How I Use AI Workflows to Accelerate User Research

I used to think proper user research had to be painful.
You had to build long surveys.
You had to collect thousands of complaint tickets.
And you had to read every single user complaint until your eyes hurt.

On paper, the process looked right.
It felt like I was working hard to defend the user.
Managers loved the thick reports.
Stakeholders nodded at the complex charts.

But by the time the report was done, the problem was stale.
The market had shifted.
Competitors had already released the feature.
And the product we built was late because we spent too much time processing data.

Why does this happen?

Imagine you want to know the fish population in the ocean.
So you decide to fish with a single line.
You catch some fish, and you can study them.
On paper, you have highly valid data.

But it takes you a full day just to get ten samples.
Meanwhile, tens of thousands of fish swim right under your boat every hour.
You just need a wide net to capture their patterns instantly.

Yet many product managers still use a single fishing line to catch data the size of an ocean.

Lately, people ask about my approach.
As a product leader, my name is often attached to efficiency initiatives.
They ask how I run user research and find root causes without spending months.
Many assume the secret is a massive research budget or a dedicated team of fifty people.

The reality is entirely different.

A habit of saving, long before AI

Building efficient companies is an old habit of mine. It started long before generative AI became a trend.
I have always pushed various cost saving initiatives with my teams as a product leader.
We fixed internal processes, optimized wasteful infrastructure, and made product decisions that were more cost efficient.
Automation was just one part of this.

The result? Total savings reached $4 million+ USD per year across the companies I have worked at.
That number was not an overnight accident.
At Flip, we cut money transfer costs by 32% in 6 months.
When I was at Tokopedia fixing the authorization platform, my team and I cut OTP costs by USD 2 million per year. We also opened the path for 1.5 million extra transactions every month.

But honestly, leverage that big used to be very expensive.
To find out where the money was leaking, I needed access and a strategic position.
I needed dozens of engineers to pull data from legacy systems.
I needed data analysts to separate noise from important signals.
And I needed expensive system architecture just to process tens of thousands of lines of problem logs.

Just researching and finding the problem took massive amounts of time.
A simple idea to validate a user bottleneck had to wait in a backlog for months.
The product team could only wait patiently while reading raw data manually in Excel.
The idea looked easy, but executing it to get the data was highly expensive.

Now? The situation is completely reversed.
AI is the sharpest tool to apply those exact same habits.
And most importantly, this automation mindset can now be trained into everyone on the team.
You no longer need to wait in an engineering queue for months.

An AI net for user complaints

When I was trusted to lead four product teams at once, the challenge was brutal.
We had to build a large loyalty platform connecting multiple systems in the Middle East.
The release target was tight.
On the other hand, about 10,000 customer complaint tickets came in every month.

If I wanted to research why users were stuck, the old way was sampling.
Pull 100 random tickets.
Read them manually one by one.
Then draw a conclusion that is supposed to represent all users.

The problem is that with sampling, you miss many important anomalies.
Small things that are actually critical get ignored because they did not make it into the 100 ticket sample.

That is why I decided to implement an AI workflow to pull insights.
In two weeks, I built FINA.
This is an AI orchestrator designed to handle customer support tickets.
Its main target was more than just replying to chats quickly.
Its real job was to research and dissect the root causes from tens of thousands of data points.

The results were striking.
From zero automation initially, this system immediately handled 70% of the 10,000 monthly tickets.
Support team costs dropped by 42%.
One interaction cost roughly ~$0.004 per turn.

But the highest value of FINA was not the cost savings.
The highest value was the research data I got instantly.

While this system was running in the pre release phase, the AI found a strange anomaly.
There was a data leakage flaw between users appearing from a repeated pattern of complaints.
This was truly a needle in a haystack.
If I had used manual research checking only 100 ticket samples, this flaw would almost certainly have been missed.
But because the AI read 100% of the tickets completely, even the smallest pattern became clearly visible.
We managed to patch that fatal leak long before the full public release.

The AI read an ocean of data using a giant net.
Meanwhile, if I forced myself to do it manually, I was just fishing with a dull hook.

This workflow is not just for corporate products

I also use this mindset of reading wider data to develop personal projects.
Audience research and testing content formats are equally exhausting if done by hand.
You have to test many writing styles, post manually, pull metrics from each platform, and read comments one by one to find out what the audience likes.

The solution?
I built an 8 channel content distribution system.
This system runs entirely on its own, spreading videos and articles to various platforms automatically.
The reports aggregate themselves, and I can instantly see the complete data on which messages attract the most attention.
Market research that used to require a full social media team now runs endlessly every day. I just set the strategy.

All these experiences building workflows became the real foundation when I started the AI Circle community.
These field materials are also the backbone of the Applied-AI Certification program I structured.

Everything starts from one mindset:
How can I process millions of user data points without having to read them manually with my own eyes?

Do not waste your time on robot work

Many people are still afraid that AI will push product managers away from user reality.
They say you have to read user complaints yourself to have empathy and feel their pain.

To me, that is a misguided myth.
Empathy does not mean you have to torture yourself processing thousands of spreadsheet rows until morning.
Real empathy happens when you already know the problem patterns from AI, and then you pick up the phone.
You call five real users who are the most impacted, listen to their voices directly, and focus on finding solutions together.

AI takes over the heavy lifting of pulling and sorting the data.
You take over the thinking and building empathy.

This is a principle I hold strictly at work:
Use AI to process the volume, use humans to decide the direction.

If you are still spending hours every week just summarizing tickets or checking survey results one by one, something is wrong.
You are throwing away your most valuable resources: your own time and focus.
Time that should be used to build strategy is instead spent doing robot tasks.

Which team are you on right now?
Do you still want to catch fish with a small line in the middle of the ocean?
Or will you start learning to cast a net that can catch all patterns at once?

If you want to learn how to apply workflows like this to change the way you work and boost your personal career, you can start by joining /ai-circle/. There, the members and I deeply discuss real world applications.

But if you are currently leading a team and you need a solid framework to increase your company efficiency from the inside, we can talk directly to dissect the initiatives at /corporate/.