The question arrives on most projects now, usually as "what AI should we add". That is the wrong shape. The useful version is which existing problem is expensive enough to justify a build, and the answer differs sharply by business.
Here is a rough ranking by return against effort, based on what tends to survive contact with real users.
Usually worth it
| Feature | Why it works | Effort |
|---|---|---|
| Search that understands intent | Site search is where high-intent buyers go and where most stores lose them. Improving it converts immediately | Medium |
| Product data enrichment | Generating attributes, descriptions and alt text from source data. Unglamorous, saves weeks, low risk | Low |
| Support ticket routing and drafting | Drafts for a human to approve. Keeps quality control where it belongs | Medium |
| Internal document search | Staff finding policy and process answers. Low risk, no customer exposure | Medium |
Depends entirely on the business
- Recommendations. Excellent with a large catalogue and real behavioural data. Pointless with 40 products.
- Customer-facing chat. See the separate piece on this. High variance, high visibility when it goes wrong.
- Content generation. Fine for structured product copy, poor for anything meant to sound like you.
- Configurators with natural language input. Genuinely good when the rules are complex, over-engineering when they are not.
Usually disappointing
- 01A chat interface bolted onto a site where the underlying content is thin. It surfaces the gap rather than filling it.
- 02AI-generated blog content at volume. Search engines now treat scaled content as spam, and the risk is a site-wide demotion rather than a post that ranks poorly.
- 03Personalisation without enough traffic. Below a few thousand sessions a day there is not enough signal, and it will confidently personalise on noise.
- 04Anything replacing a person at the exact moment a customer is upset.
How to choose
Take your three most expensive recurring problems, in hours or in lost revenue. If none of them appears above, the honest answer is that you do not have an AI project yet, and the money is better spent on the thing that is actually costing you.
Then prototype the hard part first, against real data, before committing to a build. Feasibility on your content is the question, and it is answerable in a week.
Working on something like this?
We build websites, stores and custom applications, and we will tell you honestly if the thing you are describing does not need one.