As senior leaders invest in transformational AI initiatives, they need to see clearly. “You Are Not Hallucinating” spotlights BCG’s AI experts, whose insights for the C-suite range from actionable next steps to thought-provoking ideas.
I recently met with a CEO of a manufacturing company who was proud of his robust investment in data storage and eager to become an industry leader in AI. As we discussed what an AI transformation would look like, our conversation turned to predictive maintenance, which is a practical starting point for many manufacturers.
To the company’s credit, it had diligently tracked sensor data on equipment failures for five years. Unfortunately, it did so only in minute increments. In reality, though, multiple errors often occur within the same 60-second window. Without second-by-second resolution, the data was too coarse to pinpoint the root causes of failure.
If the company had captured five years of high-quality error data, it could have trained AI agents to anticipate failures and cut unplanned downtime. Instead, the company had sunk significant resources into a trove of unusable data.
The point of this story isn’t to call out one organization’s mistake; it’s to draw attention to an incredibly common situation.
All too often, CEOs who are sitting on large data sets accept the recommendations of their CIOs or heads of engineering even when their guidance doesn’t reflect the true needs of the business. Lacking deep data or technology expertise, CEOs frequently hesitate to push back, especially when confronted with persuasive pitches, heavy on technical jargon, to invest in costly new software and cloud technologies.
In contrast, some CEOs cite poor data as an excuse to delay an AI transformation. This, too, is a mistake. There is no such thing as perfect data, and many of the most successful AI programs thrive despite incomplete or imperfect data sets.
Here’s what CEOs must do to ensure that their data strategy is AI-ready:
- Start with the goals of the business. Every data conversation should be rooted in the organization’s strategic objectives. When companies anchor their efforts in real operational challenges or growth opportunities, they start asking the right questions: What strategic decisions are we trying to support? What workflows are we trying to automate? What data insights will create value for the business?
- Be ruthless about integration decisions. Calibrate the organization’s needs to the appropriate level of investment. Real-time integration is expensive, so don’t default to it. Ask yourself, Do we need to connect live equipment directly into real-time systems, or would batch uploads and after-the-fact analyses be sufficient? Each integration choice carries tradeoffs in cost, speed, accuracy, and cybersecurity risk.
- Understand the cost of getting it wrong. Be vigilant about what data is used by communicating with the people closest to the inputs. Ask your teams, Are we getting this right? A company built a GenAI field agent for technicians, but the agent started returning incorrect instructions because it had been trained on 30-year-old manufacturer manuals that no one actually used. So, while the manuals reflected the official, documented order of operations, they were obsolete in practice. Not only is investing in bad data costly in terms of resourcing, it may undermine employees’ confidence in the technology. Building in a robust data quality assessment up front can deliver exponential returns down the road.
The beauty of starting with business outcomes is that it brings teams closer to the inner workings of their organization. In fact, data quality can have a meaningful impact on worker focus and company culture.
If leaders incentivize people to care about data quality, they will inspire enhanced trust in automated and agentic systems. This will decrease the opportunity for unforced human errors and free up space for workers to focus on challenges that require human creativity.
I’ve also watched executives fall in love with their business all over again by digging into their data needs, seeing a process with fresh eyes, and asking, Why are we still doing it this way? The executives often find that work that has been done manually for decades can now be performed—with a higher degree of accuracy, at an unprecedented pace—by turning good quality data into airtight AI solutions.
I firmly believe that in the near future, more CEOs will have a foundational background in data. That doesn’t mean that CEOs have to become a technical expert, but it does mean they should become a strategic data leader. To start, CEOs should:
- Create a strategic data and AI team that reports directly to them—not to just IT.
- Treat data as a product. Have the data and AI team report to them like a product team does and focus on usability, accessibility, and value.
- Define upfront what types of data can create value for the business. Otherwise, the company risks paying for data that looks polished but doesn’t deliver a return on investment.
- Ask the CIO to walk through the top five places where bad data impedes progress.
- Own the why—and no longer abdicate responsibility to technology teams.
Data is a strategic lever for getting ahead in a world that is rapidly transforming with AI. Data must be proactively owned by the organization’s top officer.
