Unframe Shares the Key to Successful AI Integration

In a recent AI case study webinar, Cushman & Wakefield CIO Colin Campbell spoke with Chris Slovak, CTO at Unframe, C&W’s partner in the firm’s recent AI transformation.

Unframe is managed AI delivery platform that turns enterprise needs into tailored, production-ready AI solutions in days.

Edward Cohen, director of technology and AI partnerships and host of the Tangent podcast at Commercial Observer, hosted the conversation on behalf of Commercial Observer Partner Insights.

One of the first issues the pair addressed was whether a company required a complete data refresh before taking on the task of AI integration.

“Where technology projects start to fade is when you try to take on too much at once,” said Campbell. “We always say that firms should take the 10 things that matter, start there, and then find the next 10 things that go after that.”

Campbell advises companies to start with high-impact domains and the data they believe is the cleanest, and use that as the foundational layer.

“There’s a very simple phrase that sits across the top of our AI development that says that a fancy model or tool on bad data just gets you the wrong answer faster,” said Campbell. “As you localize into a particular sub-segment in real estate — that could be in preventive maintenance, valuations, or any number of the core domains — pick the things that matter most and always start there. Don’t pick everything.”

Campbell noted that maintaining strict governance and staying laser focused on outcomes from any particular data set, and then moving on to the next one, creates momentum that eventually speeds the process of solving other issues in parallel.

Slovak said that in determining which data can and cannot be appropriately cleaned, there’s an interesting conversation now about “how agents and agentic-to-agentic communication might actually change what we think about cleaning, cleanliness and quality of data.”

“Conceptually, as humans operate in imperfect data sets, agents can too,” said Slovak. “We’re not there today, but it seems like that might be a factor in decision-making as we go forward.”

Campbell then turned to the subject of confidence scores, asking Slovak about Unframe’s approach to determining the right level of information to create a desired outcome, and how this would be handled when the communication is agent-to-agent.

“I’m always a big fan of saying consistency first, over accuracy,” said Slovak. “Not that accurate data isn’t super critical. But you get into the conversation around, how accurate does something need to be in order to be informative? I’ll take a B+ answer every time over, say, an A+ followed by an F.”

Slovak noted that it’s this sort of consistency that builds the trust executives need to assist in their decision-making.

“Executives need to know if this is the right decision to make at this point in time,” said Slovak. “They can come back around on a perfect number, often within their reporting, but most of these systems are trying to figure out how to make recommendations and provide answers.”

As agent-to-agent communications increases, then, human users will need confidence in the information being communicated.

Campbell noted that as long as the data points a firm in the right direction, that data can be made tighter and more reliable over time.

Slovak replayed that this is the road to building trust in these systems.

“If you’re consistently getting answers that are reasonable and not wildly fluctuating, the human side of you can trust it,” said Slovak.

“It takes a long time to build trust, but a very short amount of time to lose it,” added Campbell. “If you take that humanistic approach into your data journey, you’ll be very successful, because that’s what leads to consistent outcomes and a better-trained model.”

The pair then turned their attention to the issue of build vs. buy, with Campbell noting that C&W had historically been a “buy” shop, but eventually transitioned more toward “build” — without leaving “buy” completely behind — due to the ease of creating more complex and multi-feature toolsets.

“We had one tool per problem versus a toolset that would historically span a series of problems,” said Campbell.

Campbell noted that this became especially apparent as C&W worked with Unframe to transition into AI, noting that the buy strategy becomes less productive as the technology becomes more complex.

“Early on, we needed much more of a ‘buy’ from accelerators like Unframe to really drive value, because they helped us think about new ways of working,” said Campbell. “But when you come from a buy-based model, you come from a one-tool, one-problem type mindset. That one-to-one ratio no longer applies with modern technology. So that’s how the buy-versus-build conversation changed with us, and then we take that same logic and apply it everywhere.”

The conversation then turned to budgets and tokenmaxxing, as Campbell noted that while budgets have historically been viewed through the lens of dedicated projects or targeted opportunity, an increasing need for greater outcome-focused technological flexibility is shifting that approach.

“If I’m budgeting for an accounting or preventative maintenance transformation, I’m allowing myself the grace to understand that the ability to budget isn’t limited to just a singular opportunity,” said Campbell. “It’s within a domain that allows us to find more opportunities and greater potential value.”

Reinforcing that everything they do has an ROI-based outcome, Campbell said he communicates regularly with the company’s various businesses, getting a clear-eyed view of the maximization of value and adoption, and allowing budgets to adapt accordingly.

As far as tokenmaxxing goes, Campbell said the company doesn’t so much set token limits as allow for them to become more established as the platform scales to maximize value.

“We don’t set out at inception with a token limit,” said Campbell. “We achieve that in alignment with the business to get the value, and then, seeing how achievable it is, to then get a more efficient, optimized model.”

As for the issue of what happens to SasS as all these changes roll out, Campbell believes the systems will adapt, as opposed to being replaced.

“I don’t believe in the SaaSpocalypse,” said Campbell. “Your core systems are going to stay, but that doesn’t mean you shouldn’t be looking at how you maximize your differentiation — your target on what makes sense. If your differentiation is going to be in a certain area, then create the best opportunity within that space, find the right partners, and do it yourself. I would focus on the thing that makes you uniquely you and drives value.”

Brief: A webinar about how Unframe helped Cushman & Wakefield integrate AI into its systems dealt with issues of data integration, buy vs. build, tokenmaxxing and the future of SaaS.