The automations worth doing, and the ones that aren't

There's a version of automation advice that lists ten things you could automate and calls it a day. That's not useful, because most of the ten aren't worth it. The interesting question is narrower: for a small team with a website and not enough hours, which jobs actually pay for themselves, and which ones look clever but cost you a customer.

Here's how we think about it. The test we use for anything is: is the answer stable, is it already written down somewhere, and does getting it slightly wrong hurt? Stable-and-written-down-and-low-stakes is the sweet spot. The further you drift from that, the more a human should be involved.

Worth doing: answering the questions your site already answers

The best automation isn't ambitious. It's answering "do you offer annual billing?" at 11pm when nobody's staffed, using the answer that's already sitting on your pricing page. The customer gets it in three seconds instead of waiting until morning and maybe not bothering to come back.

The reason this one works is that the answer already exists and doesn't change much. You're not asking the machine to be creative or to make a judgment call. You're asking it to read your own site and repeat it back to the right person at the right moment. That's a job with almost no downside, and it's the first thing we'd point any Engine64 setup at.

Worth doing, carefully: turning a good conversation into a lead

The second one is capturing intent when it shows up. Someone's asked three specific questions about your enterprise plan — they're not browsing, they're evaluating. The automation worth building here is the one that notices and quietly offers to take their details, then hands your team a clean summary with the actual questions attached, so nobody starts the conversation from zero.

The "carefully" matters. There's a bad version of this where every visitor gets a "can I grab your email?" popup on question one, and it feels like being followed around a shop. Wait for genuine intent. A lead captured too early is worse than no lead, because it teaches people your chat is a sales trap and they stop using it for the thing it's good at.

Usually not worth it: personalized recommendations

This is the one that shows up on every automation list and rarely earns its place for a small team. "The AI recommends the perfect product for each visitor" sounds great. In practice, unless you have a big catalog and real behavioral data, the recommendations are generic enough that a customer notices, and a wrong recommendation reads as either pushy or clueless.

We're not saying it never works — for a large store with thousands of SKUs, sure. But for most businesses, the effort-to-payoff is bad, and the failure mode (recommending the thing that's obviously wrong for this person) actively erodes trust. Answer their question well first. That builds more goodwill than a guessed upsell ever will.

Our rule of thumb: automate the task you're tired of doing, not the task that sounds impressive in a demo.

The one to be genuinely careful with: triage and routing

Routing an incoming question to the right place — support versus sales, urgent versus not — is legitimately useful and legitimately easy to get wrong. Useful, because a misrouted ticket sits in the wrong queue for a day. Wrong, because if the routing is confident and incorrect, the angry customer who needed a human in the next ten minutes instead got a cheerful "here's an article!"

So we treat routing as assistive, not autonomous. The agent can suggest where something goes and pass along the full context, but the rule we'd give anyone is: when in doubt, route up, to a person. An unnecessary human glance costs you a minute. A missed escalation costs you the customer.

What this adds up to

If you only did the first thing on this list — answer the repeated questions your site already covers — you'd get most of the value and take on almost none of the risk. That's genuinely the whole recommendation. Start there. Add lead capture once you trust it. Leave recommendations alone unless your catalog demands it, and keep a human's hand on anything that routes or escalates.

Automation isn't about doing more things automatically. It's about picking the few where automatic is clearly better than manual, and being honest about the rest. Most of the ones that sound exciting in a pitch are exciting precisely because they're risky, and risky is the opposite of what you want a machine doing unattended on your website.