Clue Labs v3 launches in May 2026.
It’s the best thing we’ve ever built, and it’s taken longer to get here than I planned, cost more than I budgeted, and required a level of architectural thinking that I couldn’t have articulated two years ago.
If you’ve been watching from the outside and wondering what was happening; this is the answer.
Where We Started
The original Clue Labs launched on a thesis that I still believe completely:
Social platforms have rebuilt themselves around AI-driven discovery, and the entire social media tool industry is still selling solutions designed for the old model.
The beta proved the thesis had legs. We hit £10k MRR within fifteen minutes of opening the doors. That’s not a typo. The problem we were solving was real and the demand was there.
Then things stalled.
Not because the thesis was wrong.
Not because the market moved.
Because building a platform that genuinely models how AI discovery algorithms work is technically harder than building another scheduler or analytics dashboard; and the development team we had at the time wasn’t the right team for what we were trying to build.
I won’t get into the specifics.
What matters is that we lost time, lost momentum, and lost customers who had believed in the vision early and watched the product fail to deliver on it.
That stings. It still does.
The churn during that period wasn’t a metric problem.
It was a trust problem.
People had seen what Clue Labs could be and then watched it stop moving. I understand why they left.
What Was Happening Behind the Scenes
Here’s what the churn data didn’t show: while the visible product was stagnant, the intellectual architecture behind it was getting more rigorous, not less.
The intelligence that existing Clue Labs users have been working with; the pattern recognition, the discovery signal analysis, the prescriptive recommendations.
That was real and it was working.
The results our customers were seeing weren’t placeholders. They were the product doing what it was built to do.
What we were working on wasn’t replacing that. It was figuring out how to take it further than the original codebase could support.
The question I kept returning to was this: if I’m building a platform that’s supposed to model how social discovery actually works, what does that model look like when it’s fully realised?
Not at the level of “engagement signals matter” but at the level of how data flows across an entire account history, what it connects to, how the system learns continuously, and how you build something that compounds in intelligence over time rather than plateauing.
That work took the form of a complete architectural redesign, what we now refer to internally as the twelve-district city map. Think of social media performance data as a city.
There are districts: the content itself, the timing, the audience behaviour, the platform mechanics, the account history, the competitive landscape.
The current platform models these well.
What we’ve been building is a system that understands how all twelve interact with each other dynamically and learns from those interactions at a level that a traditional analytics architecture can’t support.
The next layer being added is a Graph Neural Network – a GNN. Unlike analytics that treat each post as an isolated data point, a GNN understands posts as nodes in a network.
It models relationships: between posts, between content attributes, between audience behaviour patterns, and between platform-level signals. It learns the topology of what works on a specific account and gets more accurate over time.
This is the architecture that takes what we already do well and makes it genuinely compounding.
Alongside this, we built out a platform adapter system; a structured way of adding new platforms without rebuilding from scratch each time. Instagram first, then LinkedIn, then everything else.
The adapters mean that when we expand, we’re extending something that already knows how to learn.
None of the architectural work is visible to users directly. That’s the nature of infrastructure. You can’t see the foundations of a building, but you feel it when they’re built for scale; and you feel it when they’re not.
The Decision to Rebuild
There was a point where we had to make a choice. We could keep iterating on what existed and ship incremental improvements, or we could treat the architectural work we’d done as the real product and build v3 properly, even if that meant more time.
The incremental path was safer in the short term.
It would have been faster.
It would have stopped the churn sooner.
But it would have meant building the wrong thing faster.
And I’ve spent enough time in this space to know that shipping a product with the wrong foundation is a trap. You end up in technical debt that slows you down for years.
You end up with a product that looks like the competition and performs like the competition; which is to say, it works, but it doesn’t do the thing that makes Clue Labs worth building in the first place.
So we rebuilt. We brought in the right technical leadership. We documented the IP properly — the GNN architecture, the discovery model, the platform adapter system, the relevance scoring logic. We built the kind of codebase that a company can actually grow on.
That decision cost us time and customers. I’m at peace with it.
What v3 Actually Is
Clue Labs v3 is not a feature release. It’s the first version of the next level of the platform where the thing users interact with and the thing powering it underneath are finally aligned.
On the surface: a cleaner onboarding experience, a freemium model that removes the commitment barrier for new users, deeper AI-generated campaign recommendations, and a more honest interface that shows you what your data actually says rather than what you want to hear.
Underneath: a rebuilt data architecture designed to support the next layer of intelligence we’re adding — the Graph Neural Network that will model your account as a network rather than a list of posts. A relevance scoring system that tracks how discovery algorithms are currently behaving, not six months ago. An AI layer that synthesises all of this into a recommendation specific to your account, your goals, and your audience, with the reasoning shown so you can learn from it. The intelligence you’ve been using is still here. v3 gives it a foundation to go further than it could before.
The freemium model matters beyond the commercial logic. One of the consistent frustrations I’ve had building in this space is that the brands who most need to understand discovery mechanics — the ones with limited budgets, operating without agency support, making strategic decisions alone — are the ones least likely to pay for tools upfront.
The new model means they can experience what the platform does before they have to commit.
That feels right to us.
Why This Matters for the Category
I coined the term Social Discovery Optimisation because there wasn’t a language for what we were building.
SEO has decades of frameworks, tools, and education built around it.
SDO has me and a growing number of practitioners who’ve quietly figured out that the old playbook is broken.
v3 is the first time Clue Labs is technically capable of delivering on the full SDO promise.
Not a version of it.
The actual thing.
What that looks like in practice: an account connected to Clue Labs should be able to post three times a week and achieve results that would previously have required posting ten times a week; because every post is precision-placed, properly sequenced, and mapped to how the discovery algorithm is currently operating.
Not in theory.
Measurably, in the data.
Post 3x, perform 30x.
That’s the north star.
It has been from the beginning. v3 is the next version of the platform where we can back it up.
To the Customers Who Left
I want to say this directly: I know some of you believed in this early and watched it stall.
The churn during the rebuild period wasn’t invisible to me it was personal. Every cancellation represented someone who’d taken a bet on a vision and felt let down.
I’m not asking you to come back on faith. I’m asking you to look at what we’ve built and make the decision based on that.
The product is ready.
The architecture is solid.
The team is right.
And the problem we set out to solve — the guesswork, the burnout, the disconnect between effort and result — hasn’t gone anywhere.
If anything, it’s bigger than it was when we started..
v3 launches 1st May 2026 →
