TL;DR (too long, didn't read)
North Energy's staff ask an AI agent instead of digging through 73 files — ops pulls permit contacts and warranty terms, reps pull scripting and finance details live at the customer's table. Answers in under 30 seconds instead of minutes of folder digging, sources named every time. Live in about a week.
The problem
Everything North Energy knew lived in 73 files in a shared drive: warranty policies, permitting contacts for each municipality, sales scripts, finance program terms, training decks. Some of it was scanned PDFs no text search could read.
In the office, staff who needed the warranty terms on a piece of equipment or the right permitting contact — a different one for every municipality they build in — either dug through folders or interrupted a supervisor, so every question cost the company twice. In the field it was worse. A rep at a customer's table who blanked on a finance program's terms had nothing to reach for. The training binder doesn't come to the appointment.
What we built
An AI support agent that lives inside the company's internal tools and works from a phone. Staff type a question in plain language and get an answer from the company's own documents in a few seconds, source files named underneath.
Behind it:
- A pipeline that re-syncs from the company's drive every night. New and edited docs get picked up; unchanged ones are left alone.
- Text recognition for scanned PDFs, so their contents became searchable.
- 2,570 searchable passages built from those 73 documents.
- A branded chat widget on any internal page. No new app, no login.
What changed
Ops: questions that went straight to a supervisor now go to a chat box first. And the knowledge base maintains itself — edit a doc, the agent knows it by morning. Nobody uploads anything anywhere.
Sales: the whole training library comes to the appointment. A rep can pull the exact objection-handling language or a finance program's real terms mid-conversation instead of promising to follow up. Follow-ups are where deals go to die. New reps also train against the same material the top closers learned on.
Every answer names its source files, so anyone can open the actual policy and check. Nobody has to take the bot's word for it.
The efficiency math
The math, shown so you can check it:
- Lookups went from 2–10 minutes to under 30 seconds — up to 95% faster. Digging through folders was the fast path; waiting on a supervisor to free up was the slow one.
- Per person, that compounds. One employee making 3 lookups a day at ~5 minutes each recovers about 15 minutes a day — over an hour a week, per person. Multiply by headcount.
- Every escalation used to cost double. A question that lands on a supervisor burns two people's time. Assume even half of daily questions used to escalate: cutting those in half hands supervisors back hours a week for the work they're actually paid for.
- Upkeep is zero. The knowledge base re-syncs itself nightly. Ongoing maintenance hours: none.
The sales-side number we won't invent: close rate. Nobody's measured it yet. What's structural is this — every "let me get back to you" that becomes an answer at the table is one less follow-up, and follow-ups are where deals die.