Growth that compounds.
Costs that don’t.
Commerce still asks people to describe what they want. The best consumer experiences learn from what they do instead.
VISII is one engine, built to understand every product, every shopper and every session, and to let each understanding teach the others: search, recommendations, personalisation, merchandising and the catalogue improve as one.
Fifteen years of R&D and a decade of measured deployments keep arriving at the same conclusion: understanding has proved to be the highest growth lever we’ve found. The numbers below are theirs.
The numbers on this page: randomised split tests against matched control traffic.
Every leap in commerce has asked less of the shopper.
The feed learned her taste from what she does: TikTok Shop reached tens of billions in sales inside three years on exactly that.
Commerce has never reversed that direction: every step asked less, no step asked more. Once behaviour beats description for the shopper, it beats description for the merchant too.
The merchant’s columns, starting the same descent.
That shift has already started. It is what we're building for.
Growth is getting dearer. Standing still is dearer still.
Traffic
New traffic is fought for: auctions reprice, channels saturate, and AI answers keep clicks that once reached you. Your team fights it session by session; you meet it as acquisition taking a growing share of the revenue it brings.
Conversion × order value
Most shoppers can’t describe what they want; sites still ask them to. What your team logs as abandoned searches and stalled baskets, you meet as conversion that won’t move and traffic that grows faster than revenue. A million products mean nothing to a shopper with no words for what she wants.
The cost line
So the plan gets rescued with spend: discounts to buy the conversion understanding didn’t, headcount thrown at tagging and merchandising, and returns absorbed as a cost of doing business, ~17% of sales, $890B a year in the US alone. The rescues spend margin, and they don’t compound.
Sources: NRF / Happy Returns 2024; BoF State of Fashion 2026In the usual stack, each term above is defended by its own tool, learning alone. Intelligence strands between them. Nothing in the stack sees your number whole.
Built where correlation fails.
The engine was built to understand where correlation fails: marketplaces where every piece is one of a kind, and “others also bought” has nothing to work with. Fifteen years with that problem produced the intent model and an architecture that has never needed a name or an email. Understanding of that kind is not a feature added to a search bar; it is the foundation the whole engine is built on. If people can choose from a catalogue, the engine can understand it.
One engine. Zero silos.
Search, recommendation, personalisation, visual and generative AI, built to communicate with each other, so you don’t have to.
BEFORE, OVERLAPPING TOOLS, NONE INTEROPERABLE
Each tool learns something the others never hear, intelligence stranded at every link of the chain.
AFTER, EVERY SIGNAL INFORMS EVERY LINK
Measured in your analytics, not ours. VISII deployments are measured as randomised split tests: a test group against a matched control, inside your live traffic. Our case studies, and whether we get paid, stand on the same number.
A watch looks nothing like a shirt, but the shopper who loves one often loves the other. Correlation engines only see that after thousands of purchases; we see it from meaning, before the first sale. Models that teach each other can go further: the search bar already teaches the recommendations what your shoppers mean, and the engine is built to pass that lesson on to your catalogue.
Sees what shoppers mean
Deep learning reads style, pattern, texture and context straight from your product images, so nothing depends on tags you don’t have.
Reads the live session
The signals shoppers already give off, what they linger on, skip, and almost buy, with no names, emails or identifiers. An intent model, patented in the US and EU, turns them into a live preference profile per shopper, and knows browsing from buying as it reads.
Knows the whole catalogue cold
Popularity, reviews and what similar shoppers did next, working from the first click, even on brand-new products.
As close to reading minds as retail gets, without collecting a single identifier.
One engine connecting every lever of your P&L.
Not another tool on your stack: five commercial levers, one understanding behind them all.
Average order value
Every product page becomes a merchandiser that never sleeps: alternative styles, complete-the-look, mix-and-match: rails of intelligent suggestions composed automatically, personalised without collecting a single identifier.
Conversion
New revenue from the catalogue you already own: digital changing rooms, wordless browsing, back-catalogue alternatives: your long tail, finally selling. Even sold-out pages pay their way.
Engagement
When the rail isn’t quite right, your shopper doesn’t start over: she steers it. A word typed above the recommendations re-composes them around her journey.
Traffic
Generative tagging, read straight off the image, and long-tail pages that finally rank: the catalogue earns its own traffic, and the engine is built to inform the campaigns you still buy.
Costs
One platform, built to replace the siloed stack, labour that stops scaling with catalogue size, and the costs that scale with stock rather than sales: storage, unsold seasonal inventory, digitalisation debt.
And behind them all: capital freedom. When more of your growth comes from understanding than from acquisition spend, traffic and stock stop being the price of growth. They become choices, made when they pay.
One snippet or API · headless or traditional · works alongside your stack, or replaces it · needs only your product feed and images · any language, no dictionaries · start with one surface, expand when it pays.
Deploy and see results on any e-commerce platform in a matter of days, not months.
…or your custom platformResults.
What changed, and by how much, client by client.
Most vendors grade their own homework. The numbers here were read in the client’s own analytics, in randomised split tests against matched controls.
Hundreds of new products a week, merchandised by hand.
“Compared to our Einstein recommendations, Visii’s recommendations drastically increased the pages visited, time spent onsite, click-through rate and revenue.”
A catalogue where every piece is unique.
“Whatever your industry, Visii are the go-to guys to figure out the magic sauce for your sales. They drove ours higher by 25%, beating all competition and never compromise on customer privacy.”
The benchmark: their own in-house build.
“Visii solved a problem no one else could: offer highly relevant product alternatives and sets at scale to our marketplace of in-situ images visitors.”
Millions of artworks; taste refuses the box.
“After demoing several products, Visii’s was the best…” 47% of purchasers use VISII before buying.
The fair, and its visitors, had to move entirely online.
+80% AOV, >200% conversion, a marketplace outlier, well beyond the ranges we quote.
“…comb through 15,000 images of artwork in less than 350 milliseconds and easily find unexpected gems.”
Jaja Liao, Business Development, Google · 2018, on a VISII deploymentFashion & luxury · art & design marketplaces · furniture & home · resale & second-hand.
How the split-test methodology works
A VISII deployment splits a share of live traffic at random into two matched groups: one experiences the store as-is, the other with the engine on. Because the split is randomised at the session level and both groups run simultaneously, seasonality, campaigns and stock effects hit both groups equally, so the difference in revenue per session, conversion or order value is attributable to the engine and nothing else. Results are read in the client’s own analytics, not a vendor dashboard, which is why VISII’s case-study figures name the client, the metric and the period, and why the same measurement can back performance-linked commercial terms. It is the standard your data team would design themselves.
What is understanding worth on your line?
Understanding is operating leverage: it grows with your catalogue and your traffic, while the cost of running it does not.
Revenue
Understood shoppers buy more, measurably, per session, per visit, per basket.
Profits
Labour stops scaling with catalogue size. Your best people move from tagging to taste.
Capital freedom
The freedom that arrives when growth no longer depends on buying traffic.
Grow faster, faster. Each lever feeds the next. That’s why the upside curves.
Put your own numbers in.
We review requests from your region by hand; merchants always get in, usually within a day. Send a note, or skip the maths and book the call directly.
Your baseline (traffic × conversion × order value): a month, a year.
And at full scale, a different conversation opens: performance terms. Once the engine runs across your whole business, we can price against the lift itself. We win when you win, provably.
Chosen by leaders in art, design, fashion and lifestyle e-commerce
Questions.
How fast is deployment?
Live within a single sprint (less than two weeks), turnkey or build-it-yourself, via one lightweight snippet. Optimisation compounds over the first cycle (~a month).
Our data is a mess.
That’s fine. That’s normal. No tags, thin metadata, missing collaborative-filtering history, offline data to fold in: the engine is designed for your starting point. All it needs to begin: your product feed and images.
What does it cost?
The test terms sit at the close of this page, in full: priced so the decision needs no committee. Beyond it, one predictable fee at full scale, set to the size of the transformation rather than seats or searches, and once the engine runs across the whole business, performance terms can enter the conversation.
Will it slow our site?
No. One minimal, unobtrusive JavaScript snippet, built to leave your page speed alone; responses typically return in under 100ms.
Can we see it before we commit?
Yes. We build a working demo on your own catalogue before anything is signed; the test then measures it properly, on live traffic.
What about privacy?
No names, no emails, no identifiers, personalisation from in-session signals alone. Sessions stay anonymous to us, and personalisation never needs to touch your CRM.
We’re considering building in-house.
The instinct is right: nobody knows your catalogue like your team. Wescover built their own, and a randomised split test refereed it: ours measured 32% ahead.
Do we have to rip out our existing providers?
Not necessarily, though you may want to. Complement the stack or replace it; the engine is built for either.
What about brand-new products with no history?
No cold start: the engine understands a product from its image and its meaning before the first click ever lands.
Stores that understand their shoppers are replacing stores that ask them to search.
Our best measured results so far:
Outliers included. The ranges we quote are lower.
Those are the numbers.
Book Your Growth Call10% more revenue, in ten days, for £10K, paid only if it works.
We run a limited number of sprints per quarter, each properly measured.
Regardless of your data quality · on a randomised test group vs matched control · integration within a sprint (less than two weeks) · if you want an alternative metric, its success criteria agreed in writing before day one.
Book Your Growth Call