How to Find Your First AI Win in 30 Minutes (The Most-Tedious-Step Method)

I was sitting with a quoting team at a contract manufacturer — good people, sharp estimators, clearly good at their jobs — and I asked a simple question.

“Walk me through one quote from RFQ to send.”

The estimator — I’ll call him Noah — launched into the description. Customer sends over an RFQ. Sometimes it’s a spreadsheet. Sometimes it’s a Word doc. Sometimes it’s a PDF. Sometimes it’s three different PDFs from three different OEM supplier portals that each have their own format, their own column names, their own decimal conventions, their own part number schemas.

Before Noah could even start estimating, he had to standardize all of it. Copy the data. Reformat it. Match part numbers to internal SKUs. Double-check quantities against the drawing revision. Then cross-reference against the current price sheet. Then check whether any of these part families had a rebate program with the OEM.

I asked him how many PDFs he’d dealt with on his last big RFQ.

He said seventeen.

The insight that changes everything: you don’t automate the whole workflow. You automate the one step that makes the expert in your building feel like a data entry clerk.

That step, for Noah, was the 17-PDF standardization. The actual estimating — the judgment about margins, risk, lead time variability, supplier relationships — that’s what he was hired to do. The PDF cleanup was just noise he had to push through to get to the real work.

Why People Get This Wrong

The most common mistake I see companies make with AI is trying to automate the whole workflow at once. They see a multi-step process and immediately imagine replacing all of it. They scope a project that takes 6 months and $200K. It fails to ship, or ships late, or ships incomplete, or the team doesn’t adopt it because it changed too much at once.

The right move is different. Identify the most tedious step. Automate just that. Stack the win. Then find the next most-tedious step.

This is lean thinking. You don’t redesign the entire floor in one kaizen event. You attack one piece of waste, validate the improvement, then go back and find the next piece.

👉 Tip: The most tedious step is almost always the one your best people complain about most — not the one that takes the longest. It’s the one that makes a skilled person feel like they’re wasting their time. Find that, and you’ve found your first win.

The Most-Tedious-Step Decomposition

Here’s how to do this in a structured way. Sit with the person who does the work — not their manager, not the process owner, the person who actually does it — and walk through one real example together.

Step	What You’re Doing
1. Pick one output	Choose one recurring deliverable: a quote, a schedule, a quality report, a purchase order. One thing.
2. Walk one real example	Have the person walk you through one actual recent example, not a description of the process — a real one they just did.
3. List every step	Write down every action, in order. Include the ugly parts: the workarounds, the manual lookups, the copy-paste between systems.
4. Rate the tedium	For each step, have the person rate it 1-5: how tedious is this? (1 = pure judgment, love it; 5 = I hate every second of this)
5. Rate the complexity	For each step, rate the cognitive complexity: how hard is it to explain to someone new? (1 = pure pattern-matching; 5 = requires years of expertise)
6. Find the 5/1s	The steps that are highly tedious (5) but low complexity (1) are your AI targets. These are tasks that are routine but just haven’t been automated yet.
7. Size the impact	How many times per week does this happen? How long does it take per occurrence? Multiply. That’s your weekly hours of recoverable time.

For Noah’s quoting workflow, the 17-PDF standardization was a 5/1. Tedious as hell (5), not intellectually complex (1) — it’s just pattern recognition and formatting. Automating that step alone recovered about 90 minutes per RFQ across his team, at the rate of 15-20 RFQs per week.

What to Do With the Result

Once you’ve identified the most tedious step, you have three choices:

Option 1: Script it. If the step is pure data transformation (copy this field, reformat that column, match these values), write a script. This is the cheapest, fastest, most reliable solution. No AI needed. A Python script that runs in 3 seconds beats an LLM that takes 30 seconds and occasionally hallucinates.

Option 2: LLM with structure. If the step requires interpretation (read this PDF, understand the intent, extract the structured data), this is where an LLM genuinely adds value. Give it a template for what you want out, let it parse the unstructured input, validate the output against known constraints.

Option 3: Voice or capture. If the step is someone explaining something repetitively — to another person, in a meeting, in a daily report — capture and synthesize that instead. The knowledge already exists; you’re just stopping it from evaporating.

👉 Tip: Don’t solve the problem you think you have. Solve the problem Noah described. “The 17 PDFs” is always more specific and more solvable than “the quoting process.” Get granular before you get ambitious.

🔧 Tool: Build a simple workflow audit table (Step / Tedium / Complexity / Frequency / Time) and fill it out with the actual practitioner. This 30-minute conversation is worth more than any discovery survey or process documentation.

Stack the Wins

Here’s what happens when you do this right. You automate the 17-PDF standardization. Noah gets back 90 minutes per RFQ. He’s not retiring those 90 minutes — he’s doing more estimates, or catching errors he used to miss because he was tired from the formatting work, or improving the margin on quotes because he now has time to actually think about the risk.

Then you go back. What’s the next most tedious step?

Now it’s the OEM rebate lookup. He has to cross-reference part numbers against two different portal sites to find out if there’s a rebate program that applies. That’s 20 minutes per RFQ, also automatable.

Then it’s the margin validation. He has a mental rule: no line over 80% markup. But he’s checking 80-line quotes by eye, occasionally missing one. That’s a script — check every line, flag anything over the threshold, stop the human from having to be the quality control on data that has objective rules.

Each win stacks. Each one is easier to fund because the previous one proved the model. Each one builds organizational confidence that this stuff actually works.

The Counter-Intuitive Part

Here’s what surprises people: the most valuable first win is usually the most boring one.

Not the most technically impressive. Not the one that looks good in the strategy deck. The one that makes your best estimator stop feeling like a data entry clerk.

Because when you give that time back to Noah, two things happen. First, the business gets a measurable ROI on a small investment — and that makes the next project easier to fund. Second, Noah becomes a believer. And a believer who does skilled work in your building is worth more to an AI rollout than any technology decision you’ll make.

Find the most tedious step. Fix it. Stack. Repeat.

That’s the whole method.

---

Want to run this exercise on your operation with someone who’s done it before? Let’s talk about a roadmap session.