The real value of AI in analytics isn’t just answering questions. It’s closing the decision loop.

AI is great at answering questions. But is it helping you make faster decisions?

If you only looked at launch pages and LinkedIn feeds, you'd think the future of analytics is one big chat window. Type a question in plain English, get a chart, call it "AI-driven insights". There's a reason it feels so compelling. For years, getting answers from data meant waiting on someone who knew the warehouse, the schema, and just enough SQL to not break production. A chat interface looks like a way around all of that. Finally, anyone can "talk to their data".

And that's a genuine step forward. But if you've spent time in the guts of how decisions actually get made, you know it's only one part of the story. The question isn't "Can I talk to my data?" It's "Does this actually change how fast and how confidently my organisation learns?"

Conversational interfaces make data more accessible. But accessibility alone doesn't close the gap between seeing a problem and doing something about it.

Dashboards are unglamorous (but essential)

Let's start with the thing chatbots are making more accessible: dashboards. They might not be the hero of any product announcement, but they still sit at the centre of how most teams understand their world.

With modern tooling and a bit of AI assistance, you can throw together a decent growth experiment dashboard in 15 minutes. I've done exactly that – "vibe‑coding" something from a rough mental model into a working view over lunch. That used to be impressive, but now it's just the baseline.

Dashboards shine in a very specific context: when you already know what matters. Before building any dashboard, you’ll have done the yardwork to know what metrics matter, what recurring questions you get from the business, and what types of decisions you need to make week after week. The dashboard becomes a shared communication layer, where everyone knows where to look on a Monday morning, and what “good” looks like.

When numbers drift from the goal, it's obvious. That kind of ambient awareness is hard to recreate inside a one‑off chat exchange, no matter how smart the model is.

So yes, AI is changing how dashboards are built and how people interact with them – and conversational interfaces are a natural companion, letting you dig into what the dashboard surfaces without switching tools. But the idea that dashboards will be replaced by a single conversational interface misunderstands what they're for. Dashboards exist to make a complex system legible at a glance. That job doesn't disappear just because we can talk to our tools.

Natural language is a real improvement, but it's not the whole picture

Natural language interfaces are genuinely useful. Instead of writing a five‑line query to check how revenue trended in Canada after a specific launch date, you can ask your chat interface in plain English and get something sensible back. There’s less friction. And in financial platforms, where teams juggle multiple currencies, entities, and regulatory contexts, that ease compounds quickly into productivity gains.

But it's important to be honest about what's actually being improved. Natural language lowers the cost of asking. It makes it cheaper – in time, skill, and context‑switching – to pull a slice of data from your business. That's especially valuable for teams where the person with the question isn't the person with the SQL skills. But what chat interfaces don’t do, on its own, is make you any better at knowing what to ask.

You still have to frame the question. You still have to decide which cohorts matter, which time windows are relevant, which comparisons are meaningful and which are noise. You still need the judgement to look at a chart and say, "This doesn't actually mean anything."

If the output of your "AI analytics" is still just a chart or a table, you haven't transformed the system. You've removed a bottleneck in the request/response loop. This still matters. But that’s just improved how you access information, not how you make decisions. And that raises the question: how should we best use AI in analytics, if the answer doesn’t quite lie between the ease of dashboard creation and chat interfaces?

Agentic analytics: From answering questions to closing the decision loop

To answer this question, we need to shift from tools that answer questions to systems that can do part of the thinking for you – this is what I call agentic analytics.

Think about how most organisations learn today:

1. Something happens in the market or the product.

2. Someone notices it, often late.

3. Someone frames a question and pulls data.

4. Someone interprets that data and proposes an action.

5. Someone approves it.

6. Eventually, someone checks to see whether it worked.

There are delays and drop‑offs at every stage. The net effect is that the time between something changing and your team responding can stretch into weeks or months.

Agentic analytics, in whatever form they take, compress that timeline. They don't just wait for you to ask, "How did our latest campaign perform?" They:

• Live inside your data and understand what your business cares about

• Know what normal looks like, and what's changed

• Watch for shifts that are worth paying attention to

• Suggest specific things to try

• Track the results and get better over time

In other words, they start to behave less like a search box and more like a teammate whose job is to watch the numbers, flag what matters, and help you move faster. This behaviour represents the shift to agentic analytics – systems that don’t just answer questions, but reason, act, and learn on your behalf, with humans governing the process. That's ‌fundamentally different from chatting with your dashboard, though conversational UI may well be one of the ways these agents will talk to you.

The quiet metric that matters: Decision latency

One metric that captures what's really at stake is decision latency. That is, how long does it take for your organisation to notice something important, decide what to do, and execute?

Dashboards help by making important numbers visible to everyone. Conversational interfaces help by letting more people pull custom views and spot anomalies without a BI bottleneck. But the ceiling on both is still defined by how quickly humans can look at the data, make sense of it, agree on a response, and act.

As AI systems get better at understanding the business context behind a change, reasoning about what probably caused it, running or simulating experiments to test a response, and measuring whether that response actually worked, decision latency can fall dramatically. That lets you move from monthly to weekly or even daily actions.

But decision speed on its own comes with pitfalls. If you optimise purely for how quickly an agent can suggest and ship a change, you risk making faster mistakes. Good agents don’t just have access to data; they sit on top of context and layers of governance.

In other words, good agents understand the business context behind the numbers, tailor actions to different situations, and interpret and adhere to policies and best practices. Without that foundation, you could end up making bad decisions faster – albeit with more confidence.

The distinction between “knowing” and “doing” matters

It's tempting to treat "AI for analytics" as a UI problem: bolt a conversational layer onto your BI stack and move on. The risk is that you pour your AI budget into making it easier to ask questions, without seeing how long it takes to act on the answers.

A more useful starting point is to work backwards from decision latency. Ask yourself:

• Where in our current process does learning stall?

• Which decisions are too slow, too noisy, or too dependent on one person's knowledge?

• What would it take for an AI system not just to answer "what happened?", but to help figure out "what should we try next?" and "did it work?"

If you start there, chat becomes one powerful tool in your overall strategy. Dashboards will keep doing what they're good at: giving everyone a shared view of the truth. Conversational interfaces will keep making data less intimidating and easier to act on. But the real advantage will come from the systems you build around them: the agents that work in the background to speed up the time between something showing up in your data and your team doing something about it.

AI in analytics isn't about replacing dashboards or building the cleverest chat interface. It's about shortening the distance between knowing and doing, and trusting that decisions are grounded in the right data and context. This is how you set up your organisation to learn reliably at the speed of its data, and not the speed of its meetings. Done well, that’s what agentic analytics looks like.