The meeting that keeps happening
A prospect calls us. They want to know if we can build "an AI agent" to handle customer support. We ask what the agent would need to do. They describe it: read tickets, pull account data, send a response, and escalate if needed.
We ask how ticket data flows into their system today. Silence. Then: "It's kind of manual. Our ops team forwards emails from a shared inbox into our support tool, and then someone pulls account info from a different dashboard."
There is no AI agent to build here. There's an email forwarder to write. An API to connect. A templating system to standardize responses. Maybe — at the very end — an LLM call to draft a first-pass response a human reviews.
Automation is the water. AI is the boat.
The businesses getting the most value out of AI are the ones who already had their automation house in order. They have APIs. They have structured data. They have reliable pipelines between systems. They have workflow state they can observe.
AI shines when it has somewhere to sit. Most companies are still building the somewhere.
What we do first
Before we even scope an AI feature, we look for:
- Manual data re-entry between systems. If your team copies and pastes between tools, that's an integration, not a language model problem.
- Scheduled tasks done by humans. A weekly report? A monthly reconciliation? Cron job territory.
- Documents handled by eye. Invoices, contracts, receipts. Deterministic extraction first, LLM extraction second.
- Escalation rules that live in someone's head. Put them in code. Then, maybe, improve them with AI.
When those systems are in place, AI becomes cheap to bolt on. When they aren't, AI becomes an expensive band-aid.
The short version
Automation gives you leverage today. AI gives you leverage on top of the automation you've already built. Sequence matters.