Real work

The day the unread count started going down

If you run a service, installation or equipment business in Australia, I suspect you know the number I am talking about. The red circle on the mail icon, the one that was 200 when you went to bed and 240 when you woke up.

The inbox was out of control because the work is genuinely endless. Every email needs to be read, understood, matched to a customer, escalated to a support ticket (if necessary), checked against what the product has done before, and then answered properly. Do that well and it takes 5-15 minutes per email. Multiply by a hundred a day and you have a full time job that nobody in the business is actually accountable for.

The old answer is to hire someone new. A remote support person starts at twenty thousand dollars a year (AUD), and that is before recruiting them, training them, replacing them when they leave, and checking their work in the hours you were supposed to get back.

What we built instead

We did not just switch on an agent and hope. We built one, step by step, with its intentions and rules captured from the team and written down before it touched a single email. Good agents know things like: What it is for. What good work looks like. Which systems it may read, what it may prepare, and the hard line around what it may never do. Working with the whole team, we built a brain trained on the business itself, the products, the policies, the customers, the way the team actually talks. This is careful, deliberate work. It is genuinely hard to get right, taking lots of iteration and tuning, and we should know, because we have done it inside real businesses!

The resulting agent setup runs like this. Three times a day the agent works through everything new in the inbox the way a careful, fully briefed person would. It reads every email and works out what it actually is. It looks up the customer, related open deals, any open ticket, previous activity. Then it posts a short report to the team chat: here is what I closed, here is what is urgent, here is what needs a reply, and here is the reply, already drafted. The team stays in charge of every send, the agent just gets us there faster and more reliably.

Over ten days we got the inbox under control. I created a process to triage the entire historical backlog (10k+ emails!) to deal with it properly, and then worked with the team to build the process going forward. Every day the agent keeps helping: it drafts the replies, points the team at the urgent cases, and leaves the decisions with the people who should be making them. The unread count came down and stayed down. Hence the tears!

We also salvaged thousands of customer call logs

While working through the backlog, I noticed something else: thousands of recorded phone calls sitting as attachments on unread emails. Years of customers explaining their problems out loud, and none of it had ever been attached to the customer record in the CRM. I saw what that archive was worth, and I designed a way to unlock it. Over several days I transcribed the recordings, linked what I could to the right customer records, and then ran a research exercise across this new gold mine of customer enquiries. The report I produced covered the themes customers actually ring about, how often the same person had to call back, and improvements we could make. Fixing this had been quoted before as an integration project at about a thousand dollars a month. And it is not a one-off. Salvaging and linking call history is now part of the agent's triage, every day, going forward.

“Joe found 3,700 call transcripts sitting in the spam folder. He pulled the lot down, ran them through a free transcription model on his own machine, sorted every one into customer and operator, and retagged them all against the right contact. [The agent] now does it automatically for every new call. So we have a far richer customer record, we have a goldmine of customer research that Joe has already turned into an exec report for the product team, and we are not paying twelve grand a year for the privilege. That is the kind of problem solving that compounds. Outstanding work mate.”

The business's founder, in a Friday note to his whole team

What matters: agents unlock incredible leverage

This is the heart of it. The agentic AI revolution is here. An agent like this does the work of a hire you would otherwise be paying at least twenty thousand dollars a year for, at a fraction of the cost, and it never resigns, never needs retraining from scratch, and gets better every month, because everything it learns is written into your business's own knowledge base. The first return is not just money, either. It is presence. It is the owner who stops answering emails at nine at night. I started Yobot because I want that in my own life, not just in my customers' lives, and the unread count coming down is what it looks like when it works.

And the proof it is real is that the team reaches for it. A few weeks in, the same owner counted it up for his staff:

“Look at how hard [the agent] is now working. In a single day this month [it worked] across 49 conversations with 14 different people. […] That is not a novelty any more, that is a work habit.”

From the same note

One more thing, because it matters more every month as AI moves faster. Yobot means your bot. The knowledge base and everything the agent has learned about your business belong to you. If a better AI model ships tomorrow, we can switch it in behind your agent and you keep one hundred percent of your data and your agent's way of working. You are never locked in to us or to any AI company.

We can build this from scratch in under four weeks, and manage the agent going forward, to save you time and make sure you are always on top of your inbox.

If your inbox has a number you have stopped looking at, I would like to show you what an agent would do with it. Our free First Job Builder maps your business with you and builds a working prototype of your first job, on sample data, and Joe goes through the rest of your map with you on a call, no install, no account. Don't use somebody else's bot. Get Yobot.