AI, but make it human

The Problem Behind the Problem

A software company turned on Fin.

It confused customers, stalled sales conversations, and they pulled it back. They thought Fin was the problem.

But when I looked under the hood, the issue wasn't what anyone thought.

What counted as "AI resolutions" often wasn't AI resolving anything — the numbers were being shaped by system behavior, not real outcomes. The content Fin relied on had gaps, outdated information, and no clear structure around how customers look for help.

Fin wasn't failing. It had never been set up to succeed.

So we rebuilt what Fin was standing on: content organized around real customer questions, structure that mirrors how people think when they're stuck, and reporting that reflects whether customers are genuinely being helped — not just whether a ticket closed.

Their Head of Support put it simply: he wanted to go on vacation, come back, and trust that the conversations marked resolved actually were.

That's what we built.

Nobody Taught Fin Where to Stop

They'd built a brand their customers genuinely loved — strong community, thoughtful programs, real care.

So when they launched Fin, they expected it to carry that same energy.

Instead, it started answering questions it shouldn't have been answering.

It found its way into premium content — material designed for a completely different context — and started confidently repurposing it as customer support answers. Because it sounded confident, customers didn't always question it.

Nobody had taught Fin where to stop.

So we gave Fin the boundaries it never had: content structured around how customers ask for help, clear lines around what it should and shouldn't touch, and guidance that helps it respond with more care — and less assumption.

Now when something nuanced comes in, Fin knows how to support the conversation — and when to hand it to a human.

60% Hidden in Plain Sight

They wanted better visibility into what their customers were asking about.

With hundreds of products, the default solution was tagging — but tagging only works if agents remember to do it, choose the right label, and stay consistent. In practice, it becomes one more thing to manage, and the data breaks down fast.

I took a different approach. Instead of adding to the team's workload, I set up AI-driven categorization — Fin handles the sorting, agents just confirm and close. No extra steps, no inconsistency.

Once the data started flowing clean, something showed up immediately:

60% of their support volume was concentrated around one issue.

It had been there all along — just buried under messy data.

I flagged it and wrote a small set of macros to handle it. When their team lead read through them, she said something that stuck:

If these had been in place months ago, think of how much time we would've gotten back.

That's what clean data does. It doesn't just make reporting easier — it shows you where to act.

From Chat Etiquette to Revenue Discovery

Their team lead came looking for chat etiquette resources.

The team had been handling live chat for a while, but it had mostly been running on autopilot — uncoached, unmonitored, and leaning heavily on the same habits they'd built in email. She'd already Googled around and shared what she found. Most of it was surface-level. "Be friendly." "Respond quickly." Nothing that would change how someone actually shows up in a conversation.

When she followed up, she named the real gap: some of the reps didn't come from a sales background. They were good at support — building rapport, solving problems — but when a conversation turned toward a purchase, they didn't always know how to move with it.

So instead of a script, I built a relational framework — grounded in how I've always approached sales, through relationships, not pressure. Then I connected it to their Intercom workspace and ran it across real conversations, so agents could see what each principle looked like inside their own environment, with their own customers.

Every agent got a personalized review: what they were already doing well, where they had room to grow, and which parts of the playbook would help them most. Their team lead — who'd been doing reviews the same way every cycle without much to draw from — now had specific, nuanced feedback she could actually use.

The rep she'd been most curious about saw her CX score climb 16 points within the first week.

And in the data, something else showed up — revenue opportunities the team didn't even know they were having. Customers were already signaling interest. They just needed someone to point it out.