How a terrible WordPress plugin, an army of GPTs, and an 80-page PDF report led to building six assessment platforms – solo, in months.
I knew exactly what I wanted to build.
That’s the thing people don’t always understand about being a non-technical founder. The problem was never the vision. I could see the whole thing – the assessment, the questions, the scoring logic, the report, the user experience, how it would feel to go through it. I had the model. I had the methodology. I had fifteen years of watching humans change, mapped out in a half-finished Google Doc spreadsheet.
What I couldn’t do was build it.
Or more precisely: every time I tried, the technology blocked me. Not because I’m not technically capable – I’m more comfortable in a cPanel server than most people would expect. But because the tools available to non-developers weren’t built for what I was trying to do. They were built for quiz lead magnets. Ten questions, a paragraph of results, a form capture. Fine for some people.
I wasn’t trying to build a quiz lead magnet. I was trying to build a psychological assessment platform.
The plugin that broke me
I won’t name it. That’s not fair – it does what it does, and for simpler use cases, it works. But for what I needed, it was a nightmare.
The UX made no sense. Information that belonged together was split across different screens. Things that should have been adjacent – the question builder, the scoring logic, the report formatting – required constant jumping between windows, losing your place, starting again. I’d raised so many support tickets I was practically on first-name terms with the founder.
I come from a category management background. In retail, you put pasta next to tomato sauce because that’s how people shop. You design around the user’s natural decision-making journey. You don’t make them walk to the other end of the store to find something that belongs with what’s already in their basket. This plugin hadn’t been built with any of that thinking. It had been built by someone technically brilliant who hadn’t prioritised the experience of the person actually using it.
And the report functionality. Oh, the report functionality.
There was no way to preview what your report would look like without going through all forty questions first. Every time. The merge tags – the codes that pull in personalised results – were documented somewhere, presumably, but finding them and getting them to work the way you expected was a mystery I never fully solved. Bold formatting in one place would bleed into body copy somewhere else. Centred headings that wouldn’t stay centred.
For a simple ten-question lead magnet, none of this would matter. For a forty-question psychological assessment with personalised, nuanced results for five different growth archetypes – it mattered enormously.
I stalled. Repeatedly. Projects that should have launched sat waiting because the tool just wasn’t up to the task.
The workaround that was also a nightmare
At some point I decided: fine. I’ll ditch the plugin’s report functionality entirely. I’ll just use it to collect the answers, pull the results into GoHighLevel, and build the reports myself using AI.
By this point I had – and I say this with equal pride and retrospective exhaustion – an army of GPTs.
Eight of them, each with a very specific job. One owned the model and the archetypes. One built assessment questions. One analysed results. One translated the analysis into consumer-facing language. One handled tone. One managed publishing and content. I’d learned early that AI loses quality fast when you give it too many things to do simultaneously, so I’d broken the whole process down into a chain – output from one GPT became the input for the next.
It was, in its own chaotic way, systems thinking. Which is how my brain works. I just didn’t have the tools yet to make it elegant.
So here’s what producing one report actually looked like:
Candidate completes forty questions in the plugin. Results come through to the backend. I copy the results table. I paste into GPT one – the analyser. I take that output. I paste into GPT two – the results interpreter. I take that output. I paste into GPT three – the report writer, which I have briefed extensively with a template, a sample report, specific section headings, tone guidance, formatting instructions.
And ChatGPT, bless it, just… doesn’t listen.
I’m giving it the template. I’m giving it the sample. I’m giving it everything. And the output is inconsistent Every. Single. Time. Different lengths. Different structures. Ignoring the formatting I’ve specified. I’d spend more time correcting the report than it would have taken to write it myself.
Then I’d take whatever I had and open Canva.
I cannot stand Canva. I know everyone loves it. I find it deeply counterintuitive – you can’t just paste formatted text in and have it behave. The layout fights you. Moving elements around is infuriating. I’m much happier in Pixelmator where I have real flexibility. But Canva was what we were using, so Canva is what I battled.
The end result: an eighty-page PDF report.
People said it was brilliant. They said it resonated. And genuinely, the content was good — because the model behind it was solid and the fifteen years of observation behind that were real. But eighty pages? For a psychological assessment result? Nobody should have to wade through eighty pages. And I knew it. The delivery mechanism was completely wrong for the content.
But it was the best I could do with the tools I had.
The day Supabase appeared
When Claude Cowork launched and I started building the Clearance Club app, Claude said: we need to set you up on Supabase.
I looked at the words SQL table and felt something close to panic.
I did a Facebook post about it at the time — something about suddenly finding myself in a foreign country where I didn’t speak the language. Claude was giving me instructions that assumed a level of technical knowledge I didn’t have, and I was just… trusting. Which requires a particular kind of courage when someone’s setting up database tables on your behalf and you genuinely don’t know what a database table is at home, even if you vaguely know what databases are in theory.
But I trusted. And it worked.
And then when it came to building the first proper assessment platform – the PRAXIS therapist assessment – Claude said: we’ll use Supabase. And I said: oh, I’ve already got an account. Check me!
That moment. That tiny, almost insignificant moment of “I’ve already got an account” – that’s what progress actually feels like. Not a dramatic breakthrough. Just the slow accumulation of capability, one build at a time.
What became possible
Six assessment platforms. Built properly. With real databases, real scoring logic, real automated report generation, real PDF pipelines, real CRM integration.
No more copying results tables and pasting them into chains of GPTs. No more battling Canva. No more eighty-page PDFs nobody asked for.
Instead a fully automated workflow.
Assessments that auto-score, auto-generate reports, auto-deliver to the candidate, and write the results back to the CRM. Dashboards instead of static PDFs. Interactive results instead of walls of text. The whole thing running on infrastructure I built myself, that I understand well enough to maintain and improve, that doesn’t depend on a plugin that might break next time someone pushes an update.
The content is the same. The methodology is the same. The fifteen years of observation behind the model is the same.
What changed was my ability to build the vessel it deserved.
What I actually learned
There’s a version of this story where I wasted years fighting tools that weren’t built for what I needed. And I did. That part is true and it was genuinely frustrating. I spent months doing things that just went straight into the bin.
But there’s another version where those years of frustration taught me exactly what I needed to build – and gave me the clarity to build it properly when the tools finally caught up.
I knew what good looked like before I could make it. I’d been a brand director, briefing agencies, working with designers, thinking about user journeys and decision trees and category management for twenty years. The vision was always clear. The technology was the gap.
AI closed that gap. Not by replacing my expertise – the model, the methodology, the understanding of how humans change – but by giving me the capability to finally build what I could already see.
Here’s what I think is true, and I think it matters:
If you have vision, if you’re a creator, if you’ve got an idea and you can see what it should be – it’s your time. The tools now exist to build what you can see, without needing to hire a team of people to translate your vision into something real.
If you’re the person who used to do that translating – the executor, the implementer, the technical layer between someone else’s idea and its realisation – the equation has shifted. Permanently.
The visionary no longer needs to wait.
Ladder of Growth assessment platforms are available for coaches, therapists, HR teams, and organisations. Explore at ladderofgrowth.io
Alexia builds AI-powered assessment infrastructure and systems for founders and organisations. Work with Alexia
