Fixing step 4 with ai could make step 5 worse

***

Most of you know my take on AI.

Business outcome first → Mapped to the workflow → Mapped to the #1 constraint → Then AI.

This week, I’m going to use an example of what this looks like inside a real business to drive the point home.

Better Coffee, an equipment servicing business out of Ottawa.

Two questions came before any tool.

The first one is strategic. Where is this business going, and what has to be true for it to get there?

Otto and Josiah at Better Coffee answered that last September when they started Levership’s 12 month program.

They decided they weren't building two good jobs for themselves. They were building something that could eventually carry other technicians across the country.

The second question: Inside the workflow that carries that outcome, which single step is the blocker?

Why the step matters more than the tool

Take any process in your business and write down its steps in order.

Most have between five and seven steps. One of them is slower than the rest, and that step sets the pace of the whole thing.

Watch what happens depending on where you point AI:

Lead generation is the classic step-4 purchase.

Sales feel slow, so more leads go in the top, and now there's a longer queue in front of the actual problem.

Mapping it: one equipment sale at Better Coffee

Better Coffee sells, installs and services commercial coffee equipment.

This was their rough process for an equipment sale a year ago.

Step 5 was their constraint.

Longest cycle time. Most owner attention.

And it was the only step where a deal could go quiet without anyone noticing, because nothing tracked a proposal once it left the building.

Better Coffee had no shortage of inquiries.

Buying into the exciting AI lead gen software and use cases last year would have pushed more work into a step that was already backed up.

The part that came before any AI

Step 5 wasn't ready to be automated.

You can't template a proposal when every price gets worked out from scratch.

So the first fix was not AI.

They moved off hourly billing to value-based pricing per contract, and added preventative maintenance contracts as recurring work.

Then a second call: stop taking new residential clients, and move the existing home clients to drop-off instead of on-site visits.

Fewer proposals, larger ones, and similar enough to each other to be repeatable.

Neither of those is an AI decision.

Boston Consulting Group puts the real split at 10-20-70. Ten percent models, twenty percent technology, seventy percent people and process.

The pricing and client-mix calls live in that seventy percent, and skipping it explains why 95% of businesses integrating AI are not seeing real money back from it.

Then AI

With the step repeatable, Josiah built the first working version of a proposal tool himself, in Claude.

Good, better and best pricing options in about five minutes, with a matching contract and an email draft ready to go. He handed that version to a developer to finish rather than paying someone to guess at the spec.

Proposal tracking is next, which is step 6, and it's worth building now that step 5 has cleared.

Days of work down to about five minutes, on the step that was setting the pace of the entire pipeline.

Same analysis, different workflow

Their delivery side had its own constraint.

Better Coffee services roughly 50 different machines, and every one has its own manual.

Prepping for annual service season means working out which parts each machine needs, which of those you buy brand-name and which have a generic equivalent that holds up, and getting all of it into inventory before the calls start.

That research step gated the inventory buy, and the inventory buy gated the season.

One person and 50 PDFs.

Otto pointed Perplexity at all of them. Every manual crawled, every part listed, grouped by service category, cheaper alternatives flagged. About $1,000 in credits, and he estimates it saved him weeks.

Look at the division of labour there.

The AI did retrieval and sorting. Otto made the call on brand versus generic. That's close to the 10-80-10 we teach: ten percent setting it up properly, eighty percent machine, ten percent a human reviewing what came back.

Josiah's words: "None of this is unique to our business."

Trunk 7, a cladding company we work with, had the same constrained step in a completely different trade. Same analysis, and they now save roughly 600 hours a year on proposals with a 17% lift in conversion. They've 2x'd their business in 6 months.

Once you can name the constraint step, the tool question mostly answers itself.

Otto and Josiah just finished their first year with us. What it bought them:

• "20-hour days, always on call, no weekends" → For Otto, time off at 4pm three days a week for his kid. Both of them have weekends back.

• Maxed out on cash and chasing debt every month → adding to reserves every month

• Two founders doing everything → first hire outside the two of them

• Crashing on a friend's couch in Montreal → a lease that expanded their service footprint

• 
  

None of that came from buying an AI tool. It came from a roadmap: find the step that sets the pace, fix the work around it, then put AI where it plugs in.

Put AI on the constraint and you get a year like this one.

They just started year two last month. Their words: "Year two, let’s do it again!”

If you want to have a conversation about your AI roadmap, the door's open.

Cheers,

Matt

PS. Something else Josiah said: “I’ve honestly read twenty-five business books in the last year. We’re not lacking for concepts. Ideas, sure. Strategy, great. But what really changed my mind, and I think makes Levership actually helpful and useful, is that it’s 10% of making a plan and 90% of getting the plan actually sorted out, and then on track, and then with follow-up. So it’s not like, ‘Here’s a bunch of work for you to do.’ It’s, ‘Here’s a bunch of work that we’re going to do together with you’ and then keep you on track and accountable to actually do that.”