Before You Redesign Your Website, Do You Actually Know What's Wrong?
Diagnose the business problem before the redesign — and what an AI audit agent can and cannot do.
Everyone’s discussing what the website should look like
Every time a website redesign comes up, the conversation tends to start in the same place.
Colors, layout, what competitors look like, whether there’s animation, how good the mobile version looks. These all matter. But they’re all downstream questions.
Before any of that, there’s a more fundamental question most people skip right past: what should this website bring the company?
Not “it should introduce the company.” Not “it should hold the product catalog.” But — what business problem should it solve? What concrete opportunity should it create?
Skip that question, and no matter how beautiful or flashy the rest turns out, you may just be renovating a house that’s facing the wrong direction to begin with.
I wanted to test whether AI could make this run more efficiently
The idea came from the same problem I’ve seen over and over across nearly twenty years of UX work:
How do you make this critical — but hard-for-most-people-to-value — early exploration phase run more efficiently?
Because this phase is usually the one that gets skipped or brushed over lightly — and yet it’s so often where a project’s success or failure is actually decided.
Over the past few months, I decided to use AI to build a semi-automated system — combining my experience mentoring newcomers and training researchers — to make the current-state audit of the exploration phase faster, with results tailored to each client but consistent in quality.
I built this prototype AI audit agent as an MVP, focused first on Taiwan’s B2B manufacturers. They typically have very strong products — top-two in a global niche market — but websites that do almost no work at all. The main reason: their sites usually weren’t designed to generate inquiries.
The first step wasn’t building the AI agent
Before training any intelligent audit-agent model, I did one thing first: I designed a questionnaire.
The reason is direct — the output quality of an AI audit agent depends on the quality of the input. And for the input to be meaningful, you have to ask the questions clearly first.
This questionnaire comes from years of my own experience in deep client interviews at the start of a project — the things you have to clarify before any analysis begins, such as: what is your website mainly trying to achieve? Which pages are most critical? What problems have you already noticed yourself? How do your overseas customers find you?
Letting clients fill it in themselves, and importing it into a data format the AI can use directly, replaces the traditional deep-interview-and-problem-framing process that eats up so much time at the start of a project.
The move looks simple, but it’s critical — because it aims the entire scope of the audit at the real business problem, not just the surface symptoms visible on screen.
What the AI audit agent can do
Once the questionnaire is filled in, the AI audit agent goes to work.
It applies the audit framework, analyzes the site structure, finds usability problems and patterns, and produces a prioritized report — which problems have the biggest impact, how they relate to the business goals, and where it recommends starting.
It’s fast. It’s highly consistent.
What does that mean for the client? No waiting through a long manual analysis cycle, and no wildly varying report quality depending on how experienced the person handling it is. Within a clear scope, every audit is consistent in depth and structure.
But the AI audit agent can’t judge what actually matters
During testing, I also saw its edges.
When the situation is clear, the output is very good. But once the context turns fuzzy — a more complex business model, or a finding that points somewhere the framework didn’t anticipate — the AI audit agent keeps running and keeps producing results, but can’t judge what this particular company actually needs to prioritize.
That’s exactly why our process isn’t just handing it to the AI and calling it done. The AI audit agent handles the analysis; a senior consultant comes in to review, add judgment, clarify context, and finally synthesize it into recommendations that can genuinely be executed.
The experienced human in the loop isn’t an add-on. It’s the difference between a report that’s actually usable and one that isn’t.
Fast isn’t the same as pointed in the right direction
This experiment made me more certain of one thing: the AI audit agent can make the whole process run faster. But it can’t get past this gate —
If you haven’t figured out at the start what business problem the website should solve, AI’s help may just be producing, more efficiently, a pile of findings that have nothing to do with the real problem.
That’s also why we put that questionnaire at the very front of the process — just as, in the past, whether a project succeeded often had a lot to do with how well the early deep interviews with the client were done.
After all, the measure of a project’s success isn’t whether the technical features are compliant, but whether it delivers the best possible effect. And that comes down to whether this one thinking gate is truly under control.
If your website has problems you can’t quite put into words
Maybe traffic feels okay, but the inquiries just aren’t coming.
Maybe you’re planning a redesign, but can’t articulate where the problem actually is.
Maybe the site has been around for years, but deep down you’re not sure whether it’s really bringing the company any opportunities.
These are all worth looking at clearly.
We offer an AI-augmented, expert UX audit service — starting from your business goals, finding the problems your website actually needs to solve, and delivering prioritized, executable recommendations.