on real user behaviour
Frontend applications of AI show promise — but the goal isn’t to replace design with automation. It’s to make interfaces more responsive to real user behavior.
“AI can make the interface feel more adaptive. Reordering menus, updating banners, and surfacing the right products at the right time — all based on how users interact.”
— Aleksei Sinkov, Frontend Engineer
Key opportunities include:
- Real-time product recommendations that adapt as users browse
- Dynamic category menus that reflect interest and time of day
- Autocomplete and search hints based on behaviour
- Predictive reordering prompts for previously purchased items
- Smart banners that update by context (evening, weekend, location)
For retailers with thousands of SKUs and frequent product turnover, simplifying navigation and surfacing relevant items is critical. This is where adaptive interfaces can reduce bounce and speed up conversions.
These are the kinds of subtle optimizations that improve UX without overwhelming it. But as Aleksei noted, too much personalization without transparency can create friction: “Users should understand when AI is making decisions for them. Otherwise it feels unpredictable.”
What isn’t working (yet)
In practice, many AI initiatives still fall short of expectations — especially when not grounded in operational needs.
- 3D shopping assistants often feel gimmicky and slow, with limited real impact on conversions
- Automated product descriptions still require editing — they’re fast, but often lack nuance or accuracy
- Complex agent workflows remain difficult to scale and break easily when business logic changes
“AI features can't be considered in isolation — we need to start with the business process and see if AI simplifies it.”
— Internal, Engineering
Marat Bolatov, CEO:
“If used carelessly, it creates more work — not less. AI doesn’t remove the need to think — it increases the need to think clearly about how it’s applied.”
Otherwise, they risk solving problems that don’t exist — or adding more layers of complexity instead of removing them.
Where the opportunity lies
Some of the most effective applications of AI aren’t user-facing. They’re in the backend — addressing the operational friction that slows teams down.
Setup is a typical example: misformatted files, inconsistent data, and small errors that escalate into delays. Applied correctly, AI can reduce that manual burden and keep implementation on track.
Across our engineering team, one pattern kept coming up: AI works best when it removes friction from already existing workflows. Language models, in particular, are proving helpful in places like onboarding, support, and documentation — where users often get blocked by formatting issues or unclear steps.
“Right now, the biggest use case for AI is in Large Language Models (LLMs),” said Aleksei Kolchanov, one of our backend engineers. “And we need to look at cases where they can make life easier — like support, knowledge bases, onboarding, and working with text data.”
That same logic applies elsewhere in the stack. Teams flagged routine friction points that AI could meaningfully reduce — like broken product imports, courier matching errors, or messy data uploads that stall configuration.
In onboarding and setup, AI can reduce the effort required to resolve routine questions, clean up data issues before they create delays, and surface the right documentation at the right moment. These aren’t headline features — but they remove day-to-day friction that slows teams down.
That’s where we see the real opportunity: not in adding complexity, but in making existing processes faster, more reliable, and easier to navigate.
Setup is a typical example: misformatted files, inconsistent data, and small errors that escalate into delays. Applied correctly, AI can reduce that manual burden and keep implementation on track.
Across our engineering team, one pattern kept coming up: AI works best when it removes friction from already existing workflows. Language models, in particular, are proving helpful in places like onboarding, support, and documentation — where users often get blocked by formatting issues or unclear steps.
“Right now, the biggest use case for AI is in Large Language Models (LLMs),” said Aleksei Kolchanov, one of our backend engineers. “And we need to look at cases where they can make life easier — like support, knowledge bases, onboarding, and working with text data.”
That same logic applies elsewhere in the stack. Teams flagged routine friction points that AI could meaningfully reduce — like broken product imports, courier matching errors, or messy data uploads that stall configuration.
In onboarding and setup, AI can reduce the effort required to resolve routine questions, clean up data issues before they create delays, and surface the right documentation at the right moment. These aren’t headline features — but they remove day-to-day friction that slows teams down.
That’s where we see the real opportunity: not in adding complexity, but in making existing processes faster, more reliable, and easier to navigate.
Final take
In eCommerce, the most valuable uses of AI are often the least visible. Not the splashy features — but the quiet fixes that reduce manual work, prevent delays, and help teams move faster.
Throughout our work, we’ve seen AI create real impact when it’s applied to the right problems: improving search relevance, streamlining onboarding, fixing broken imports before they escalate, or making documentation easier to access.
Throughout our work, we’ve seen AI create real impact when it’s applied to the right problems: improving search relevance, streamlining onboarding, fixing broken imports before they escalate, or making documentation easier to access.
