Webinar Recap: Agentic AI for Enterprise Decisions: Demystifying the Build vs Buy Dilemma

Summary

In our Future.Now webinar, “Agentic AI for Enterprise Decisions: Demystifying the Build vs Buy Dilemma,” we took on a question that gets complicated fast. Should you build your own decision system from the tools you already have, or buy a platform made for the job? We set out to answer it honestly, and to keep it specific to enterprise decisions and AI. To set a clear baseline, we started with how Gartner defines a decision intelligence platform, then looked at the real ways teams try to assemble one on their own.

We walked through five common configurations. They range from AI development platforms to frontier models connected to data lakes or transaction systems, along with workflow tools and the AI now built into modern ERP systems. Each does real work and earns its place in the stack, and the point is not to diminish any of them. But each stops short of owning the decision itself, and that gap is what the discussion centered on. We closed by examining where enterprise AI budgets actually go, why so little of that spending reaches operational decisions today, and what a platform built around the decision does differently.

Key Takeaways

• Six capabilities have to work together. Gartner says a decision intelligence platform must support, augment, and automate decisions across people and machines. To qualify, it has to handle decision modeling, collaboration, execution, monitoring, and governance, all at the same time. These are not separate features on a checklist. They work as one system in real time. Think of a self-driving car: it senses, decides, and acts as a single loop, not as a series of steps.
• A prediction is not a decision. Platforms like Vertex, SageMaker, and Databricks are great at building and deploying models. What they hand back is a score or a prediction through an API. To turn that into an action, you still need a decision model, a policy check, governed execution, an audit trail, and a feedback loop. All of that sits outside the platform, and your own team has to build it and keep it running.
• A frontier model on your data gives you answers, not governed decisions. Connect a model like GPT, Claude, or Gemini to a data lake and you get a strong chat interface. It can spot a supply disruption, explain the cause, and suggest a next step. But the output is still just an answer to a question. There is no record of what was decided or why, and no built-in way to apply policies, track outcomes, or learn. In short, that’s just AI on top of BI.
• A frontier model with connectors is a strong prototype, not a finished system. Add connectors and the model can pull data and take action, so it can feel like an agent. But the model is only one part. The harness around it is the real product. A coding assistant works because of everything wrapped around the model, and an enterprise decision needs the same: memory between steps, governed execution, outcome tracking, and cost control across thousands of runs. Building and maintaining that harness is a big, ongoing job.
• Workflow tools run decisions but are not built around them. These tools are great at running a process once the logic is set, and they run it the same way every time. But execution is only one part of decision intelligence. Workflow tools turn every problem into a workflow, and a decision is a different thing altogether. They also struggle with brand-new events, like a stuck ship or a sudden tariff. Those moments need real intelligence to read the situation and weigh the options, not a script set up in advance.
• Application agents see only one slice of the decision. Agents inside enterprise apps reason within a single object model. A CRM agent thinks in accounts and deals, while a service agent thinks in tickets and queues. Each is strong inside its own app. But a real disruption, like a weather event that hits supply, involves many decisions at once: a procurement call, a logistics call, a revenue call, and a customer call. No single app sees the whole picture, because the decision is not an object in any of them.
• When the decision is the unit of work, you can govern it, audit it, and learn from it. A purpose-built platform applies governance at one point, so the model reasons but the platform decides what is allowed. Every decision carries a full trail from trigger to data to reasoning to action to outcome, and you can rebuild it on demand months later. Captured this way, each decision becomes a first-class object you can study in bulk. That improves the next thousand decisions and turns know-how into a lasting edge.

Speakers

Ram Krishnan, SVP, Platform Product Marketing, Aera Technology Ram leads the GTM strategy for Aera Decision Cloud with over two decades of experience in enterprise, software, product strategy and customer success. Ram brings a deep understanding of how to align solutions with customer values. Under his leadership, Aera has earned a strong reputation for delivering measurable business outcomes and customer satisfaction.

Lalitha Sundaramurthy, SVP and Head of Product, Aera Technology Lalitha leads product strategy and innovation across Aera’s platform. Lalitha brings nearly two decades of experience building enterprise data, analytics, and AI solutions, including earlier product work at Informatica.

Mustafa Kabul, SVP of Data Science, Machine Learning, and AI, Aera Technology Mustafa works at the intersection of machine learning, optimization, and generative AI, where he develops the intelligence that powers Aera, the decision intelligence agent. He focuses on advancing the models, heuristics, and learning systems that enable Aera to deliver fast, accurate, and goal-aligned decisions at enterprise scale.

Full Recording

Access the full webinar recording here.

Q&A

At the end of the presentation, we held a short session dedicated to answering attendees’ questions. Below are several of the key questions and answers.

Q: Where do you start, and how fast can you show value?

A: We live in a world that expects value almost instantly. If a project is slow to deliver, it will not get off the ground. The good news is you do not have to wait years. With customers who are ready to go, the first value tends to show up in weeks. The exact pace depends on practical factors like security clearance and what data is available. In the best cases, organizations have seen full payback in as little as 12 weeks, and there are case studies that back it up.

Q: What does the total cost of ownership (TCO) look like?

A: Total cost of ownership drops over time, and the reason comes down to the foundation. You start by building a decision data model that captures how your business works, and that foundation gets reused with every new use case. Say you optimize inventory today and move to order fulfillment tomorrow. When that work needs to look at inventory, you do not recreate the underlying model; you build on what you already have. So the cost of adding each new decision keeps falling while the value keeps rising. Our head of customer success calls it a domino effect.

Q: Can I use our own LLM for decision intelligence?

A: Yes. The Aera platform is open by design, and the agentic AI engine works with any model. That includes open-source models, frontier models, and your own large language models, whether you fine-tune them or host them yourself. Some customers already do this. The goal is to give you the best harness for your decisions while leaving the choice of model in your hands, so you can pick whichever one fits your needs.