Sales
AI for personalized recommendations and segments
Start with the purchase and browsing data your e-shop and email tool already hold. Turn on built-in features that suggest related products on product pages and in emails. Ask an AI assistant to propose 4–6 customer segments from an anonymised export. Then test one segment-specific campaign against your usual one and keep whichever brings more orders.
Effort:
Medium
Time saved:
Roughly 2–4 hours a week on campaign planning; measure it
The problem today
Most small e-shops send the same newsletter and show the same "popular products" to everyone. A customer who bought a winter coat last week keeps seeing coat offers, and loyal buyers are treated like first-time visitors.
How it works, in plain words
Think of a shop assistant who remembers what each regular bought last time and suggests what usually goes with it, for every customer at once.
What goes in: your order history, product categories and, where your platform tracks it, browsing data. For analysis outside the shop, names and emails are removed first.
What the AI does: it spots patterns, such as products often bought together or customers who have gone quiet, and groups similar customers into segments.
What a person checks: you confirm that each segment makes sense for your business and approve every email or offer before it goes out.
What comes out: product suggestions on your site and in emails, plus a short list of segments with one tailored campaign each.
It cannot fix thin data. With few orders the suggestions stay generic, and it never knows why someone bought something.
How it could work for businesses like yours
Illustrative examples, not client case studies.
A 6-person natural cosmetics e-shop in Thessaloniki
Every subscriber gets the same monthly newsletter, whatever they bought. The shop would turn on its platform's related-products block and create a refill reminder for customers who bought a face cream about two months ago. They would compare orders from the reminder segment with the usual newsletter over one quarter.
A 5-person digital agency in Patras
The agency runs newsletters for several local retailers and sends one version per client list. It would ask an AI assistant to propose segments from each client's anonymised order export, then agree two segments per client before writing anything. A realistic first target could be a higher click rate on segment emails than on the standard send.
A 25-room seaside hotel in Chania
Pre-arrival emails list every excursion and extra to every guest. The hotel would group past bookings by party type, such as families or couples, and suggest a different short list of extras to each group. They would track how many guests book an extra before arrival.
How to do it, step by step
Check your data first. Export 12–24 months of orders with customer ID, date, product category and order value. Remove names, emails and addresses before any analysis outside your shop platform.
Turn on the recommendation features your platform already has, such as related products on product pages and "frequently bought together" blocks. They run inside your shop, so customer data stays there.
Ask an AI assistant to propose segments from the anonymised export, for example "repeat buyers of one category" or "high value, no order in six months". Check each segment against what you know about your customers.
Write one email flow per segment with a clear reason to exist: a refill reminder, a complementary product, a win-back offer. Keep 3–4 segments, not 20.
Run a simple test. Half of a segment gets the tailored email, half gets your usual one. Compare clicks, orders and unsubscribes after 4–6 weeks.
Review segment rules every quarter. Seasonal products and price changes make old rules go stale.
Tools to consider
Shopify Search & Discovery: Shopify's app for related and complementary product recommendations. Caveat: it only works on Shopify, and good results need enough order history.
Klaviyo: email and SMS platform with predictive fields such as expected next order date and churn risk. Caveat: predictions only appear once you have enough customers and orders, so very small shops may get weak signals.
HubSpot: CRM with lists and AI-assisted segmentation, useful for B2B or mixed businesses. Caveat: the richer AI features sit in higher plans; check what yours includes.
ChatGPT or Claude: good for exploring an anonymised CSV and suggesting segment ideas in plain language. Caveat: they can state patterns with false confidence; verify the numbers in your own spreadsheet.
Data protection and GDPR
Segmenting customers and recommending products is profiling under the GDPR. You need a clear legal basis, an updated privacy notice and an easy way for people to object to profiling for direct marketing. Marketing emails need prior consent, apart from the limited exception for similar products sold to existing customers. Website tracking for recommendations needs cookie consent. Avoid segments that infer sensitive traits, such as health, religion or pregnancy, from purchases. If you profile a large customer base or combine many data sources, a DPIA may be required. Check with your DPO or legal advisor.
Mistakes to avoid
Creating 20 micro-segments that nobody has time to write content for.
Uploading full customer exports with names and emails into a public chatbot.
Judging success on open rates alone instead of orders and unsubscribes.