AI, Analytics and the Art of Asking Better Questions

Most businesses already have more data than they realise. Website analytics, product analytics, finance data, CRM notes, support tickets, sales conversations and internal reports all contain useful signals about what is happening.

The problem is rarely that the data does not exist. The problem is that people do not always have the time, head space or confidence to interrogate it properly.

Google Analytics might show where visitors came from. Amplitude or Mixpanel might show what users did inside a product. Xero might show what is happening commercially. Support tickets might reveal where customers are getting stuck. Each system gives part of the picture, but the more useful insight is often found between them.

This is where AI becomes genuinely useful. Not as a magic answer machine, and not as a replacement for experienced people, but as a way to make data easier to question.

Dashboards are still valuable, but they are usually passive. They show what has already been defined as important. That is helpful when you know what you are looking for, but less useful when you are trying to understand why something has changed.

AI changes the starting point. Instead of waiting for someone to build a report, you can begin with a question. Why did conversion drop? Which traffic sources are bringing users who actually convert? Where are people dropping out of onboarding? Are support issues linked to a product change? Which clients or services are taking more time than they are worth?

The value is not just speed. It is the ability to keep asking better follow-up questions.

This is already becoming visible in the tools many businesses use. Amplitude and Mixpanel are moving product analytics towards more conversational workflows. Google Analytics 4 can surface generated insights around meaningful changes in website performance. Claude can analyse data and create charts directly inside a conversation. ChatGPT can explore uploaded files, structure messy information and turn data into summaries or recommendations. Gemini is increasingly relevant for organisations already working across Google Workspace, Google Cloud, BigQuery, Looker and GA4.

The wider shift is connected AI. Tools like Claude, ChatGPT, Gemini and Codex are increasingly able to connect with external systems, documents, codebases, design tools and business platforms. MCP, or Model Context Protocol, is part of that story. In simple terms, it works like a universal adapter for AI tools, giving them a more standard way to connect with external data and services.

Most businesses do not need to care about the technical detail. They should care about the direction of travel.

AI is becoming less like an isolated chat window and more like an interface to the systems where the business already operates.

That matters because useful business insight rarely lives in one place. Imagine a subscription product where website traffic is up, conversion is down and support tickets are increasing. GA4 might show that a new campaign is driving more visitors. Product analytics might show that users are dropping out during onboarding. Support tickets might reveal confusion around a recently launched feature. Revenue data might show that the traffic increase is not translating into better customers.

The useful insight is not inside one dashboard. It sits between analytics, product behaviour, customer feedback, support, finance and human context.

AI can help join those dots more quickly, but it should not be treated as a source of truth. It can misread a metric, trust broken tracking, confuse correlation with causation or create a polished chart from a flawed question. If the data is messy, inconsistent or poorly defined, AI will inherit that confusion.

Good data habits still matter. Clear metric definitions, sensible event naming, clean exports, appropriate permissions and shared understanding across the team become more important, not less. As tools become more connected, businesses also need to think carefully about privacy, access, governance and how outputs are checked.

The best way to start is not by connecting everything. Start with one useful question.

Pick an area of the business where data already exists but is rarely explored properly. That might be website performance, product onboarding, sales conversion, client profitability, support issues or monthly revenue. Export the data, ask AI what changed, ask what looks unusual and ask what it would investigate next. Then check the response like you would check the work of a junior analyst.

If the output is useful, you have found a workflow worth improving. That might lead to cleaner tracking, a better dashboard, a more useful report or eventually a direct connection between AI and your business systems.

This is where we can help. At Little Thunder, we work with teams to understand what data they already have, where it lives and what questions it should help answer. That might mean improving analytics tracking, connecting data sources, designing clearer dashboards or creating AI-assisted workflows that make insight easier to access.

The goal is not to add another layer of complexity. It is to help teams make better decisions with the information they already have.

The opportunity is not to automate judgement. It is to make better questions part of the way your business works.

The businesses that get the most value from AI will not necessarily be the ones with the most data. They will be the ones that learn how to question it better.