<- all tokdocs

Map a Process Before Adding AI: The Quote Workflow Example

By , founder of Mavric and Apso

View on TikTok ->

Before adding AI to a small business, map one process, measure where work waits, then decide which steps go to AI, which to plain automation and which stay with a person. A TikTok from @pro.glitch (Professor Glitch) walks through this with a heating company whose quotes took four days. The method is a reasonable operations habit, and n8n does support the human-approval step it relies on, but the company and its "4 days → same evening" result are a staged example with no measured data behind them.

The saved copy of this video had no audio track or transcript, so this breakdown uses the frames and the post description.

Why not just plug an AI agent into your inbox?

The video opens with a cautionary story. On Friday an owner "Hears about a new agent", on Saturday "Hooks it up to the inbox", and on Monday the agent "replies to the biggest supplier" with a discount code, "THANKYOU". The description says "most owners plug a new AI agent into their inbox first and nothing improves, because nobody wrote down what was broken."

The story is a hypothetical. The video shows the agent causing harm while the description says "nothing improves", and "most owners" has no survey or count behind it. The point it sets up is fair: decide what is broken before you automate.

How to map the process

The example company is "Corbett Heating & Air", which the description calls "a 12-person heating and cooling company". Its swim-lane map runs: customer calls, office takes the details, tech visits and measures, owner writes the quote, customer says yes, office books the install, crew puts it in, office sends the invoice.

The description lists eight steps:

  1. Map the business in one line.
  2. Name the owner of each step and circle every handoff.
  3. Measure each step.
  4. Pick the highest-return area.
  5. Break it down and list the edge cases.
  6. Label each step AI, automation or by hand.
  7. Build it in n8n with a human approving.
  8. Measure again with the same number.

The on-screen banners use six numbered steps instead, and the edge-case screen carries no number. The order of activities is the same.

Where do quotes get stuck?

The video's claim is that handoffs are "where work sits and waits". In the example, the measure step puts "8/wk · 4 days" on "Owner writes the quote" and "4/wk · 1 day" on "Crew puts it in". A mock customer card reads "Waiting on a quote · day 4" and then "Called another company". The description's advice is to "write down one number before you build: days from visit to quote."

That number is the useful part. It gives you a baseline to compare against after the build.

AI vs. automation vs. by hand

A three-column board defines the labels:

  • AI: "messy reading + writing", such as turning a voice note and photos into clean notes, or drafting an answer to a question, with "✓ owner approves first"
  • Automation: "same way every time", such as putting the right prices on the quote, sending it, and following up after 3 days and then a week
  • By hand: "judgment + trust"

The description keeps site visits, unusual prices, discounts and upset customers with people. The edge-case screen shows why: "Old house: ductwork nobody has priced". A price-list lookup cannot fill that, so a person has to.

How does n8n handle human approval?

The build combines an AI branch that writes notes with a rules branch that fills prices from a spreadsheet, landing as a draft quote on the owner's phone with Edit and Send buttons. The phone screen is a mock-up, not an n8n interface.

n8n's human-in-the-loop documentation describes a "Human review" section in the AI Agent node's Tools Panel. A reviewer can approve, which runs the tool with the AI's inputs, or deny, which cancels the call and tells the AI it was rejected. Approval channels include n8n Chat, Slack, Discord, Telegram, Microsoft Teams, Gmail, WhatsApp Business Cloud, Google Chat and Microsoft Outlook, and the approval channel can differ from the chat channel. The docs also say to describe the approval setup in the system prompt, which the video skips.

Approval applies per tool call. "Approving every quote" works only if you route the quote-sending tool through review, because n8n does not enforce it by default.

Did quote time really drop from 4 days to same evening?

The video shows "DAYS FROM VISIT TO QUOTE: 4 days → same evening". Corbett Heating & Air is an example company, and the "same evening" figure has no run count, date range or data source. The calendar and email panels are drawn mock-ups. Treat both numbers as a picture of how to record a baseline and a target, not as evidence the method worked.

Where the video sends you

The final screens advertise "The Thinker's Mind", an 11-episode series inside the creator's community, plus automation courses, "8 courses in total", with "Link in my profile". The description has no URL, so price and contents are unknown. The hashtags include #claude, but no frame names a model, and n8n is the only tool shown.

Bottom line

Mapping a process and measuring one number before you automate is sound advice, and the AI, automation and by-hand split is a practical way to assign steps. n8n's human review can gate an AI agent's tool calls if you configure it. The heating company and its four-day-to-same-evening result are illustrations, so measure your own baseline and your own result.

Sources

Published October 29, 2026. Based on a TikTok by @pro.glitch. Full video notes (markdown)