This is a real project. All details have been anonymized to protect the client. But the process is real, and so are the pitfalls we hit along the way.
Background
A mid-size service organization, a few dozen employees. Their core work is serving people — but that work comes with a mountain of records, reports, and checklists.
Their problem was specific: every time there was an important audit, the entire team spent nearly two weeks just getting ready for it.
Two weeks. Not doing their actual work — just gathering data, filling in documentation, and matching formats.
But the deeper problem was: those records were supposed to be accumulating naturally through daily work. There should be no scrambling before an audit. Because everything was on paper, information was scattered everywhere — handwritten notes, LINE group chats, Excel spreadsheets, Google Drive — so every audit was essentially starting from scratch.
When the director first reached out to me, she said something that stuck with me: "Our expertise is serving people, not writing reports. But now everyone spends more time on reports than on actual work."
Step 1: I Didn't Open My Computer
After our initial call, the first thing I did wasn't write code. It was go on-site and watch how they actually worked.
I sat in on their meetings for three days, looked through their paper records, and observed their daily routines.
After three days, I realized the real problem wasn't just "audit prep takes too long" — it was something more fundamental:
-
Everything was on paper. Daily work logs, meeting notes, activity photos, assessment forms — all handwritten or scattered across personal phones. Need to find a record from a specific day? You're digging through paper.
-
Staff didn't know how to write reports — it wasn't a motivation problem, it was a skill gap. The frontline staff were hardworking and genuinely good at their jobs. But ask them to write a report that meets government format requirements, and they had no idea where to start. Not a bad attitude — a missing skill.
-
All the paperwork fell on two or three people. The director and one or two senior staff were carrying the entire documentation burden. Everyone else couldn't help even if they wanted to.
This wasn't an AI problem. It was a paper-to-digital problem that hadn't been solved yet.
Step 2: Digitize the Paper Workflows First
The client originally wanted an "AI management system." I told them: let's put AI aside for now. Let's do one thing first:
Get the paper processes online.
Concretely, I built a simple management system:
- Daily work logs: instead of handwriting, staff fill them in on the system
- Activity photos: instead of sitting on individual phones, they get uploaded directly to the system and auto-sorted by date and activity
- Various forms and checklists: instead of paper, they're filled out online in a consistent format
The point wasn't a fancy tool — it was a process change. Before, everyone recorded things their own way, with different formats in different places. Now everything went to one place in a consistent format.
Just this step alone solved the "can't find the data" problem.
Step 3: Let Three People Try It First
After the system was built, I didn't roll it out to the whole organization. I found three people — one manager, two frontline staff — and had them use it for a week first.
On the third day, one of the staff said to me: "There are some parts of this interface I don't understand."
Perfect. That's exactly what I needed to hear.
Most of these staff members were in their 40s to 60s, and not familiar with digital tools. Their biggest fear was "pressing the wrong button and breaking something." So I was especially careful: big buttons, few steps, plain language they actually used in conversation — not system jargon.
When a staff member is willing to tell you what's not working, it means they see value in the tool and want it to be better. If they didn't care at all, they wouldn't bother complaining.
I simplified the workflow based on their feedback and removed one feature I thought was important but that nobody actually used.
Step 4: Add AI for the Last Mile
Once the digitization was stable, the interesting part started.
The system now had a large accumulation of daily records — photos, meeting notes, work logs. The data was there, and the format was consistent. This was the moment for AI to step in.
I added two AI features:
First: auto-generate reports. Staff just take photos and make voice recordings. The AI reads the photos and recordings and automatically generates reports in the required government format. They just need to review and tweak. What used to take a senior staff member three to five hours to write now takes fifteen minutes.
This wasn't making "people who can write reports" write them faster. It was making "people who couldn't write reports" able to produce compliant ones. The director could finally delegate that work.
Second: auto-classify and archive records. For each entry, the AI determines which category in the management framework it belongs to and files it in the right place automatically. What used to require two weeks of scrambling before an audit was now accumulating automatically every single day.
By the time the audit came, the director opened the system and everything was already there.
She told me: "What used to take two weeks to prepare for now just takes half a day to confirm."
Looking Back: Why This Worked
1. Digitize first, then add AI. Without getting the paper workflows online first and standardizing the data format, AI would have nothing to work with. A lot of people want to skip this step and go straight to AI — and the result is garbage output, because the input was garbage.
2. We solved a capability problem, not just an efficiency problem. Not making the people who could do something do it faster — making it possible for people who couldn't do it at all. That's where the real value is.
3. Involve the actual users. I didn't sit in an office imagining what they'd need. I let them use it, and then they told me. Especially when your users aren't young people and aren't familiar with digital tools, you need to go on-site and watch them use it.
4. Features emerged from use — they weren't designed upfront. The AI report generation and auto-classification were both added after the digitization was stable. If I'd tried to add them at the start, forcing AI onto a paper-based process, the result would have been wrong.
What Does This Mean for You?
If you're thinking about adopting AI — in any industry — ask yourself one question first:
Are your processes digitized?
If you're still working out of paper, Excel, or group chats, the first step isn't AI. It's getting all of that into one unified system.
The logic of the whole process looks like this:
- Go on-site and observe (don't just take a briefing — watch how they actually work)
- Digitize first (move paper to systems, consolidate scattered data)
- Let the actual users try it (three people is enough)
- Then add AI once things are stable (this is when AI has something to work with)
Sounds slow? It's actually faster. Because every step you take is in the right direction — you won't get three months in and realize you've been heading the wrong way the whole time.
Build the foundation before you build the house.
I'm Young. I help organizations with digital transformation. From process mapping and system building to data integration and AI adoption — I start from wherever your problem actually lives.