Industries Retail & E-Commerce
Systems behindbillions checked out carts.
AI Engineering and software development for retail companies.
Most of a catalogue is invisible: a shopper searches a term nobody mapped, a size shows available and isn't, a recommendation still thinks it's last season. We build the discovery, pricing, and inventory systems behind retail and DTC platforms, including the ones that have to survive the twenty minutes after a drop goes live.
FIGSWhere Real Heroes Shop Apparel
The direct-to-consumer platform behind the world's best-known medical apparel brand. Search, product, and checkout for a catalog forty colorways deep, clearing orders in 85 markets, for 3.1M+ customers.
- $630M+
- In annual revenue powered
- 3.1M+
- Customers reached
- 85
- International markets served
Why retail companiestrust us with delivery.
An oversold size, or a search that returns nothing, costs real revenue without ever paging anyone. This is why we pay special attention to the most sensitive parts of the system.
The catalogue is the ceiling
Missing attributes, supplier feeds that disagree, and a taxonomy built by whoever set up the store cap everything downstream. Search and recommendations are only ever as good as the product data underneath.
Zero-result searches are demand data
Every query that returns nothing is a customer telling you what they wanted, in their own words, at the moment they were ready to buy. Most stores log them and nobody reads them.
Returns are the hidden P&L
A conversion that comes back costs picking, shipping, handling, and often the margin twice over. Fit, sizing guidance, and honest product data are worth more than another point of add-to-cart.
Inventory truth is distributed
Stock sits in stores, distribution centres, and in transit, and each system is confident about a different number. Every checkout makes a promise on top of that, and the gap is where oversells come from.
Peak is a day, not a season
A drop or a sold-out restock concentrates a quarter's traffic into twenty minutes. Queueing, caching, and what your platform can shed while checkout stays up are decided months earlier.
What retail demandsfrom engineering and AI.
01Search that understands the query
Shoppers type descriptions, not SKUs, and keyword matching fails exactly where intent is clearest. Meeting that takes semantic retrieval, synonym and misspelling handling, and merchandising rules a buyer can still override, which together turn the search box into the highest-converting surface on the site.
02Recommendations that survive the first visit
Most sessions are anonymous and most products are new, so a recommender judged on logged-in repeat customers flatters itself. The useful ones work from session context, catalogue similarity, and cold-start behaviour, and are measured against a control rather than against last month.
03Pricing and markdown timing
Discounting is usually reactive: too late on slow lines, too deep once it happens. Elasticity by category, competitive position, and inventory age turn markdown into a scheduled decision with a forecast attached, which is the difference between clearing stock and giving away margin on stock that would have sold.
04Forecasting and allocation
Sizes, colours, and stores have to be committed on a lead time so long the answer is locked before the season starts. Getting allocation slightly less wrong compounds through availability, markdown, and the working capital sitting in a warehouse.
Real-life stories of triumph.
Get In Touch
We were searching for a partner who can help us scale for the long term, as our vision is big and evolving. We decided on Nortik, as they showed a very clear understanding of our product and the initiatives we were trying to deploy.
Nortik’s Impact
An AI virtual try-on platform taken from MVP to production, with bulk generation that runs a whole catalog per item and cuts physical sample production and CO₂ by up to 75%.
Discover more of our insights.
Recent Posts
The martech stack has 15,000 tools. AI earns a slot by deleting handoffs, not adding features
Marketing teams don't need another AI-powered tool. The use cases that stick, from campaign assembly to audience queries to creative testing, remove handoffs between people, not clicks between screens.
Fraud, KYC, and disputes: the unglamorous AI that pays for itself in fintech
Fintech AI headlines chase robo-advice and chatbots. The returns live in the back office: KYC document review, transaction enrichment, dispute handling, and fraud ops, with the audit trail regulators expect.
Why AI prototypes fail in production (and how to ship one that doesn't)
Most AI prototypes stall somewhere between the demo and the deploy. The problem is rarely the model. It's the evals, guardrails, and infrastructure nobody scoped.
Frequently asked questions
Everything commerce teams usually want to know before rebuilding discovery. Not here? Ask us directly.
We work with it. Shopify, commercetools, Salesforce Commerce, SAP, or a custom stack: the intelligence layer goes in alongside what you already run, not in place of it.
A replatform is a business decision with its own case. It shouldn't arrive as a side effect of wanting better search.
Usually not. Most of the gain comes from what surrounds the engine: cleaner product data, query understanding in front of it, and merchandising rules that let buyers steer results without a developer.
If the engine genuinely is the constraint, we'll say so and show you the evidence rather than starting from that conclusion.
A/B or holdout from the start, measured on revenue per session rather than click-through, because it's easy to raise engagement while selling less.
We agree the metric and the guardrails before launch, so the result isn't a debate about which dashboard to believe afterwards.
Yes. Session-based and catalogue-driven personalisation works without identifying anyone, which covers most anonymous traffic anyway.
Where identity is involved, consent state travels with the data and suppression propagates to every downstream system, including the ad platforms, which is where most retailers' consent stories quietly break.
That's the design point. Checkout and availability get the headroom and the caching; discovery and personalisation are built to degrade gracefully so the paths that take money stay up.
We load-test against your actual peak shape: the twenty-minute spike, not the busy afternoon.
FIGS is the reference: we built part of the storefront 3.1 million clinicians buy through, including the checkout that clears an order in any of 85 markets.
You do. Code, trained models, pipelines, and documentation are yours, along with the infrastructure access to run them without us.
We hand over a codebase your own team can pick up. Nothing we build is designed to keep you dependent on us.