Chatbots aren’t the answer

“Everything should be made as simple as possible, but not simpler” - Einstein (aka the most mis-quoted person ever)

We started Count.co a shockingly long time ago, almost eight years to the day in fact.

In the startup world (if you can even call us that), that makes us a bit of an outlier. We’ve kept going thanks to a blend of stubbornness, a dash of arrogance and a total obsession with the problem of analytics.

We started the business because we saw the near-universal adoption of an interface we believed was fundamentally flawed, the dashboard. Dashboards are an incredibly crude way to understand your business, they block collaboration, trap data teams in a service role, and yet, within this billion-dollar industry, almost everyone had accepted them as the best way to see and work with data.

So after taking that big gulp of founder naïve optimism we set out to find a better solution and failed, many times. But finally, a few years ago, we did find the answer - the canvas. Collaboration first and flexible, it was (and remains) a totally new way to work and see data. We clearly got something right because Count is now the highest rated analytics tool on G2.

But now in 2025, looking at the analytics market, I feel like we’re back at the start. As the industry rushes to embrace the promise of AI there once again seems to be a universal adoption of a single interface for AI + data, and that’s the chat thread and it feels…. unsatisfactory.

Chat threads are dead… I’m not falling for it this time

To be fair, after many years, my view on interfaces is slightly more nuanced than the early days. I don’t think there’s just one perfect interface. It’s much more about having the right interface for the right use case.

Lots of our customers at Count for example, use it to build dashboards. And that’s absolutely fine, because they also build slide decks, metric trees or just dump charts into a big collaborative canvas - whatever they need to make a good decision as fast as possible. The key is to have the right interface for the right workflow (or for the right decision in the case of analytics).

The problem is the chat thread is almost certainly the wrong interface for working with data and almost certainly wrong for many other use cases too. We’ve known this since 2014 when Slack led us into thread land with the promise of custom emojis.

Obviously, for a new technology which you can talk to, a chat thread is a very natural place to start but that doesn’t make it the right interface. It makes it the safest for people to copy.

There’s a reason, for example, you don’t design your house with an architect by whatsapp. You may use chat to arrange a meeting but it’s not the place to discuss the project, work through design complexities and have confidence you’re on the same page. A chat thread just doesn’t deal with that kind of visual information density.

I’d argue analytics is more analogous to designing a house than booking a meeting.

Yes, I do think a canvas is better, but in your heart you do too

So if we know that chatbots aren’t necessarily the right interface for all agentic use cases - including analytics - the big question is, what is the right interface?

Yes, having spent almost a decade of my life looking for a better analytical interface and then building it for the last 3, yes I have a bias. But I genuinely believe, whether by chance or design, that the canvas is one of, if not the best interface for agentic analytics and possibly beyond.

Here are three reasons why:

1. Flexibility

I asked ChatGPT to generate an image which represented the concept of flexibility and freedom of thought and this is what it gave me. This does kind of look fun doesn’t it.

As AI has become more powerful it’s becoming clear that the limiting factor is no longer the model but the model’s ability to communicate with it’s human master.

I can already ask an AI to bring back information on every single user in my business or every single article on analytics grouped by themes the challenge is how limited the AI is at presenting all this information back to me in the most efficient and valuable way. The human <> AI interface is the weak link, holding back how well the AI can present ideas, help humans think and ultimately creating a more natural interaction fluid interaction between AI and human.

Having an interface which gives an AI room to breath where it can show it’s full power and potential is the only interface which really makes sense given the trajectory we’re on.

2. Collaboration

This is clearly not the ultimate work set up.

Collaboration is what humans need. If we look at all the best in class tools just before the AI race began - Figma, Notion, Slack as examples they all brought real-time collaboration to the fore because ultimately humans work and make decisions together.

Right now, most of our AI workflows are single-player. That was ok to start but it’s not optimum. At their core BI tools and LLM apps are actually very similar. Both are information retrieval systems, BI for qualitative data an LLM app mostly for qualitative data. The big dirty secret in BI which is now becoming true for LLM apps is that information retrieval or “insight creation” hasn’t really been the bottleneck - it’s been in the decision-making workflow that follows.

The best way to use AI is to use it with others, to support creativity and discussion and turn that into a consensus and a case for change. Any use case which involves humans using information to make decisions should be biased to collaboration as much as possible.

3. Transparency

Anyone who’s worked in data knows analytics is all about trust. Giving the AI detailed prompts and context is definitely part of the answer but they can ultimately only go as far. Eventually someone will ask something the prompter hadn’t thought of and chaos will ensue.

The catch-all design principle you also need is transparency. Any peer-reviewed scientific paper is only accepted if the method of any experiment is explained alongside the results. The results and the method have to be seen together for the results to be trusted and the same is true in analytics. It’s something perhaps more unique in analytics than other software industries like code or design where (mostly) the quality of the output is all that matters.

An interface where workflow and outcome can be shared together is the right interface for complex AI workflows.

Over the past few years our users have shown us the power of these three principles for turning data into value and for this reason and those above we’re incredibly bullish on the canvas as the foundation of an AI-first analytics workflow.

We’ve already seen why dashboards weren’t the right answer, and we think the industry is now defaulting to chatbots out of convenience rather than good design thinking.

Stand by for something special…