The exact process mapping methodology we use before building any AI system. Most AI projects fail because they skip this step.
Article text
How We Diagnose a Broken Business Process Before We Touch AI
A client handed me a spreadsheet last month with 14 tabs, 6 formulas that referenced deleted columns, and a sticky note that said "ask Janet." That was their entire customer onboarding process.
They'd called us because they wanted "an AI chatbot for new clients." What they actually needed was to figure out what their onboarding process even was before any technology could improve it.
This happens more than you'd think. About 70% of the companies that contact us asking for AI don't have a process problem that AI solves. They have a process problem that needs to be
visible
before anything can solve it.
Why This Matters More Than the Technology
If you're running a 20- to 80-person company and you're thinking about AI, the most important thing I can tell you is this: the technology is the easy part. The hard part is understanding what your business actually does, step by step, in enough detail that you can decide where automation fits.
If this sounds like your business, we do free 30-minute discovery calls where we can talk through whether your processes are ready for AI or need mapping first.
Most AI projects fail not because the technology doesn't work. They fail because someone automated a broken process, and now it's broken faster. A 2024 RAND Corporation study found that roughly 80% of AI projects don't make it to deployment. In our experience working with small and mid-sized businesses, the number one reason is the same every time: nobody mapped the process first.
So we don't start with AI. We start with a diagnosis.
The Five-Step Process Diagnosis We Run Before Every Engagement
This is the exact methodology we use with every client before we write a line of code or configure a single tool. We've run this process with HVAC companies, staffing agencies, law firms, dental practices, and commercial cleaning operations. The industries change. The methodology doesn't.
Step 1: Follow the Work, Not the Org Chart
The org chart tells you who reports to whom. It tells you nothing about how work actually moves through the business.
We start by asking one question: "Walk me through what happens when [trigger event] occurs." The trigger event depends on the business. For a plumbing company, it's "a customer calls with an emergency." For a staffing agency, it's "a new job order comes in." For an accounting firm, it's "a new client signs the engagement letter."
Then we listen. We take notes. We draw the flow on a whiteboard or a shared doc. And we ask follow-up questions that most people skip:
"What happens when that person is out sick?"
"Where does this information live when it's between systems?"
"Who checks this for errors?"
"How do you know when this step is done?"
What we're looking for isn't efficiency. Not yet. We're looking for the
actual
process, which is almost never what the owner or manager thinks it is.
A property management company told us their tenant maintenance request process had 5 steps. When we mapped it by talking to the people who actually handled the requests, it had 13. Eight of those steps were invisible to management because they happened in text messages, personal email inboxes, and a shared Google Sheet that only two people knew about.
Step 2: Find the Human Glue
Every small business has at least one person I call "the human glue." This is the person who makes the broken process work through sheer willpower, institutional knowledge, and a bunch of workarounds they've invented over the years.
Janet, from the sticky-note spreadsheet? She was the human glue. She knew that column F in Tab 3 was actually supposed to pull from the CRM, but the integration broke in 2023, so she copies the data manually every Monday. She knew that the "status" field in their project management tool was wrong for about 30% of projects because the field team doesn't update it, so she cross-references it with the dispatch calendar.
The human glue is critically important for two reasons:
They know where all the real problems are.
They've been patching them for years. They can tell you exactly where information gets lost, where delays happen, and where errors slip through.
They're a single point of failure.
If Janet quits, retires, or gets hit by a bus (the morbid but useful planning scenario), that entire process collapses.
When we identify the human glue, we're not trying to replace them. We're trying to document what they know so the business doesn't depend on one person's memory. Often, the best AI implementations we build are directly informed by the workarounds these people invented.
Step 3: Map the Data Flow (Where Does Information Actually Live?)
This is where most businesses get an uncomfortable surprise.
We ask every person involved in the process: "Where do you get the information you need to do this step? Where do you put the information when you're done?"
The answers usually involve a combination of:
A CRM that's partially filled out
A shared spreadsheet that three people edit simultaneously
Email threads that serve as the "official record"
Text messages between the field team and the office
A whiteboard in the break room
Someone's personal notebook
"I just remember it"
We draw a data flow diagram. It usually looks like a plate of spaghetti. That's normal.
The reason this matters for AI is simple: AI systems need data. If the data is scattered across 8 different places, half of it is in someone's head, and the other half is in an email thread from last Tuesday, no AI tool can work with that. You'd be asking a machine to read sticky notes.
One electrical contractor we worked with had customer information in QuickBooks, job details in a dispatch app, material costs in a spreadsheet, and change orders in a filing cabinet. The same customer's information existed in four places, and none of them matched. Before we could automate anything, we had to get the data into a shape that made sense.
Step 4: Quantify the Friction
"It takes too long" isn't a diagnosis. We need numbers.
For every step in the process, we measure (or estimate with the team's help):
Time per occurrence
: How long does this step take each time?
Frequency
: How often does it happen? Daily? Per job? Per customer?
Error rate
: How often does this step produce a mistake that someone downstream has to fix?
Wait time
: How long does work sit idle between this step and the next?
Rework rate
: How often does this step need to be redone?
We multiply these out to get a weekly and monthly cost of each friction point. This is where eyes usually go wide.
A dental practice we assessed was spending 11 hours per week on insurance verification calls. Not because the calls were long, but because the information needed for the calls was spread across three systems, and the front desk staff had to manually look up and cross-reference patient data before every call. The call itself took 4 minutes. The prep took 12 minutes. For 40+ calls a week.
That's $34,000 a year in staff time on a process that could be reduced to 2 minutes per call with the right data integration. Not even AI. Just data integration.
Step 5: Rank and Recommend
Not every broken process needs AI. Some need a better spreadsheet. Some need a real project management tool. Some need two people to stop using text messages for official communication.
We rank every friction point we've found on two axes:
Impact
: How much time, money, or risk does this friction point cost the business?
Solvability
: How hard is this to fix? Is it a tool configuration? A workflow change? A custom automation? A full AI agent?
The combination creates four categories:
High impact, easy fix
: Do these first. Often they're not AI at all. They're workflow changes, tool configurations, or simple automations.
High impact, hard fix
: These are your AI candidates. The problems where the complexity of the task justifies the investment in a custom solution.
Low impact, easy fix
: Nice-to-haves. Do them if you have time. Don't prioritize them.
Low impact, hard fix
: Skip these entirely. The ROI isn't there.
Most businesses expect us to come back and say "you need AI everywhere." The reality? Across our last 15 engagements, an average of 2 out of 7 identified friction points actually warranted AI. The rest were solved with better tools, better workflows, or better data hygiene.
Red Flags We Look For
After running this diagnosis dozens of times, there are patterns. If three or more of these show up in your business, your processes probably need mapping before you invest in any automation:
The "only Janet knows" problem
: Critical business knowledge lives in one person's head
The copy-paste bridge
: Someone manually moves data between two systems multiple times a day
The status check loop
: People regularly ping each other (Slack, text, walk to someone's desk) to find out "where is this at?"
The phantom queue
: Work sits in someone's inbox or on a clipboard waiting to be processed, and nobody knows how long the queue is
The reconciliation ritual
: Someone spends hours each week or month making sure two data sources match
The tribal workflow
: "That's just how we've always done it" is the explanation for at least one critical process
The override rate
: More than 20% of automated or system-generated outputs get manually overridden
What You Can Do Right Now (Without Hiring Anyone)
You don't need us to start this work. Here's a version you can run yourself this week:
Pick your most painful process.
The one that generates the most complaints, errors, or fire drills.
Talk to the person who actually does it.
Not the manager. The person whose hands are on the keyboard.
Write down every step.
Not the official process. The real one. Include the workarounds.
For each step, note where data comes from and where it goes.
Draw the arrows.
Estimate the time and frequency for each step.
Multiply it out to a monthly number.