Bin Mu is chief data officer and vice president of enterprise data and analytics at Adobe. In a video interview published by CDO Magazine on 7 August 2025, the first part of a three-part series, he spoke with Rohit Choudhary, chief executive of Acceldata. Their conversation covered why Adobe rewrote the charter for its enterprise data team, how agents are changing the way business rules are managed, and how the team keeps financial figures accurate while working faster.
A charter rewritten around AI
Mu explained that his team’s previous charter centred on data. It covered data strategy and roadmap, a unified data platform, analytics, governance and the delivery of AI-enabled products. Then the brief changed.
“The reason for the refresh is AI, which fundamentally changes everything.”
The new charter is AI-first, with an AI-driven data strategy and a roadmap built around it. He named three additions: AI-driven intelligence that people can act on, agentic AI running on an internal Agent OS, and training to build AI fluency. He presented the refresh as his team, the wider organisation and the company getting ready for the AI era. Agents running on Agent OS, he said, can automate many functions and reduce the need for people to step in.
He was candid that he was still working out the full implications. The early benefits he listed were simpler workflows, shorter cycle times and fewer conflicts, because the AI keeps optimising for business outcomes.
Turning business rules into building blocks
The impact, he said, reaches both internal teams and outside business partners, and he used annual recurring revenue (ARR) to show how. Calculating ARR involves hundreds of business rules. Under the old process, any change to pricing policy or to the commercial model was slow and error-prone to put into effect.
His team broke those rules into foundational building blocks, which he likened to Lego, so they could be reconfigured quickly whenever a rule changed. The first step was drag-and-drop configuration, which cut the amount of coding needed. The second assigned an AI agent to represent each distinct block, which made the whole system more adaptable.
He said the approach significantly lifts productivity. The plan is to give business partners a flexible interface so they can adjust rules themselves, with changes flowing automatically through the network of agents. Automating that groundwork frees his team to concentrate on strategic work.
Accuracy first for financial metrics
Choudhary raised the tension between intelligence and accuracy, and Mu said his peers share the concern. His priority is to keep the semantic layer, the shared business definitions that sit on top of the data, clean enough to act as the gold source of truth.
He illustrated the old problem with a request from the chief financial officer for April’s ARR for Adobe Firefly in Germany. A question like that could take days to answer and might still produce wrong figures, because rigid business intelligence dashboards had been built across many dimensions.
The current approach uses a modern data stack and the medallion framework, which organises data into progressively refined layers. All the relevant business rules and metrics sit in the top, gold layer, and the attributes for each entity are consolidated into a single record. He said this improves speed, accuracy and governance together. Structuring business rules as taxonomies, the same building blocks used for ARR, carries data from its raw form to a coherent semantic model, and AI works well with that structure.
His most striking example was personal. Using the AI coding tool Cursor, he built a semantic financial table in two hours. He set that against the two months a data engineer would traditionally have needed for the same job.
What it means for leaders
- For data leaders, the lesson is that a team charter written before generative AI probably needs rewriting. Mu’s three additions, covering actionable intelligence, agents and fluency, make a useful checklist.
- Breaking business rules into small, named blocks pays off twice, because people can change them faster and agents can work with them.
- Fast answers on financial metrics depend on a clean semantic layer. Our reading is that teams should fix the gold layer before putting an AI interface in front of the finance function.
- His two-hours-against-two-months example is one person’s experience, yet it suggests data leaders should try the tools themselves before setting targets for their teams.
- Letting business partners change rules directly shifts accountability, so governance has to cover who may change a rule and how each change is checked.
