The AI hype settled; the data problems didn’t.
Two years into production AI, the projects that stall rarely stall because of the model. They stall because of the data layer underneath. Accounts with no industry, leads that never matched to anything, hierarchies built around a legal entity instead of how your team actually sells. An agent reading that data doesn’t fix it, it repeats it faster and is more confidently wrong than ever.
At Dreamforce, Ernesto Valdes (CTO, Traction Complete) and Chris Pickering (Sales Operations Lead, Snap) walked through the 10 data quality requirements needed before any agent touches your data. You’ll take away:
- Data foundations. What has to be clean, connected, and matched before an agent reads any of it.
- Enrichment. Where AI can outperform data providers, and where it doesn’t.
- Token costs. How to plan and manage what an AI job costs before you run it.
- Governance. Who owns an agent once it’s built, and who should be gated out.
- Success metrics. How to define what working means before the pilot ship.
