Introducing Phronel. Private intelligence for early-stage company decisions. Live 1 November.

Introducing Phronel v1.0. Live 1st November.

I was at a Russell Group university recently, waiting to present to around forty people from the investor community. Funds, accelerators, angels, university venture teams. The kind of room where everyone’s politely half-checking their phone until something big lands.

The chair opened the session. Before I’d said a single word, he said this:

“Please, do not use general-purpose AI to carry out diligence on companies.”

Forty people nodded.

Then I stood up.

I’ve had a lot of introductions over the years. This wasn’t technically one, but it was certainly my favourite.

He was right, and everyone in that room knew it.

This wasn’t an anti-AI room, It was full of people already using AI every day. And it wasn't a Hugging Face shockwave, it's much bigger than that. Data privacy remains a fundamental concern, actually the no.1 concern among venture capitalists adopting AI according to the VC Tech Pulse stack published earlier this year.

Ultimately, more companies are entering the ecosystem. Teams are drowning in information and Institutions need to find the strongest companies faster.

The problem isn’t that they’re using AI, but It’s the AI that they’re using.

Ask a general-purpose model to assess a company today and again on Thursday, and you can get two different answers. Without a consistent evidence standard underneath the output, you can’t confidently compare one company against another. You can’t defend a conclusion to an investment committee or an LP if you can’t show how you reached it.

And then there’s the part nobody enjoys talking about. If you’re pasting a confidential pitch deck, unpublished research or data-room material into an AI tool without knowing how that information’s handled, that’s a red flag. It’s a red flag to me, and it’s very much a red flag to the founder whose information you’ve just uploaded.

Just take a step back for a moment. You’re a PhD student, and you’ve committed years of your life to researching something. You’ve uncovered a fundamental need and decided to build a business around research that isn’t yet published.

You share it with an investor, trusting them to treat it confidentially. They upload it to an AI tool.

It isn’t as black and white as “AI learns your business model”. It’s about where that information goes, who can access it, how long it’s retained, and whether you ever agreed to it being processed there.

You’ve trusted someone with years of your work. You shouldn’t have to wonder what they’ve done with it.

So, I introduced Phronel to the floor… prematurely of-course.

Phronel is the intelligence layer that powers Caplia. It’s a model built for one job - the secure evaluation of early-stage companies.

We launched Caplia in April as the enterprise platform that helps funds, accelerators, banks and universities make better decisions on early-stage companies. The question we got, every single time, was what actually powers the intelligence underneath. The CRI, Iris, the deep research.

The answer’s Phronel, and from 1 November you can use it directly.

Give it your thesis and it calibrates to you. Your mandate, your screening prerequisites, your definition of a good company. In testing, we’re seeing 87% alignment with human judgement. That number still makes me sit up.

Every reading is standardised and evidence-backed. Tuned for early-stage company evaluation, consistent signals, consistent frameworks, and every claim tied back to a source. Assessments follow the same standard, so you can compare companies and understand what’s changed. That’s the whole point.

It’s private, and that matters more than anything else on this list. Phronel runs on our own infrastructure, based here in the UK and soon in the US, or on your own if you’d rather. It doesn’t learn from your prompts or founder data. Your pitch decks, sensitive materials and data-room information stay within the deployment environment.

And it’s agentic. Phronel already lives in my Slack. It also works through Teams, a Chrome extension, you can build with it using our API and even our MCP if you seek guardrails, which means it sits inside the systems where your decisions already happen.

Want to check readiness or thesis fit? Done. Want to run a P15 (a real time traction index that measures 15 signals) to surface contradictions and traction? Easy. At a demo day and want to assess the company presenting? Take a photo of its slide. Want pre-diligence, or a read on the current state of a company you already back? Phronel does that too. Want to use it like you would use Claude? Run deep research inside your Slack. Easy.

What comes out is structured, evidence-backed reporting on an early-stage company. Reporting you can scrutinise, challenge and take into a discussion with senior management or LPs.

What happens next

In October, our first empirical study begins with Oxford University (Oxford Caplia Investment Readiness Study, OCIRS). It lays the foundations for a continuous recalibration loop across the frameworks that sit on top of Phronel, using research and outcomes to refine how companies are assessed. That’s the part I’m most excited about.

Phronel launches on 1 November 2026.

If you’re making decisions on early-stage companies and you’re still screening them with a general-purpose assistant, we’ve built Phronel for you. Build your systems with it, or load it into the workflows you use today.

Purpose-built. Evidence-backed. Consistent. Private.

We've been building with the best, and we're blown away. We can’t wait to put it in your hands.