Most conversations about AI focus on what models and agentic systems can do. Kelly Murray is more interested in what they need to know. As a Senior Staff Outbound Product Manager at Pyramid from ServiceNow (formerly Pyramid Analytics), Kelly works at the intersection of business intelligence, machine learning, and enterprise AI – helping organizations move beyond the hype over agentic systems and build the data foundations that actually make them work. With a career spanning analytics, strategy, and AI product management, she has developed a practitioner's perspective on what separates AI deployments that deliver measurable outcomes from those that simply add noise to an already complex environment.
In this interview, Kelly explains why the biggest bottleneck in enterprise AI is almost always the data, how machine learning has shifted business intelligence from reactive rule-following to proactive decision support, and what it really takes for an AI agent to understand a business rather than just process information about it. She also shares her thinking on trust, governance, and why the goal was never for AI to replace human judgment – but to earn enough trust to be genuinely useful.
Interview Highlights
Q. What common mistakes do you see organizations making as they begin deploying agentic AI initiatives?
Sometimes businesses understandably get a little too excited and end up over-focused on the agents themselves and what they can automate, instead of asking some important questions, like what business decisions, workflows or customer experience do we need to improve?
Some organizations also give the agents too much autonomy too quickly. I recommend starting with agents giving assistance and recommendations and then move to automation once we’ve had a chance to establish trust and also identify some gaps.
AI shouldn’t be treated like a switch that you turn on or off. Successful organizations are treating AI like a journey, one that can change and evolve over time. They focus on their business objectives, establish guardrails, keep humans in the loop where appropriate, and gradually increase autonomy as trust, governance, and confidence mature over time.
Q. Enterprises have been connecting business intelligence (BI) output to automated workflows since long before LLMs went mainstream. How has machine learning changed the way this all works?
Machine learning is helping us to scale and be more proactive.
Traditionally, BI and workflow automation were completely rule-based. If sales drop by 10% then do this. If returns go above a threshold, then do that. It’s difficult to manage these rules for all products and segments.
Machine learning allows us to identify these outliers or data points that are outside of expected bounds automatically. But also allows us to identify issues before they get to that point. We can now see trends and the likelihood of a threshold being broken.
This enables us to be much more proactive. So instead of trying to use workflows that fix something that has gone wrong, we can use these systems for prevention instead.
Q. What are some of the biggest differences between agentic AI that has a data analytics component and agentic AI that doesn’t?
An agent without a data analytics component is probably reasoning from text, documents, policies, and instructions alone. It may know how a business process works and what actions are typically recommended, but it doesn't know what is actually happening in the business right now.
In short, it’s another layer of context that makes AI with data analytics more powerful and relevant.
For example, an AI agent might know when inventory shortages typically occur and what actions should be taken to address them. But an analytics-enabled agent can see that inventory for a specific product is declining faster than forecast, supplier lead times have increased by 15%, and demand is expected to spike over the next month. The recommendation is no longer based on generic best practices – it's based on what is actually happening in the business and understands the impact of the recommendation.
Q. What role does business context have in building out agentic logic models? What types of data are best for ensuring the AI system has a maximum “understanding” of business context?
That’s a tough one. Business context exists in so many layers, and it's not just about accessing the data. It's in the KPIs and metrics, the business definitions, organizational structures, policies, how the data connects with each other, and some difficult to capture items like strategic objectives and even organizational culture.
But the general rule is that the most important context is semantic rather than facts. That’s why semantic layers, meta data, synonyms and governed models are becoming increasingly important foundations for agentic systems.
Will AI ever be able to fully capture the context of a business? I don't know. I'd like to think there's always going to be a human element that is difficult to completely encode into a model. Every organization has unwritten rules, knowledge, political realities, cultural norms, and strategic considerations that aren't neatly stored in a database or documented in a policy.
Two people looking at exactly the same data may reach different conclusions, because they understand nuances that come from experience, relationships, and business judgement.
That's why I don't think the goal is necessarily for AI to perfectly replicate human understanding. The goal is to provide enough business context that it can make informed recommendations. There will likely always be situations where human judgement remains an important part of the process.
Q. When you talk to business leaders about what AI can and can't do with their data, what do they find most surprising?
I think it's often a surprise that bottlenecks don’t result from what the AI is able to do but rather how ready the data is to support the AI.
AI isn’t a miracle cure. It's not going to fix your data – it simply uses it. So upgrading to the latest and greatest version of AI models isn’t necessarily going to produce better results. It might just produce bad results faster. It’s the quality of your data, the collection and arrangement of business context and governance, that will make the biggest impact here.
Model capability is not an issue anymore. But unfortunately, there is no shortcut if the data isn’t right. So before enabling AI, we have to honestly evaluate the AI readiness of our organizations. The successful ones are those who have put in the groundwork. They’ve invested in trusted data and created consistent definitions across their business. With these foundations, AI becomes much more effective. Without it, you are just adding one more variable into the chaos, and it can just as easily amplify confusion and poor decisioning making.
Q. Enterprises often talk about trust in AI. How does integrating business context contribute to trust and governance?
I think trust in AI is pretty simple – it’s the same as trust in a person. If a person told us what to do, didn’t explain why, gave no context, we would naturally be sceptical. The same applies to AI.
If we want people to be on board with a decision, we want to be transparent, be on the same page, give them the journey to the decision and not just the final recommendation.
Imagine an analyst presenting to the board and recommending that marketing spend should be reduced. The immediate reaction from senior leaders would be questions like, Why? What data supports this? What analysis have you considered? Which channels are underperforming? What impact do we expect on revenue? What alternatives were considered?
AI is no different. At the most fundamental level it must be transparent. It must show the data it accessed and the analysis applied. Even better, connect the recommendation back to business context and policies, so we understand not just what is recommended but why it makes sense for our business.
Ultimately, however, there is an element of time. Consistently reliable performance will create trust over time, just like with people. This is where governance becomes critical. Governance provides the guardrails, accountability, and oversight that allow organizations to safely build that trust. It ensures decisions can be audited, policies are followed, and risks are managed.
Over time, strong governance combined with proven outcomes is what turns AI from an exciting new technology into a trusted business capability.
Connect with Kelly Murray
If you find this interview valuable, you can follow Kelly’s work and connect with her directly.
LinkedIn: Click Here
Website: Click Here
About Business Outstanders
Business Outstanders is a business magazine featuring interviews, growth strategies, and leadership insights for entrepreneurs and small business owners. If you know a business leader with a story worth telling, nominate them for a feature interview.
Business Outstanders brings you sharp insights on tech, business, entrepreneurship, law, crypto, and more. We uncover what’s next. Stay updated, sign up for our newsletter and be part of the future!
