Editor's note:Siqi Chen is the CEO of Runway, a financial software company headquartered in San Francisco.RunwayThe views expressed in this article are solely those of the author.

Everyone craves the "magic" that AI brings, yet few are willing to invest in the data groundwork that makes it possible.

This is exactly why many AI projects fail. Algorithms are rarely the problem; data preparation is where things commonly break down.

Teams often face incomplete data, conflicting definitions, forgotten fields, and competing "sources of truth." They rely on shadow spreadsheets that no one acknowledges and systems that cannot sync. These are not edge cases—they are the norm.

The key to solving these problems lies in establishing a unified source of truth that all departments within the organization can rely on. This requires the unglamorous foundational work that most people keep putting off. You cannot automate your way around structural weaknesses like data fragmentation. AI amplifies whatever foundation it rests on: solid data pays off over time, while weak data collapses at scale.

But fixing the foundation is not just an IT matter—it requires senior leadership involvement.

Data governance used to belong entirely to IT, but this responsibility is increasingly falling to the finance department. CFOs have visibility into every major data touchpoint, giving them a unique advantage in reducing data silos and establishing a shared source of truth.

Here are six steps CFOs can take to ensure AI has a solid data foundation:

1. Map data usage

Start by assessing the organization's existing data practices, including where data is stored, how it is maintained, how it is structured, and how it is updated. Along the way, identify conflicting data sources. For example, sales, finance, and marketing systems may define "active customer" differently, leading to inconsistent results. Investigate the root causes of these discrepancies so that AI can rely on accurate, aligned inputs.

2. Appoint formal data stewards

After clarifying how data is used and who touches it, assign clear, formal ownership. Each data domain should have a clearly responsible steward who ensures data is accurate, reliable, and follows company-wide standards. Responsibility should not be delegated to committees or "teams"—it must be assigned to specific individuals. Clear ownership creates accountability, while committees often lead to diffusion of responsibility.

3. Define what data truly means

Accurate data alone is not enough—AI needs clear definitions and context. What counts as a customer? What does "active" mean? AI cannot fix problems that humans have never clearly defined.

4. Prioritize integrating critical data

You don't need a perfect, universal source of truth immediately—you just need to reduce contradictions. Every isolated spreadsheet or workaround weakens the data signal. Start with data that directly supports key decisions, then expand gradually. Unfortunately, this step can become political. Every VP and analyst may have their own dashboards and definitions, and standardization may be seen as giving up territory. But in the end, results matter most: centralized data governance—though painful in the short term—resolves conflicts caused by departmental silos.

5. Promote cross-departmental data flow

Data fragmentation across finance, sales, operations, marketing, and HR limits the insights AI can provide. Only by connecting these sources can organizations unlock reliable, actionable intelligence. Establish mechanisms for data sharing, automated integration, and cross-departmental review to reduce bottlenecks and unify perspectives.

6. Keep the process running

This cannot be a quarterly project. Establish processes so that data hygiene, shared definitions, and cross-departmental review become part of the team's daily operations. With every new integration and every product launch, ask: "Does this add clarity, or does it introduce more ambiguity?" By continuously reinforcing collaboration and consistency, you can prevent silos from re-emerging and maintain a foundation that AI can reliably build on.