Why AI Readiness Starts With Your Data Foundations, Not Your Copilot Strategy — Murali Kallem, Box

Murali Kallem leads Data and AI at Box, where his remit spans internal data strategy, data architecture, and the semantic layer that underpins the company’s AI agents, alongside a forward-deployed engineering model that partners directly with business teams to automate their processes. His path to this role runs through GE, Toyota, and Salesforce, and it was in these earlier roles, including formative data management work at Motorola and IRI, that he first saw how the quality and structure of an organisation’s data directly shapes the decisions built on top of it. That grounding now informs how he thinks about AI: not as a layer to be added on top of existing systems, but as a capability that depends entirely on the strength of the data foundations beneath it.

How would you describe your responsibilities leading Data and AI at Box?

Murali Kallem outlines the scope of his role, from data strategy and architecture to the semantic layer that powers Box’s AI agents.

“My responsibilities within Box are to lead Data and AI for internal business user consumption. Along with that, data and AI come with the responsibility of creating a data strategy, a data ecosystem, and a data architecture, making sure we are pulling in the right data, providing the right context, and building a semantic layer that actually provides the right context to the AI agents we are building and soon to build.

There is quite a bit of that activity happening. My team also works directly with the business in a forward-deployed engineering model, working with them to automate their business processes by leveraging AI and automation techniques. So that’s essentially my role: ultimately providing business value to the business users.”

Was there a particular role or moment that changed how you think about data and its value?

Reflecting on his path through Motorola, IRI, GE, Toyota, and Salesforce, Murali Kallem points to how early he encountered the cost of poor data quality.

“I think that happened very early on in my career, understanding how not having the right amount or the right quality of data impacted the business. This was way back when I was working for Motorola, where we worked with the business to help them make decisions, and then also at IRI, where we built a master data management technology. That was when I truly understood the data management challenges and how they were impacting the company. The sooner you realise this, the better you are able to facilitate the business with the right data ecosystem.

At IRI, in the CPG industry, we built a master data management solution for our internal business users. It created a single source of truth for key master dimensions, given that we dealt with a lot of consumer items and the stores that sold them. We wanted to make sure we had the right context for what each store meant: that stores weren’t duplicated, that they had the right information, and that we did the necessary address standardisation. We also enriched the data with where those stores resided, the demographic information around them, and what kinds of consumer products they were selling.”

What has changed and what has stayed the same in the role of data as AI has advanced?

Murali Kallem argues that data remains the foundation of decision-making, but that AI raises the stakes on getting that foundation right, while also offering new ways to build it.

“The fundamental requirement that data is the oil for all business decision-making still stays true, as does the fact that you need strong foundations for AI to make the right decisions. It’s even more important now, in the hallucinated world of AI, where it can take non-deterministic paths. It’s important that we serve up the right amount of data with the right context and the right metadata.

In the pre-AI days, it was important to have your data management foundations in order so people knew where to find your tables and your business metrics. Now, in the agentic world, it’s even more important, because you don’t have a human inferring things; you have AI agents inferring things. On the other hand, you can also leverage AI to help you build the foundations. Building data definitions, metric definitions, and classifying data used to be a very manual and tedious effort with no automated, repeatable, trustworthy way to do it. Now, with AI, you can make good investments in automating that work.”

What do leaders most commonly misunderstand about becoming genuinely AI-ready?

Murali Kallem contrasts the copilot-and-token conversations he hears at industry events with the outcome-driven thinking that mature organisations apply instead.

“I go to a lot of networking events and conferences, and I see the evolution of different companies in terms of being AI-ready. A lot of people talk about copilots, about how they’re measuring the productivity of their engineers, how many tokens they’re consuming, and how much code they’re generating through AI. That was the conversation until a few months ago. Now they’re talking about having a token problem- too many tokens consumed- so they’re working on budgeting tokens.

A mature company thinks differently. What needs to happen is to take a strong step back and look at what business value you need to provide. Look at your business processes and think about how you can apply AI to them, how you can automate things, and, while doing so, keep a peripheral view of how to reinvent that process entirely using AI. Is there a way to come up with a new process that achieves what the business wants, but in a completely different, more automated manner? You’re still figuring out where to keep a human in the loop, but the aim is autonomous agents running your marketing campaigns or your CPQ process.”

How should organisations approach rethinking a process rather than simply layering AI onto it?

Murali Kallem sets out a phased approach: build individual fluency with AI first, then let that experience inform a longer-term redesign of the process.

“I think you have to take baby steps. Don’t just jump at reengineering the process. You need to first let people understand the strengths of AI and how to leverage it in their day-to-day activities. If you’re a go-to-market engineer or a business user, what can you do? Can you do your research prior to a customer meeting using AI? During the meeting, provided it’s recorded, can you leverage AI to be more productive, automating some of the documentation, updating your CRM system, taking action items, or opening internal tickets?

That helps people take baby steps, but along the way, you need a thought leader, or leaders, to look at what the process should mean two years down the line, with a completely redesigned AI process.”

What is the big question every senior leader should be asking themselves about AI right now?

Murali Kallem closes on the risk-aware mindset he believes leaders need as AI agents gain more autonomy inside organisations.

“How are you leveraging AI? What is your approach? How are you planning to implement it within your organisation, and what are the blind spots as you’re scaling AI? If you’re in banking and you’ve built an agent that automatically approves a loan, is your blind spot that you don’t have a sufficient explainability ecosystem, or that the model is skewing against applicants of a certain race, exposing you to a lawsuit you haven’t thought through?

A risk-aware purview is what I would expect all leaders to have. Have an awareness of what AI could do that could put your organisation at risk, whether that’s building a secure solution or exposing more internal data than you should. But also enable users to adopt and leverage AI as much as possible in a self-serve manner. True innovation and true productivity only happen if people are empowered to take advantage of the technology, provided you’ve drawn the lines: what the limitations are and what they should not cross. Balance opening up the doors for mass adoption with making sure you know how things are actually being used.”

Join the Conversation: The Ortus Club’s Executive Network

As Murali Kallem has shown, becoming genuinely AI-ready is less about adopting the latest tool and more about the discipline of building trusted data foundations before scaling autonomous agents. Leaders at this level increasingly rely on structured peer dialogue to pressure-test their approach, compare notes on governance and process redesign, and avoid the blind spots that only become visible once AI is operating at scale.

His emphasis on risk-aware adoption reflects a broader reality: as data and AI leaders move from experimentation to enterprise-wide rollout, they cannot rely on internal thinking alone. The most effective leaders actively seek out peer perspectives to stress-test their governance models and reinvention strategies before committing to them at scale.

At The Ortus Club, we host curated executive roundtables that bring together senior data and AI leaders facing these exact challenges. Step away from the copilot conversation and engage in the kind of open, high-value dialogue that helps you build AI foundations you can actually trust.

FAQs Section

Q: What is a semantic layer, and why does it matter for enterprise AI?

A: A semantic layer provides AI agents with the right business context around an organisation’s data, helping ensure the data they act on is accurately understood and correctly applied.

Q: Why is AI described as producing ‘non-deterministic’ outcomes?

A: Unlike traditional deterministic systems that repeat the same steps identically, AI can take different paths to an outcome. Leaders building processes that require reliable, repeatable execution need to design for that limitation explicitly.

Q: What does it mean to ‘reinvent’ a business process with AI rather than layer AI on top of it?

A: Rather than simply automating an existing workflow, reinvention means stepping back to redesign the process itself, using AI to achieve the same business outcome in a fundamentally more automated way.

Q: Why does domain expertise remain valuable as AI takes over more routine and analytical work?

A: As AI lowers the barrier to building technical solutions, understanding the specific business function you are serving becomes what allows you to ask the right questions of AI and apply it effectively.

Q: What is a ‘risk-aware purview’ when scaling AI agents across an organisation?

A: It means maintaining awareness of what AI could do that would put the organisation at risk, such as insufficient explainability, biased decision-making, or exposing more internal data than intended, while still enabling broad, self-serve adoption within clear guardrails.