The AI is very good at answering questions. But can it help you make decisions faster?
Judging from launch pages and LinkedIn feeds, the future of analytics looks like a huge window of discussion. Just enter a natural language question to get a graph and call it “AI-powered analysis.” The idea is attractive. For a long time, getting answers from data meant finding someone who understood the data warehouse, schema, and enough SQL to not disrupt the production environment. A chat interface appears to be the perfect solution to get around all these obstacles. Now everyone can “talk with their data”.
This progress should not be minimized. But anyone who has delved into the nuts and bolts of decision-making knows that the story doesn’t end there. The question is not whether you can talk to your data, but whether it impacts how quickly and reliably the organization learns from it.
Conversational interfaces make data more accessible. However, simply accessing data is not enough to move from problem detection to resolution.
There’s nothing sexy about dashboards, but you can’t live without them.
Let’s start with the tool that chatbots make more accessible: dashboards. This is not what we highlight during product launches, but they nonetheless remain a central element of understanding for the teams.
With modern technology and a little help from AI, one can create an acceptable dashboard for growth experiments in fifteen minutes. This is what I did: bring an idea of a model to life using vibe-coding over lunch. It used to be an impressive performance. Today is the basis.
Dashboards are particularly effective when you are already aware of what matters. Before creating your dashboard, you need to know which metrics are important, which questions come up regularly, and what types of decisions you need to make week after week. The dashboard becomes a shared communication tool, where everyone knows what to look for on Monday morning and what a good result looks like.
When the numbers deviate from the target, it is visible. This general perception is difficult to reproduce in the context of a one-off exchange with a chatbot, regardless of the quality of the model.
Certainly, AI is transforming the way we design and interact with dashboards and conversational interfaces are natural companions that help dig beneath the surface of the dashboard without changing tools. But to believe that dashboards will be replaced by a conversational interface is to misunderstand their function. Dashboards are used to make a complex system readable at a glance. This necessity doesn’t disappear just because we can converse with our tools.
Natural language is a real improvement, but it is not everything.
Natural language interfaces have real utility. Instead of writing a five-line query to check revenue trends in Canada after a given launch date, you can query your chat interface in everyday words and get a sensible answer. There is less friction. On financial platforms where teams juggle currencies, entities and regulatory contexts, this comfort quickly translates into productivity gains.
But we must remain honest about the nature of the improvement. Natural language simplifies the request. It reduces the cost in terms of time, skills and context switching to extract a slice of data from your business. This is particularly useful in teams where the person asking the questions is not the one who is proficient in SQL. But at no point do chat interfaces help you know what to ask.
It’s still up to you to ask the question. It is still up to you to decide which cohorts are relevant, which periods are significant, which comparisons to study and which to ignore. You have to maintain the ability to look at a graph and estimate, “That doesn’t mean anything. »
If the result of your “AI analysis” is reduced to a table or a graph, you have not revolutionized the system. You’ve only eliminated a bottleneck in the question/answer loop. It’s useful. However, this improves your access to information and not your decision-making ability. All of these thoughts raise a question: how can we best leverage AI for analytics if the answer lies neither in the ease of creating tables nor the comfort of chat interfaces?
Agentic analysis or how to move from simple questions/answers to the notion of a decision-making loop
To answer this question, you need to transition from tools that answer questions to systems that do some of the thinking for you. This is what I call agentic analysis.
Think about the current learning journey of most organizations:
- Something is happening in the market or product.
- Someone realizes it, often late.
- Someone frames the question and draws data from it.
- Someone interprets this data and proposes an action.
- Someone validates this action.
- Finally, someone checks to see if it worked.
There are delays and performance issues at every step. The end result is that it can take weeks or even months between a change in the market and your team’s reaction.
Agentic analysis, whatever its form, reduces this delay. Instead of waiting for you to ask “How did the last campaign perform,” she:
- Lives inside your data and understands what matters to your business
- Knows what is normal and what has changed
- Find developments that deserve your attention
- Suggests specific actions to try
- Tracks results and improves over time
In other words, instead of acting as a simple search engine, it takes on the role of a colleague responsible for monitoring the numbers, flagging important items, and helping you move forward faster. This behavior corresponds to the move toward agentic analytics, a system that doesn’t just answer questions, but reasons, acts, and learns for you, with humans at the controls. This is fundamentally different from a discussion with your dashboard, even if interactions with these agents can take the form of a conversational interface.
The indicator to watch: decision-making time
An indicator captures the real issues: decision-making time. How long does it take for your organization to notice something important, decide on a course of action, and implement that decision?
Dashboards make important numbers visible to everyone. Conversational interfaces help more people get personalized views and spot anomalies without a business intelligence bottleneck. But the limit of these two tools remains determined by the speed of humans to examine the data, analyze it, agree on an answer and act.
Improving AI systems to understand the business context behind a change, explain what caused it, run or simulate experiments to test a response, and measure the effectiveness of that response will cause decision-making time to plummet. What used to take a month will now only take a week, or even a day.
However, speed of decision has pitfalls. If you just optimize for how quickly an agent can suggest and execute a change, you risk making mistakes more quickly. Good agents don’t just access data. They master the context and the different levels of governance.
In other words, good agents understand the business context behind the numbers, adapt their actions to each situation and know how to interpret and respect policies and best practices. Without this foundation, you are likely to make bad decisions faster and with more confidence.
The importance of the distinction between knowing and acting
One might be tempted to see AI-assisted analysis as a simple interface problem. All you need to do is add a conversational layer to your BI infrastructure and that’s it. The risk is spending your AI budget on making it easier to ask questions without considering how long it will take to implement the answers.
It would be more relevant to look at the problem in reverse, starting from the decision-making time. Ask yourself:
- At what point in the current process does learning stall?
- Which decisions are too slow, too complex, or too dependent on the knowledge of one person?
- What would it take for an AI system to go beyond just responding to “What happened?” ”, but helps to determine “What should we try to do now?” and “Did it work?” »
By taking this starting point, chat becomes a powerful tool within your overall strategy. The dashboards will continue to fulfill their function: giving everyone a common vision of the facts. Conversational interfaces will continue to make data less intimidating and easier to use. But the real benefit will come from the systems you build: the agents who work behind the scenes to reduce the time between the appearance of something in your data and the reaction of your teams.
AI-assisted analysis isn’t meant to replace dashboards or create the best chat interfaces. It is about closing the gap between knowledge and action and knowing that decisions are based on data and context. This is how you arm your organization to reliably learn at the speed of its data rather than the speed of its meetings. This is what well-executed agentic analysis looks like.
Airwallex helps you bring all your financial data together in one place to make decisions faster. Find out how by clicking here.
