Data Storytelling in the Age of AI: Breakthrough Conversation with IIA Expert Travis Grunewald

In this blog series, Jason Larson, head of content at IIA, interviews IIA Experts who help enterprise data, analytics, and AI leaders navigate their hardest challenges. With 150+ highly-vetted domain experts, IIA's Expert Network delivers tailored decision support, unbiased plan validation, and ongoing guidance to advance data and AI outcomes.

In this Breakthrough Conversation, Travis Grunewald, IIA Expert and senior executive specializing in AI, analytics, data science, and pricing strategy, explores why data storytelling matters more than ever as AI accelerates analysis. Grunewald unpacks the three questions every data story must answer — what happened, why it matters, and what's next — and explains why human judgment, context, and accountability remain irreplaceable. From executive buy-in to building storytelling as an organizational capability, this conversation offers practical guidance for D&A leaders navigating the next era of insight communication. 

First, how do you define or frame data storytelling in the age of AI?

The art of data storytelling is more important than ever now that AI can quickly crunch numbers and spit out analysis. As a result, it shifts the value from simply producing numbers to creating clarity, confidence, and action. That is where data storytelling comes in.  

To me, data storytelling is how you take facts and analysis and connect them to context, emotion, and a clear decision path. It helps people understand not only what happened, but why it matters and what they should do about it.

AI can quickly create dashboards and reports, summaries and analysis, and first-draft presentations, but people still decide what matters to the business, why the insight should be trusted, and what action should happen next.

The way I often say it is, AI makes technical production faster, but data storytelling makes it matter. It’s what turns analysis into business impact.

What are the “must-have” elements in a data story to influence decision-making?  

Every effective data story needs to answer three basic questions, although you do not always have to answer them in the same order.

The first is "What happened?" That is the evidence. What does the analysis show? What changed? What is the insight? It is the logical foundation of the story. Say churn is up 12 percent. That's the fact on the table.

The second is "Why does it matter?" That is where you connect the evidence to business impact, risk, opportunity, or the needs of the audience. It is the emotional stake of why they should care. Churn's up 12 percent, so what? It could mean we're losing customers faster than we're replacing them, and we need to lean harder into acquisition to fill the bucket or we will lose revenue. That's the difference between reporting a number and tying it to the business.

The third is "What's next?" That could be a recommendation, a decision, a test, or a specific path forward. This answers what we are going to do with it. For churn, that might mean recommending we shift spend toward acquisition next quarter, or testing a retention offer to bring them back. Now there's an actual decision on the table, not just a number.

The biggest mistake is stopping after the first question. A lot of analytical work tells people what happened and assumes the meaning is obvious. Usually, it is not. The audience is left to interpret the implications and decide what to do. A complete data story answers all three! If I had stopped after telling you churn was up 12 percent, you would have had a fact, but likely several different interpretations of why it mattered and what to do next.

For example, early in my career, I thought a campaign readout to my marketing partners stopped after I told them if the test worked or not. I eventually learned the reason they were repeating the campaign, even though the results were not successful, was because I was not clearly communicating the implications and recommendations going forward.  

When AI can generate a chart or narrative in seconds, what's the piece of storytelling judgment that still can't be automated?

AI can absolutely propose a story. It can identify a pattern, recommend a headline, or suggest what might matter. But it cannot own the judgment, context, trade-offs, or consequences behind that story.

A lot of what makes a story resonate comes from institutional knowledge. What decisions have already been made? What constraints are we working within? What are the strategic priorities? Which trade-offs is the organization willing to make? AI may have access to some emails, documents, or meeting notes, but it rarely sees the full picture. It sees the information it has been given, not everything happening around the decision.

It also struggles with organizational realities. It may not understand the politics in the room, the history between stakeholders, the level of change readiness, or the emotional response a recommendation may create.

Then there is accountability. AI can generate a recommendation, but it does not own the consequences. The leader or analyst presenting the recommendation still has to stand behind it.

I also find that AI is not always great at fine-tuning. It often gives me something that is mostly right, but not exactly what I want. It may also add extra info I don’t need, overstate something, or miss the tone. So, you still have to guide it.

So, the question is not whether AI can generate a story, it can. The harder question is whether it is the right story for this organization, this audience, and this decision. That is where human judgment and accountability still matter.  

Most D&A leaders say their teams struggle to get executive buy-in despite "good data." Is that a storytelling failure, a trust failure, or something else?

It can be any of those, and it is usually a combination.

Sometimes the analysis is good, but the story is not clear. The team may have the right answer, but they are presenting it in a way that is too technical, too detailed, or disconnected from the executive's priorities.

Sometimes it is a trust issue. The executive may not trust the data, the methodology, or the team presenting it. Even a well-designed story will struggle if the audience does not believe the underlying evidence.

One important point is that buy-in is often built before the presentation. A good story cannot fully compensate for weak stakeholder engagement throughout the work. If leaders first see the work when the final presentation is delivered, they may not understand the assumptions, feel ownership of the problem, or be ready to accept the recommendation. Never catch a business leader by surprise if you can help it. The best data teams engage stakeholders early, pressure-test the questions, and build confidence along the way.

Storytelling sets the tone, but trust comes from the full experience. It comes from the quality of the analysis, the relevance of the question, the credibility of the team, and how well stakeholders were brought into the process.

Generative AI can now draft the narrative around a dashboard. Where does that help, and where does it introduce risk (e.g. hallucinated causality, false confidence, oversimplification)?

I see AI as a great accelerator for data storytelling, but it should be the partner, not the pilot.

There are a lot of places where it can help throughout the process, not simply at the end to write the narrative. It can generate hypotheses, identify potential blind spots, explore alternative interpretations, recommend visualizations, draft headlines, create executive summaries, critique a presentation, or role-play a skeptical stakeholder.  

The potential risk with the automatic narratives is that the output can sound more credible than the underlying thinking deserves. I tend to group the risks into three areas. The first is accuracy risk where AI can invent facts, misread the data, or overstate the relationship. The second is reasoning risk where it may confuse correlation with causation or skip over important assumptions. The third is communication risk where it wraps a weak argument into polished, confident, executive-ready language, treating every insight like it's equally important, and sounding just as certain right up until you push back on it. That last one is the one I'd watch most closely. Generative AI can make anything sound right. That is not just a hallucination issue, it’s a false-confidence issue!

We have all probably had the experience of AI giving us a confident answer, then immediately changing its position when we challenge it. That should remind us that the quality and tone of the writing are not proof that the reasoning is sound. The human still has to ask: Is this accurate? Is the conclusion supported? Is anything overstated? Does this reflect the real business context? Would I be willing to stand behind this recommendation?

How should a D&A leader think differently about storytelling for senior leadership and the board versus storytelling for the frontline or business stakeholders?

The first two questions I ask before building any story are: “What is my objective” and “who is my audience?” Am I trying to inform, persuade, get a decision, or drive a specific action? Then, who am I speaking to? What is their role, data fluency, history with the topic, goals, and likely concerns?

Different audiences need different things and understanding that is the key to building a successful story. Tailor the story before you build it. Who you’re talking to changes everything.

An analyst or category team may need evidence, definitions, methodologies, and enough detail to evaluate or build on the work. A manager or functional leader usually needs the implications, feasibility, impact, and what will be required to execute. An executive or board member is often looking for the priority, recommendation, risk, trade-offs, and decision. This does not simply mean executives get less detail and analysts get more. It means each audience gets the detail that matters to the decision they are being asked to make.

Who the audience is can also change the order of the story. For an executive, I may lead with the recommendation. For an operational team, I may lead with the problem and then explain how the change affects them. For a technical audience, I may need to establish confidence in the evidence before moving to action.

I learned this firsthand when I was presenting a retention analysis to a customer support team. I walked them through the work step-by-step – methodology, data, findings – thinking I was building credibility. But by the time I got to the recommendation, we were out of time and no decision was made on what to do. Later, my leader told me to flip the structure: start with the headline, provide a recommendation, and then support it with the evidence. The next time I presented, we led with the decision and used the analysis to reinforce it, and the conversation immediately became more focused and productive.

A complete data story answers all three questions: 1) What happened? 2) Why it matters? 3) and What’s next? A compelling story answers them in the order the audience needs to hear them, and that order can make the difference between confusion and action.

As AI-generated insights scale, do you still need a human “translator” between the data team and the business, or does that role change shape?

I think about this one two different ways…

The first is whether you still need a person to make sense of what AI produces. There I don't have much doubt: YES. AI can hand you a lot of output, but somebody still has to sift through it, decide which version of the story is right, and understand the context around it. That doesn't go away just because the output got faster.

The second question is more interesting, and I'll be honest, I'm still working through it. Does the business need a dedicated analytics translator at all, or can AI close that gap directly? I can see it going either way. If a data scientist hands off technical work, I could increasingly just ask AI to translate what they did instead of routing it through a person, and that gets more plausible as the models improve. But I still think you need someone who can sit down with the data scientist, understand the trade-offs they made, and vouch for the translation, not just summarize the output, but stand behind it.

So, my answer is that the function isn’t going away, but I’m not certain the dedicated role survives in its current form. What I am confident about is the shape of what’s left – less translating technical language into plain English, more owning the judgment on what the story should be, and being willing to put your name on it.

What’s the most common mistake technically strong data teams make when they try to tell a story with their analysis?

The most common mistake is confusing analytical rigor with communication value. Technically strong teams often want to show all the “cool” things they did to reach the answer. They show the data preparation, methodology, models, diagnostics, charts, and every interesting finding they uncovered. They present the journey instead of the destination.

I am guilty of this myself. When I first moved into leading a pricing team, I would fall back on what I knew and bring 20 slides of analytical content into leadership discussions. It was reinforced because the CEO liked and engaged in the work, but we were not getting needed decisions made during our weekly meetings with the broader executive team. My boss (the CFO) finally told me that I need to focus on three or four slides to tell the story with everything else going into the appendix. That was an important lesson. The goal was not to prove how much work we had done. The goal was to help the leadership team make a decision in a limited amount of time.

Analysts often focus too much on the data and observation around it instead of turning it into an insight (what happened), sharing the implication (why it matters), and then providing recommended action (what’s next). That does not mean methodology is unimportant. It matters when it affects trust, uncertainty, or the decision. But it should support the story rather than become the story.

It's not only about too many slides. I've also gone the other way, crammed everything onto one slide thinking I was being disciplined, and watched someone's eyes glaze over trying to figure out what I was telling them. Both mistakes come from the same place: showing all the work instead of deciding what the work means.

A rule I often use is that if you cannot explain the core story in one slide, you may not have finished the thinking yet. You can have more supporting slides, but you should be able to clearly articulate the main insight, implication, and action.  

If the story is not driving toward a decision, it probably needs to go back to the drawing board.

How do you build storytelling as an institutional capability — not just a skill some analysts happen to have — across a data organization?

The first thing I would say is that you don’t have to be an extrovert to be a good storyteller. Storytelling is not about having a big personality or being the most charismatic person in the room. It is a structured skill that people can learn and practice. I'd put myself in that camp. I wasn't naturally wired for this, I built the muscle over time, and that's exactly why I believe it's teachable.

At the individual level, I teach teams to ask the three basic questions of “What happened?” “Why does it matter?” and “What’s next?” Another useful tool to help get you there is what I call the "so what?" ladder. What does the data show? Why is it happening? Why does it matter? What decision does it affect? What should we do next? This helps analysts move from observation to insight and from insight to action.

But an organization does not build the capability through a single training session. It must become part of how analytical work is scoped, reviewed, presented, and rewarded.

One of the most important things you can do is pair analysts with business partners early in the work. Analysts become better storytellers when they understand the decisions their partners make, the questions they ask, what keeps them up at night, and how they respond to information.

Two other mechanisms that can make a big difference include requiring every analysis to name its audience, objective, and decision before the work begins, and running peer story reviews before anything goes in front of executives. Both force the storytelling thinking to happen early, instead of getting bolted on the night before a presentation.

Finally, leaders need to reward business impact, not simply technical complexity or the volume of output. Teams will focus on what the organization values.

With AI lowering the cost of producing analysis, is the bottleneck now insight generation or insight communication? How does that shift what a D&A leader should be hiring and training for?

The bottleneck is shifting, but I do not think it is simply moving from insight generation to insight communication. AI will generate more findings, more charts, and more possible narratives than organizations can realistically absorb. The new bottleneck is deciding which insights deserve attention and converting them into action.

That can change what data, analytics, and AI leaders should hire and train for. Technical depth still matters because someone must understand whether the analysis is valid. But it is likely no longer sufficient on its own. Teams also need people who can frame ambiguous business problems, distinguish an interesting pattern from a meaningful insight, communicate uncertainty, understand stakeholder incentives, and make a recommendation.

I would hire for curiosity, judgment, business fluency, and communication alongside technical capability. I would also train analysts to ask a different set of questions: What decision will this affect? What evidence would change that decision? What are the trade-offs? What might we be missing? The highest-value analyst will not necessarily be the person who produces the most analysis. It will be the person who helps the organization make a better decision. 

Five years from now, what does "data storytelling" mean when most of the audience — human or AI agent — is consuming the story differently than today?

I think this is the same principle I described earlier about tailoring a story to the audience, just taken to its logical extreme. The idea that you shape delivery around who's receiving it isn't new, I'm doing that today when I lead with the recommendation for an executive and the methodology for a technical audience. What's new is the scale and the automation.

Five years from now, an AI system may be doing that tailoring in real time, for a hundred different consumers at once, human or AI agent, each pulling a different layer out of the same underlying insight. An executive gets the recommendation and the risk. An operational team gets the instructions. An AI agent downstream gets the structured evidence, the assumptions, and the decision rules it needs to act on its own.

I believe what won't change is the core job of a data story: establishing what happened, why it matters, and what should happen next. If anything, that job gets more important, not less, because AI will make it cheap to produce a dozen plausible but conflicting narratives from the same data. The differentiator won't be who can generate the most polished story. It'll be who can produce one that's grounded, trusted, and tied to an actual decision, no matter which layer someone is reading.