Blog · Published · Updated
AI E-Commerce Personalization: Decide What to Test First
Short answer
AI personalisation in an online shop matches products and content to how each customer behaves, and you measure its effect with an A/B test in your own shop, with no comparisons to Netflix. With 50,000 sessions a month, a 2% conversion rate and a €46 basket, 5% more sales bring €2,300 of revenue and €690 of margin a month. At Syntalith the rollout costs from €3,000 net.
How we workDetails ↓
- 01Call30 minutes with an engineer, free
- 02PrototypeA working system on your process and data, free
- 03ImplementationTarget and fixed price in the contract; 50% at signing
- 04Acceptance50% once the agreed target is met
- net per agent for one process, build paid once
- from €3,000
- from signing the contract to acceptance
- 6–16 weeks
- free prototype on your cases
- €0
We belong to the OpenAI and Anthropic partner programmes and learn through their official courses. What this means ↓
Author: Syntalith · 2 min read
Statistics
Personalisation and AI: statistics 2026
€623
more margin a month at 50,000 sessions, 2% conversion, a €46 basket and 5% more sales at a 30% margin, after the model cost. €3,000 pays back in 4.8 months. Scenario.
78%
According to Eurostat (2025), 78% of EU internet users bought or ordered goods or services online in the previous 12 months.
Source: Eurostat, e-commerce statistics for individuals (2025) ↗
48.2%
According to Eurostat (2025), 48.2% of EU retail companies that use AI use it for marketing or sales.
Source: Eurostat, use of artificial intelligence in enterprises (2025) ↗
20.0%
According to Eurostat (2025), 20.0% of EU companies with at least 10 employees used AI.
Source: Eurostat, 20% of EU enterprises use AI technologies (2025) ↗
84%
According to the European Commission (2025), 84% of Europeans think AI at work needs careful management to protect privacy and ensure transparency.
Source: European Commission, Special Eurobarometer 554 (2025) ↗
Figures checked: 22 October 2026
E-commerce personalisation without control
A shop switches on recommendations and after a month sees more clicks on the carousel. Nobody knows whether sales went up, because there was no control group. The carousel may only have taken clicks the customer would have made anyway. The second trap is data: the more information about the customer, the bigger the legal risk, and the gain from data from outside the shop is often zero.
A personalisation agent chooses products and content based on behaviour in the shop: the categories viewed, the basket and the purchase history. The team sets the business rules: low-margin or out-of-stock products do not go into recommendations, and promotions follow the marketing calendar.
AI recommendation models and their limits
A simple „customers who bought X also bought Y” model works well with a long purchase history and poorly with new products and new customers. A language model can match products to a description of the customer's need, for example „running shoes for tarmac, wide fit”, and it costs more per session. A sensible architecture combines both: cheap recommendations from history for most of the traffic, and a language model where the customer describes the need in words, for example in search or chat.
| Element | What it does | Who controls it |
|---|---|---|
| Business rules | excludes out-of-stock and low-margin products | marketing |
| Recommendations from history | similar and complementary products | the e-commerce team |
| Language model | matches products to a description of the need | the e-commerce team |
| A/B test | comparison with a control group | the analyst |
Data, consent and profiling
- Only data from your own shop: browsing, basket, purchases
- Consent to cookies and similar technologies before data is collected
- Information about profiling in the privacy policy
- No sensitive data and no inferences about it
Profiling falls under the GDPR, and the EDPB guidelines on consent describe when consent is valid. Large platforms have extra obligations under the Digital Services Act.
What AI personalisation can bring
- Sales today: 50,000 × 2% × €46
- €46,000
- 5% more sales
- €2,300 of revenue
- 30% margin
- €690
- Model for 30% of sessions, Claude Sonnet 5.5
- €67
€623 a month is left, so €3,000 pays back in 4.8 months
Traffic, conversion, basket, growth and margin are the scenario's assumptions, which the A/B test checks. Model: 1,500 input tokens and 200 output tokens per session at Anthropic's rates, converted at the ECB rate of 1.1205 USD per euro on 22 October 2026. Recommendations without a language model cost less.
Measurement and control
- 01Control grouppart of the traffic sees the shop without personalisation
- 02Measurerevenue per session and margin; carousel clicks are not enough
- 03Timeat least a few weeks, with a full promotion cycle
- 04Decisionroll out to all traffic only with a result better than the control
Personalisation makes no sense at a few thousand sessions a month, because the test will settle nothing, or with a few dozen products a customer can browse in a minute. Then it is better to refine the categories and the search. In an average shop, start in one place, for example the basket or the product page, and extend personalisation further only after a winning test. We build recommendations for shops as part of our AI apps, and for the free process review the figures for traffic, conversion and the number of products are enough.
Worked example · Scenario
How we count the payback: AI in e-commerce
We run the same count for your shop in the free process review.
€2,792
a month for handling 800 orders that bypass the checkout
Payback: 1.3 mo.
6 minutes to retype an order or answer a question and checking in 15% of the old time are our assumptions. €34.90 an hour is the EU average hourly labour cost for 2025 according to Eurostat, for companies with 10 or more employees. Model cost: reading the order in our system (about €0.0006) plus matching the lines (Claude Sonnet 5.5, about 3,000 + 500 tokens), at the ECB rate of 1.1205 USD per euro on 22 October 2026.
See the full calculation →
