Operations Redesign
We don't start with AI.We start with how your work actually runs. AI comes last, and only where it pays.
The process on paper isn't the process that runs.
Ask an owner how quoting works and you'll hear four steps. Watch it happen and you'll often count twelve: re-typing the request, chasing missing details, chasing suppliers, fixing errors, chasing the answer. None of those are in the written process. All of them cost you days.
Automate the written version and you automate a fiction. Drop AI on the real version without cleaning it up and you just make the mess run faster.
Map. Sort. Measure. Build.
Six steps. The first five don't involve AI.
- 1
Map the real process
We talk to the people who do the work, and we read what your systems already record: emails, spreadsheets, software logs. People tell us why. The records tell us what, how often, and how long.
- 2
Separate work from waiting
Most processes are minutes of work spread over days of waiting. We time both, so you can see where the days go.
- 3
Name one owner
One person who can approve changes across every team the process touches. Without this, good changes stall between departments.
- 4
Sort every step
Delete, plain code, AI agent, or your people. In that order.
- 5
Set the baseline
We write down today's numbers before we build anything. Same policies, same approvers, before and after.
- 6
Build, roll out, re-measure
We build inside the tools you already use, train your team, and measure the same numbers again.
Every step goes in one bucket.
We ask four questions about every step, in this order. Most people jump to AI first and skip the cheaper answers.
- Question 1Does this step need to exist?Delete
Rework, double-entry, and “just checking” emails. Gone, not automated.
- Question 2Is it a fixed rule?Plain code
Fixed rules with no judgment. Cheap, reliable, never guesses.
- Question 3Does it need reading or drafting?AI agent
Reading, chasing, drafting, assembling. It prepares the work, then stops.
- Question 4Does it need judgment?Your people
Approvals, pricing, relationships. The calls only you should make.
Agents prepare. People decide.
It proves itself before it runs alone.
No AI agent touches your live work until it has passed a test on your own past cases. We run it on 20 to 50 real examples from your business, the typical ones and the tricky ones, and grade it against what your best person would have done. We agree in writing what it must hand to a person. High-stakes actions always go to a person.
- Stage 1
Shadow
Works alongside your team on live cases but changes nothing. We compare its answers to your team's.
Moves on when: Its answers match your team's on an agreed share of cases.
- Stage 2
Assist
Prepares the work. A person approves every action.
Moves on when: Approvals rarely need changes.
- Stage 3
Limited automation
Handles routine, low-risk cases on its own and sends everything else to a person.
Moves on when: It runs cleanly for an agreed period with no out-of-bounds actions.
- Stage 4
Expand
Takes on more case types, one at a time.
Moves on when: Each new type passes the same tests.
Every stage has an owner on your side and a way back to the stage before. And we build on top of the systems you already use, so nothing gets ripped out.
If we can't measure it, we didn't do it.
Before we build, we agree on your numbers in writing. After launch, we measure them again the same way. Your policies and approvers stay the same, so if the numbers move, it's because the process got better.
- Steps in the process
- How much was removed, not just automated
- Start to finish
- How long your customer or your team waits
- Hands-on time
- How much of that time is real work
- Sent back to be redone
- How often work loops
- Straight-through rate
- Share of cases that finish with no exception
- Cost per case
- Staff time and tools per quote, invoice, job, or file
- Revenue impact
- More quotes sent, faster follow-up, higher win rate
- Risk reduced
- Fewer errors, missed deadlines, and payments made in error
We report the value three ways: money saved, money earned, and risk avoided. A faster quote that wins more jobs can be worth more than the hours it saves.
A typical quote: 4 steps on paper, 12 in real life.
On paper, quoting is four steps: get the request, price it, send it, win it. Mapping what actually happens shows twelve. Here is where each one landed.
- 1
Request comes in by email, phone, or web form
Plain codeone intake form for every channel - 2
Re-type the request into a spreadsheet or CRM
Not in the written process
Deletecaptured once, at intake - 3
Chase the customer for missing details
Not in the written process
AI agent - 4
Ask a supplier for a current price
Not in the written process
AI agent - 5
Wait for and chase the supplier's reply
Not in the written process
AI agentscheduled follow-ups - 6
Work out the price
Plain codea quote template built from your price list - 7
Owner reviews price and margin
Not in the written process
Your people - 8
Quote sent back for fixes
Not in the written process
Deleterules catch errors before review - 9
Send the quote
Your peoplewith a personal note - 10
Follow up on the unanswered quote
Not in the written process
AI agentscheduled, polite follow-ups - 11
Customer accepts
Your peopleyou confirm the details - 12
Enter the job into your accounting software
Not in the written process
Plain codethe accepted quote creates it
Result: 2 steps deleted, 3 handed to plain code, 4 prepared by agents, and 3 kept with people. The owner still sets every price and talks to every customer. What's gone is the re-typing, the chasing, and the quotes that bounce back.
Start with one process. Go as far as you want.
Method steps 1 to 5
Process Sprint
Pick one process. We map how it really runs, time the work against the waiting, sort every step, and write down your baseline numbers. You leave with a redesigned process and a clear plan, whether or not we build it.
Fixed scope, agreed before we start.
Method step 6
Redesign and build
We build the redesigned process inside the tools you already run, prove it on your own past cases, and roll it out alongside your team.
Scoped from the Sprint, so you know the cost before work starts.
After launch
Measure and expand
We measure the same numbers the same way and report the value as money saved, money earned, and risk avoided. Then we take on the next process.
Your policies and approvers stay the same, so the numbers are honest.
Only one task is the problem? Start with an AI Tune-Up →
Questions
Do you just add AI to what we already do?+
No. We usually remove steps first. In most processes, a good share of the steps are rework, duplicate checks, or waiting. Those get deleted, not automated. AI goes only where a step needs reading, sorting, or drafting, and a person still makes the decision.
Will this replace my team?+
No. It takes the chasing, re-typing, and checking off their plate so they can spend their day on the work that needs them: decisions, customers, and the tricky cases. Every agent we build hands its work to a person before anything is approved, paid, or sent. And we involve your team from day one, because they know where the process really breaks.
Do we have to replace our software?+
No. We build on top of the systems you already use. You keep your accounting software, CRM, or ERP.
How do we know it worked?+
We write down your current numbers before we build: how long the process takes, how many steps, how often work gets sent back, and the cost per case. After launch we measure the same numbers the same way, with the same policies and approvers.
Pick one process. We'll show you what's really in it.
Book 15 minutes and tell us about the workflow you're tired of. We'll show you which steps to cut, which to automate, and where AI actually helps.
Book 15 minutes →