Designing an AI Roadmap When Your Data Infrastructure Isn't Ready

CDO (Chief Data Officer)

Everyone wants an AI roadmap. The board wants it. The CEO wants it. But the truth is, our data is a mess. Siloed systems, inconsistent formats, no single source of truth for half our core entities. I feel like I'm being asked to build a mansion on quicksand. How do I create an honest roadmap that doesn't either overpromise or sound like I'm making excuses?

AI Strategy Advisor — Data Strategy

You're living the most common paradox in enterprise AI: the people demanding the roadmap don't want to hear that the foundation isn't ready. But here's what experienced leaders know — you don't need perfect data to start. You need honest data about your data.

Every organization that waited for perfect data readiness before starting their AI journey is still waiting. The ones that succeeded started with what they had and improved the data as a byproduct of AI implementation, not a prerequisite.

1. Build a dual-track roadmap. Track one: AI use cases you can execute with today's data, however imperfect. Track two: data infrastructure improvements that unlock progressively more ambitious use cases. Present them together so leadership sees that AI investment and data investment are the same investment.
2. Identify your 'good enough' data domains. You said half your core entities lack a single source of truth. That means the other half do. Start there. Find the two or three data domains that are cleanest and map AI use cases to those specifically.
3. Reframe data cleanup as AI-funded. The hardest part of your job has been getting budget for data infrastructure because it's invisible work. AI gives you the narrative to fund it. 'We need $2M for data integration' gets rejected. 'We need $2M for the data foundation that enables $10M in AI-driven efficiency' gets approved.

What are the two or three data domains where you have the most confidence in quality and accessibility right now?

CDO (Chief Data Officer)

Customer transaction data is solid — we've invested there for years. Employee data is decent. Everything else — product data, supply chain, marketing attribution — is scattered across dozens of systems with no common identifiers. How do I sequence the cleanup without it taking three years before we show any AI results?

AI Strategy Advisor — Data Strategy

Perfect. You just named your first two AI domains: customer intelligence and workforce optimization. Ship AI wins in those areas within 90 days while using the remaining domains as your infrastructure improvement roadmap.

1. Launch two quick-win AI projects using your clean data: predictive customer churn modeling on your transaction data, and AI-assisted workforce planning on your employee data. These can be built in weeks, not months, because the data foundation already exists.
2. For the messy domains, don't try to clean everything. Pick one — the one with the highest-value AI use case behind it — and create a 'minimum viable data product.' A focused, curated subset of supply chain data that's clean enough for one specific AI application. Prove the value, then expand.
3. Set a rule: every new data initiative must include an integration standard. Going forward, no new system gets purchased or built without a data contract that specifies format, identifiers, and API access. You can't fix the past overnight, but you can stop making the problem worse.

The roadmap that works isn't 'clean data, then AI.' It's 'AI where data is ready, data improvement where AI demands it.' They fuel each other.

Want to sketch out the 90-day plan for those two quick wins? That gives you visible progress to show the board while the harder data work begins underneath.