Sandeep Davé knows the value of experimentation as well as anyone. As chief digital and technology officer at CBRE, Davé recognized early that the commercial real estate industry was ripe for AI and machine learning enhancements, and he and his team have tested countless use cases across the enterprise ever since.

And those experiments have paid off. Over time, using machine learning and AI, CBRE has managed to reduce manual lease processing times by 25% and cut positive false alarms in managed commercial facilities by 65%. CBRE has also used AI to optimize portfolios for several clients, and recently launched a self-service generative AI product that enables employees to interact with CBRE and external data in a conversational manner.

Recently, the company announced a significant milestone: deploying its AI-enabled Smart Facilities Management Solutions at more than 20,000 Global Workplace Solutions client sites, totaling 1 billion square feet. Even so, “we are still in the early innings” with AI, says Davé. 

Davé and his team’s achievements in AI are due in large part to creating opportunities for experimentation — and ensuring those experiments align with CBRE’s business strategy. While many CIOs may still be wondering how to begin their organization’s AI journey, Davé’s work at CBRE shows that facilitating experimentation, even if it leads to some failures, can deliver outsize success.

Here’s how Davé has made AI experimentation work for CBRE’s bottom line, as well as his advice to IT leaders looking to do the same at their organizations.

Build a self-service foundation and capture ideas

Eager to deploy AI, many organizations begin by defining and ranking use cases. But those who have success with AI know training data is key. So, a better approach might be to build a data foundation and give your people time to explore the possibilities first.

Davé and his team made did just that when they recognized the power of data at scale. CBRE has a vast amount of transaction data, as well as a plethora of asset intelligence from sensors, workflows, and human interaction with physical space within the billions of square feet the company manages globally. This early work has enabled automation in areas of the business such as lease abstraction or automating work order classifications.

With hype around generative AI exploding of late, the CBRE team developed a multi-LLM, self-service generative AI platform that enables employees to use gen AI for a range of tasks, such as gaining insights from proprietary data and documents, using chatbots to work through problems, generating new content, and translating forms. By making the platform widely available, “we’ve created an appetite and an interest across the organization,” Davé says. “[The product] has hundreds of users and is growing weekly. And it’s unlocking a lot of productivity.” It’s also setting the stage for more innovation across the company. When the technology is available in a self-service manner, “lightning can strike according to its own schedule.”

Still, Davé stresses the importance of safety restrictions when it comes to AI. “You’ve got to be careful how you use [AI] and how you educate your users,” he says. “Human intervention is still necessary. Validation is still necessary. And it’s very important to remain mindful of technical limitations — such as hallucinations — and legal obligations with respect to how we use client data.”

Choose use cases aligned with business priorities

Once you’ve given your people time and resources to experiment and you’ve captured good ideas, it’s time to select the best opportunities to pursue. Here the key is to separate the flashy from the substantive. “We’ve seen so many initiatives fail when it’s technology for technology’s sake,” says Davé, who suggests two means of avoiding this mistake: prioritization models aligned to your business strategy and strategic partnerships.

Let’s start with the models. Davé and his team filter use cases using a simple, time-honored method: plotting them in a two-by-two grid that takes as its axes “value” and “feasibility.” Davé starts with cases that are both high-value and high-feasibility for quick wins to generate excitement and buy-in from stakeholders. “These have the most potential because, typically, they draw on data we already have access to and that we’ve already made good use of,” he says. “In the case of AI, many of these are productivity enhancers. They eliminate manual and repetitive processes.”

The quadrant Davé’s team attacks next is either “high-value, low-feasibility” or “low-value, high-feasibility,” depending on their objectives. It’s a choice between low-hanging fruit and big investments. For AI, the high-value quadrant is where you’ll find most predictive modeling. “These are not easy, but they have a big impact if you get them right,” says Davé, adding that IT leaders should consider picking a use case from each of these two quadrants: one that’s high-value and one that’s highly feasible. That way, your team can demonstrate early results while helping to develop momentum for the larger initiative.

As elegant as it is, though, the value-feasibility matrix suffers one serious drawback: It’s plagued by the same ambiguity that almost all prioritization models face. After all, how do you assess the value and feasibility of use cases that rely on an emerging technology so little understood, or that require building capabilities that may not immediately yield benefits? Here, partnerships can make a huge difference in both risk mitigation and improving time to market.

The importance of strategic partnerships