Cisco Is Using AI Agents to Run Forecasting at Scale—Here’s Why That Matters
Cisco Is Using AI Agents to Run Forecasting at Scale—Here’s Why That Matters
“[The causaLens agent] acts as a PhD-level Economist to assist teams to interpret what is going on”
When Cisco plans for the future, it does so with more than just models, it does it with machines that think like analysts.
At the cAI Conference, Puneet Gupta, Director of Data Science for Global Planning at Cisco, revealed how his team is turning to AI Agents to transform demand forecasting across 10,000+ products, 10 business units, and a multi-billion-dollar global supply chain.
“It’s not about doing data science faster,” said Gupta. “It’s about doing all of it; every model, every use case, at once. That’s where agents change the game.”
And for companies wrestling with analytics bottlenecks, unpredictable demand, and complex planning environments, Cisco’s approach offers a preview of what enterprise AI might look like when it finally escapes the dashboard.
The Problem: A Manufacturing Line of Models
Cisco’s supply chain spans geographies, customer segments, and verticals. Forecasting demand across all of it is one of the hardest problems in enterprise operations. And for the last four years, Puneet’s team has been building a causal AI system to do it, with over 3,000 models, 6,000+ features, and hierarchical causal graphs that span products, timeframes, and business units.
“We spent two years building the right data structure just to support meaningful modeling,” Gupta said.
But even with a mature stack, the process hit a wall. Feature engineering took weeks. Evaluating models took human judgment. The backlog grew with every new business requirement. And trust, especially from non-technical stakeholders, was fragile.
The solution wasn’t just automation. It was agents.
The Shift: Agents as Teammates, Not Tools
Cisco now uses causaLens agents to support everything from EDA to model inspection and explanation. These agents don’t just assist, they learn from past modeling work, apply reasoning across business contexts, and help junior data scientists perform like seniors.
One example? A forecasting agent that can:
- Read a user’s question (e.g. “What will net bookings look like six months from now?”)
- Pull the right model
- Generate a forecast
- And explain the causal drivers behind it using enterprise-specific language
“If you don’t have a PhD economist on your team, this agent is the next best thing,” Gupta said.
Why This Matters for Supply Chain Leaders
Forecasting in supply chains has long been limited by two things: data complexity and human bandwidth. Most companies build models in silos, assign them to senior analysts, and struggle to scale when new variables emerge (like regional macro factors, product transitions, or sales-driven demand shocks).
Cisco’s approach points to a different future, where AI agents don’t just produce forecasts but manage the entire analytics workflow across every tier of the business.
Here’s what that unlocks:
- Faster model development: Agents rapidly select and test features, even from unstructured sources
- Scalable insight: Long-tail models (the 300th or 1,000th) get the same attention as top-priority SKUs
- Business-friendly output: Forecasts are delivered with causal narratives, not code or jargon
- Reusable intelligence: Agents trained on one problem generalize across regions, segments, and new products
And perhaps most importantly, they reduce dependency on a few power users.
“There’s no one person on my team who can ‘make the call,’” Gupta explained. “Agents help us build trust in decisions, even when they’re complex.”
Where It’s Going Next
The goal isn’t to replace humans, it’s to make the humans better, faster, and more focused. Over the next 12 months, Cisco expects agents to handle more of the middle tier of forecasting tasks, models that are too important to ignore, but too time-consuming to own manually.
“Today, output is still mostly PowerPoint,” Gupta said. “Twelve months from now, I see agents summarizing model families, surfacing key scenarios, and driving decisions directly.”
That future, where AI Agents don’t just assist but operate, is what we call the digital workforce.
Cisco is already building it.