The chain sees the effect of personalisation before investing in a big platform
Case study · National electronics chain
AI recommendation system for an electronics chain
A national electronics chain wanted to know how much personalisation actually brings before putting money into a big platform. We built a working sandbox in the cloud on simulated catalogue events: seven store screens, twenty-four recommendation scenarios, ranking on LightGBM and cold-start scenarios for new users. Right there in the sandbox an A/B comparison runs against a control, and we agreed the pilot's metrics — CTR, conversion and A/B — before the start.
- ClientNational electronics chain
- Integrationfrom $2,000
- Supportfrom $150/mo
Implementation results
What changed after launch
24 recommendation scenarios can be switched on and compared in the sandbox
The cold-start question for new users is settled with separate scenarios
The next step is agreed: a pilot on the chain's real events
Solutions
What the sandbox does
- Models the storeSeven screens in the cloud — from the home page and catalogue to the product page, cart, search, account and thank-you page.
- Ranks the recommendationsA LightGBM model picks products instead of simple rules.
- Solves the cold startSeparate scenarios for new users there's no data on yet.
- Compares the variantsAn A/B test against a control right in the sandbox, with nothing run on live traffic.
- Is configured in the admin panelEach of the 24 scenarios is switched on and configured separately.
Features
What's included in the solution
A sandbox in the cloud on simulated catalogue events
24 personalisation scenarios managed in the admin panel
Ranking on LightGBM and cold-start scenarios
An A/B comparison against a control in the sandbox
Pilot metrics fixed before the start — CTR, conversion and the A/B result
AI recommendation system for e-commerce: a sandbox before the pilot
Before putting budget into a big personalisation platform, it's worth knowing how much it will bring. For a national electronics chain we built a working sandbox on simulated catalogue events: seven store screens, 24 recommendation scenarios, ranking on LightGBM and an A/B comparison against a control right in the sandbox. We agreed the pilot's metrics before the start, so the effect can be measured honestly.
Read more
Before this, recommendations in the store were built on simple rules, and there was no way to compare variants before going live on traffic. The sandbox recreates seven key screens — from the home page and catalogue to the product page, cart, search, account and thank-you page — and lets you play with scenarios in a safe environment. Ranking runs on LightGBM, and separate cold-start scenarios answer the question of what to show a user there's no data on yet.
The point of the sandbox is honest measurement. Each of the 24 personalisation scenarios is switched on and configured in the admin panel, and the A/B comparison against a control runs right here, with no risk to live traffic. We fixed the metrics before the start: CTR, conversion and the A/B result. So the decision about a big platform is made on numbers, not impressions, and the next step is a pilot on the chain's real events.
The approach suits retail and e-commerce with a large catalogue, national chains, and marketing and product teams that need to weigh up personalisation without big money up front. We build the sandbox around your catalogue structure and your hypotheses. Describe the task and we'll work out, free of charge, what such a sandbox and pilot would look like for your store; a rough price is from $2,000 for the build and from $150/month for support.
Pricing
The same solution for your business
A project like this starts from that amount. The final price depends on the volume of data, the features and the integrations, so we begin with a free consultation, prepare a rough estimate and send you a commercial proposal.
Who it's for
Who it suits
Retail and e-commerce
A large catalogue where personalisation moves revenue.
National chains
The effect needs weighing up before money goes into a big platform.
Marketing and product
Personalisation metrics agreed before the start, not after.
Analysts
An honest A/B comparison instead of subjective impressions of the recommendations.
How it works
How we'll launch it for you
- Step 01
We discuss the task
We analyse your needs and define the best solution.
- Step 02
We build the solution
We configure AI agents or adapt ready-made solutions.
- Step 03
We launch
We integrate, test, and train your team.
- Step 04
We deliver results
Measurable results and ongoing support.
Other case studies
Similar solutions already at work
ADINDEXAI Google Ads audit and campaign optimisation
Campaign audit twice as fast
AI logistics automation for a transport company
Logistics costs down by up to 30%
NeuroshopAI meeting automation: recording and summaries in Google Meet
Meeting notes three times faster
View case studyFAQ
Before you start
Why a sandbox, if you could just put recommendations on the site straight away?
A sandbox lets you measure the effect of personalisation safely and cheaply: 24 scenarios and an A/B comparison run on simulated events, with no risk to live traffic. You make the decision about a big platform on numbers, not guesswork.
How is the cold-start problem solved?
For users there's no history on yet, there are separate cold-start scenarios. They decide what to show on the first visit, while the system hasn't yet gathered enough data about preferences.
How much does a recommendation system cost?
Roughly from $2,000 for the build and from $150/month for support. The cost depends on the size of the catalogue, the number of scenarios and the integrations; we make the preliminary estimate free of charge.
Free consultation
Want the same result?
Tell us about your task — we'll show how this solution would work for you and give you a price. The first consultation is free.
What you'll get from the consultation





