Calmer, Smarter, Stricter: Inside the Upgraded ZenGrants Discovery Engine

We have some big news: ZenGrants is officially out of private beta and has launched as a fully paid platform. But before diving into the details of the release, I want to say a massive thank you to everyone in the UK startup community, particularly on Reddit, who helped stress-test our early builds. Your unvarnished feedback and bug reports directly shaped this engine, and we are incredibly grateful.

Adapting to the Restructure

A few weeks ago, we had some initial concerns. When Innovate UK announced their sweeping structural overhaul to paused SMART grants and began introducing new sector-focused competitions, rumours swirled that they were shifting to interview-only scoring. As someone who had just spent months building a structured grant-writing engine aligned with the old formats, I was worried that months of development work might have been for nothing.

However, after spending a week analysing the newly released assessor guidelines for the Breakthrough Next Wave rounds, I realised the reality is quite different. The core requirements haven't vanished; they've just become far more granular and strict. Rather than breaking our tool, this shift made it clear that generic ChatGPT drafts are going to be rejected faster than ever.

To help founders navigate these tougher guidelines, we spent the last fortnight completely overhauling the ZenGrants Discovery Phase. Here is what is new and how it works.

1. Unsparing Assessor Scoring

Standard AI tools are terrible at grading grant drafts because LLMs are naturally polite. If you ask a generic chatbot to review your market analysis, it usually replies with some variation of "This looks great!"

Real Innovate UK assessors are not polite. They are looking for reasons to mark you down. To replicate this, our upgraded Discovery assistant now adopts an unsparing assessor persona. It assumes a low baseline score and demands concrete evidence, such as quantified market sizing, specific subcontractor costs, and risk mitigation plans, before you get a pass mark.

2. Transparent Score Deductions

We've removed the mystery behind readiness scores. When the assistant evaluates a draft, it doesn't just give you a high-level percentage. It now shows you a detailed breakdown of exact deductions. For example, if your score drops because you didn't define a UK-specific benefit or missed an EDI risk, you will see the exact reason and clear instructions on how to reclaim those marks.

3. Step-by-Step Focus

Grant writing is overwhelming, and looking at ten blank boxes can cause immediate writer's block. To fix this, the assistant now acts like an active consultant. It evaluates your entire application draft and highlights exactly one weak section to target next. By breaking the bid down into individual, bite-sized tasks, you can progress methodically without feeling overloaded.

4. Full-Text Document Ingestion

Most AI tools read long PDFs by summarizing them first, which strips out technical specifications, specific financial figures, and product nuances. We upgraded our backend ingestion to read the unabridged text of your uploaded documents from the very first run. The AI captures the subtle metrics buried deep in your source files, ensuring your drafts reflect actual data, not generic placeholders.

5. Mid-Chat File Awareness

A common pain point in early betas was uploading a document halfway through writing and having to reset the chat. The assistant is now fully aware of newly uploaded files dynamically during active chats. It will automatically detect a new upload, incorporate its context, and prompt you to explore its contents within the same session.

6. Deferred Gaps to Deep Research

Some questions cannot be answered immediately because they require deep competitor analysis or market sizing that you haven't done yet. In the new Discovery brief, you can tag up to two gaps to temporarily waive their score deductions. This allows you to focus on polishing your core proposal now, and our dedicated Deep Research agent will run search queries to find the missing details in the background later.

Try ZenGrants

We've officially transitioned out of private beta and launched ZenGrants as a paid platform. But as a huge thank you to the community that helped us shape and stress-test the product, we want to give back: if you're currently preparing an application for the new Innovate UK rounds and want to try out our new post-restructure engine, drop us a line or DM the founder directly for a full 2-week free trial.