James Cham, partner at seed fund Bloomberg BETA, was at Cisco Collaboration Summit today talking about the importance of models to the future of machine intelligence.

Models fit between code and data, he said. They do some of the hard work that’s different than what apps do. But they’re not always right, he cautioned.

“If you could always get the right answer,” he said, “you wouldn’t need a model.”

Cham talked about an array of companies that use machine intelligence and models to deliver value.

• One example was AppZen. The company automatically detects unauthorized charges from receipt images. Those receipts may include boarding passes, cell phone bills, and the like. AppZen helps businesses spot possible expense fraud – and without dedicating an army of people to sort through piles of receipts. “It finds money on the table” and may encourage employees to avoid doing things like double-charging dinner expenses, said Cham.

• He also talked about Motiva, a marketing automation layer on top of Oracle’s Eloqua. More often than not, marketing messages are dumber than they could be, he said. But with better segmentation and other efforts, businesses can deliver more targeted marketing messages to people. Motiva allows that.

• And Primer is a company that collects articles on popular topics and synthesizes them into single, Wikipedia-type pieces. That way, analysts don’t have to read a sea of content. The model in this case identifies certain kinds of content. Cham said you can build apps on top of that to do things like see how media in different parts of the world cover terrorism, for example.

Models will be small and specialized, he added, and will be integrated into workflows. In addition to models, it’s important to have the right data and managers involved in driving machine learning efforts forward.

But data may not be easily accessible for organizational reasons (like someone simply doesn’t want you to see it) and/or because people aren’t sure where to find it, he said. Data may be intentially wrong (like if a doctor inputs information based on how they want to be compensated instead of based on patient diagnoses or needs), he suggested. Or an organization may not be considering all the data they should to get the best picture of what’s happening.

Cham also suggested there’s a middle management gap when it comes to transforming organizations using machine intelligence. Leaders throughout business need to be involved to move such efforts forward.

Machine intelligence as Cham likes to call it (or AI or machine learning as it’s more commonly known) is about doing math really cheap, he said. And that can unlock some surprising opportunities.

In the MI/model-centric world, he added, things iterate much faster than code. But although these models can improve quickly, they can also go very very wrong. (Without mentioning Microsoft’s AI bot Tay by name, Cham mentioned how these things can suddenly become racist or spawn other negative results.) So, he suggested, you need to monitor them.

Another challenge is there are no methodologies in the MI world, in part because the people involved with these efforts are not yet clear on what they’re doing.

However, he suggested, you could use the following metholodologies to build models:

• Split labeled data set

• Build models until you’re happy with performance

• Retrain the best model on all labeled data

Cham concluded his presentation by displaying photos of Bill Gates and Marc Benioff on the screen. Before these leaders created the software and as-a-service markets, the idea that software was a product or anyone would want to run their applications in someone else’s network were new and strange, he reminded the audience in Phoenix. Now, he said, there’s an opportunity for businesses to use machine intelligence to define new markets. And while companies like Amazon have already introduced consumer-facing solutions like Alexa, Cham believes that MI-powered models that companies use internally to enable improvements offer the biggest market potential.