More eCommerce Revenue Through AI and Automation

AI and automation in eCommerce: what really works, and what is just an expensive toy

Artificial intelligence is the buzzword of the decade. Every other eCommerce agency promises “AI-powered conversion optimisation”. Most of them mean a chatbot that says “Hello, how can I help you?” and a product recommendation widget that shows “Customers also bought”. That is not AI.

It is automation with a fancy label.

What AI can really do in eCommerce, and where its limits are

AI can spot patterns people miss

That is the real value of AI. A person can read 1,000 support tickets and form a picture. An AI can analyse 100,000 tickets, find correlations and build clusters, in minutes instead of weeks.

A real example: a fashion shop with 50,000 support tickets a year. The team read the tickets, but without a system. The AI analysis showed that 34% of all tickets asked the same question: “Where is my order?”, even though the shop had shipment tracking.

The cause: the tracking email was landing in the spam folder. A simple DNS entry (SPF record) removed 34% of the support load. No AI implementation, no expensive software. Just a pattern nobody had seen.

AI cannot think strategically

AI can tell you WHAT is happening. It cannot tell you WHAT TO DO. An AI says: “The conversion rate drops by 12% between 8 pm and 10 pm.” It does not say: “The conversion rate drops because your payment API is slow between 8 pm and 10 pm, because a third-party update is running.” Interpretation is still human work.

An electronics shop used an AI tool for “automatic conversion optimisation”. The tool recommended: “Reduce the price by 5% between 8 pm and 10 pm.” The shop took the advice.

The conversion rate rose by 2%. Revenue fell by 8%, because the lower price cut the margin more than the conversion rate rose. The AI had found the right correlation but suggested the wrong solution.

Automation that really brings in revenue

1. Automated email flows based on behaviour

Not “welcome email to all new customers”. Instead: “email to customers who left their basket AND have bought once in the last 30 days AND have a basket value over €100.”

A home & living shop switched from time-based to behaviour-based email flows. Instead of “email after 24 hours”, it became “email when the customer leaves the basket AND does not return to the site within 30 minutes”. The open rate rose from 18% to 34%.

The conversion rate of the emails rose from 2.1% to 5.8%. The difference: the email arrived at the right moment, not at the next scheduled time.

2. Dynamic product recommendations based on real buying patterns

Most “Customers also bought” widgets are static. They show products that are often bought together, but not for this specific customer. Real dynamic recommendations analyse: what has this customer looked at? What did they put in the basket? What did they buy? What did similar customers buy?

A beauty shop switched from static to dynamic recommendations. The click rate on recommendations rose from 1.2% to 4.7%. The conversion rate of the recommendations rose from 0.3% to 1.8%. Average order value rose by 11%. The investment: an email marketing tool with segmentation. No AI, no machine learning. Just better use of data.

3. Automated price adjustments based on stock levels

Not dynamic pricing based on demand, which is complex and risky. Instead: automatic price adjustments based on stock levels.

A furniture shop had products that had been in the warehouse for 6 months. The costs of storage and tied-up capital were high. The solution: an automated flow. When a product has been in stock for 90 days, its price drops by 10%.

After 120 days, by 20%. After 150 days, it moves to the sale section. Stock turnover rose by 35%. Liquidity improved. And the conversion rate in the sale section was 4.2%, well above the regular range.

4. Automated A/B test stops and alerts

Running tests in parallel is where automation pays off. Every test has to be monitored: traffic split, conversion rate, SRM (sample ratio mismatch), unexpected drops. That used to be a person’s job, three times a day, by hand.

Today an automated alert system monitors the tests. If a test shows a traffic drop of >20%: instant alert. If a test shows SRM (e.g. 52/48 instead of 50/50): instant alert.

If a test has no conversions after 48 hours: reminder. The system does not replace strategy. It replaces manual monitoring. People still decide what gets tested and how results are read. The automation makes sure nothing slips through the cracks.

AI tools: what they deliver, and where they fall short

AI chatbots for customer service

Worked example: a shop chatbot answers 60% to 70% of standard questions correctly. The remaining 30% to 40% are exactly the important ones, the questions that influence a purchase decision.

A chatbot that answers “Where is my order?” saves support time. A chatbot that gets “Will this product work with my existing setup?” wrong costs revenue. Verdict: chatbots for FAQs, yes. Chatbots for advice, no.

AI-generated product descriptions

The pattern in practice: the descriptions are grammatically correct and SEO-optimised. But they are interchangeable. “This high-quality product made from first-class materials offers you maximum quality and reliability.” That could be a toaster, a sofa or a bicycle.

Product descriptions that convert need specifics. They have to show customers why THIS product is the right one for THEM. AI cannot do that yet. It can only reproduce the average.

AI-based predictive analytics for churn

Worked example: a churn model correctly identifies 78% of the customers who are genuinely about to leave, and its false positive rate is 45%.

When almost half of the “churn risks” would have stayed anyway, the consequence is too many retention emails, and they annoy customers instead of keeping them. Verdict: predictive analytics is promising but not yet ready for autonomous decisions.

The investment rule for AI and automation

Before you invest in an AI tool, ask yourself three questions:

1. 1
   

   What exactly does the tool do, and what does it claim is “AI”?

   An if-then rule set is not AI. A statistical model is not AI. AI starts where the system learns from data and improves its decisions without explicit programming.

2. 2
   

   Can a simpler process deliver the same result?

   If so, use the simpler process. Complexity costs money and time and makes things more error-prone.

3. Who interprets the results? AI delivers data. People interpret data. If nobody on your team has the time to interpret and act on AI results, the tool is worthless.

Frequently asked questions about AI and automation in eCommerce

Do I need AI for conversion optimisation?

No. The biggest conversion gains come from method, not technology. A well-run A/B test with a clear hypothesis framework beats any AI algorithm. AI is an amplifier, not a substitute for strategic thinking.

Which automation pays off first?

Start with email automation. It is the simplest, cheapest automation with the strongest ROI. A well-configured email flow can generate 15–25% of your revenue, with a one-off setup and minimal maintenance.

How much should I invest in AI tools?

No more than 5% of your marketing budget until you have proven, measurable ROI. Most AI tools cost €500–5,000 a month. That money is better spent on tests, traffic and content until you have a specific use case that AI really solves.

Will AI replace CRO experts?

No. AI will support CRO experts, just as calculators support accountants. The strategic decision about WHAT to test stays human. The execution (build, monitoring, analysis) is increasingly automated. The expert who uses AI will beat the expert without AI. But the expert without AI will still beat the AI algorithm without an expert.

Which AI tool do you recommend?

None across the board. The choice depends on your specific problem. For email automation: Klaviyo or Mailchimp. For product recommendations: Nosto or Dynamic Yield. For analytics: Google Analytics 4 with BigQuery. But before choosing a tool, define the problem. Then find the tool, not the other way round.

You can learn the topic step by step in the Academy track Pricing & Offer Psychology ↗.