Case study: automating the new-lead copy-paste nobody has time for

Most writing about AI agents in the workplace either oversells ("AI will run your entire back office!") or undersells ("it's just a fancier macro"). Neither is true. Here's the more useful, more concrete version — built around one task almost every office has, not a list of hypotheticals.

The case: a new lead, typed in three times

A contact form gets filled in, or an email comes through. Someone then copies that same handful of details — name, email, what they're asking for — into the CRM, so sales knows to follow up. Then into the quoting or invoicing tool, so finance has a record if it turns into work. Then, often, into a shared tracking sheet, so whoever's running the pipeline has a single view of everything in flight.

Same information, three destinations, three slightly different formats. Nothing about it is hard. It's just tedious enough to demand attention without ever feeling like real work — the tax every office pays quietly, over and over, without ever quite noticing the total.

Why a template never fully fixes it

The instinct is to solve this with a template or a checklist. It helps, right up until the lead who mentions two things they're interested in in one message, or the form submission with a typo in the email address, or the enquiry that arrives by phone and gets scribbled on a sticky note first. A static process handles the clean 90% and quietly mishandles the rest.

That gap — repetitive but not identical — is exactly where an agent is useful, and exactly where it isn't. It's good at reading the message, recognising the pattern, and filling in the three destinations correctly most of the time. It's not something you should trust alone with the one that doesn't fit the pattern.

What actually changes

The realistic claim isn't "the agent handles new leads." It's: the agent handles the recognisable 90% of the copying, and hands the ambiguous one back — flagged, with everything it does understand already filled in — instead of leaving a person to start from a blank form. Someone still reviews the odd one out. They're just not the one manually re-typing the other nine first.

This is a genuinely small, concrete job, which is exactly why it's a good place to start. It's also the kind of task tools like Hermes Agent — a self-hostable agent built for exactly this sort of long-running, repetitive, tool-using workflow — are actually suited to, rather than the bigger, vaguer "automate my business" pitch most AI tools lead with.

The part nobody budgets time for

Even for a task this small, three things need a real decision, not a default setting:

• Access. The agent needs real credentials into the CRM, the invoicing tool, and wherever the tracking sheet lives. Who can see what it did, and when?
• What happens when it's wrong. Not if — when. If it mis-files the lead with two interests as only having one, does anyone notice before the follow-up email goes out?
• Accountability. If it touches a client record or a quote, someone needs to be able to say exactly why it filed it the way it did — not shrug and say "the agent did it."

None of that is a reason to skip automating this. It's the reason to treat even a small, boring task as a real engineering decision rather than a plugin you switch on.

Where to start

Not with "automate the whole intake process." With this one thing: new leads, copied into your systems. Run it for a few weeks. Watch where the odd one out actually shows up, and fix the access, the failure handling, and the accountability question for that one case before you touch anything bigger.

That's a manageable project. It's also the difference between automation that quietly makes your Mondays better, and automation that quietly becomes the thing nobody can explain six months from now.