In this era of digital transformation, the specter of spending unwisely on digital projects has emerged as a significant financial risk for companies.
So says Tim Raiswell, a financial-data analyst and a research vice president at Gartner, who recently completed a broad analysis of first-half 2018 financials and earnings calls.
On the one hand, robotic process automation (RPA) initiatives are generally succeeding in their goals to save money by having robotic software perform rote tasks historically performed by humans.
On the other hand, many companies are struggling to find impactful use cases for machine learning and artificial intelligence applications, according to Raiswell.
“There’s a lot of machine learning and AI out there,” he says. “Is it all attached to clear bottom-line or top-line outcomes? I don’t think so. It’s more frequently a branding or product feature that gets thrown out there.”
In a subset of Raiswell’s analysis focused on 10 technology companies, he found the fourth-most-common two-word term used in recent earnings calls was machine learning.
“It came above terms like ‘operating margins’ that you would expect to be in an earnings conversation,” he says. “It shows how pervasive it has become as a branding strategy — a must-mention, must-have-in-our-story item. If we’re a technology company and some part of our value chain does not involve machine learning or AI, it must mean that we’re insufficient.”
Raiswell says he “knows for a fact” from talking to companies that a lot of them “hear from vendors who come and say, ‘Hey, have we got a machine learning or AI solution for you,’ and then they struggle to work out what the use cases are.”
With RPA, use cases are much more readily identifiable. For that reason, the technology is becoming very popular, Raiswell notes. “It will, say, grab this data from Excel, drop it in an email, and save me two hours every day. Bang. Easy,” he says.
By comparison, AI typically requires significant customization before it can provide value, which may impact the cost-benefit relationship.
For example, an AI-driven natural language generation solution may aim to translate financial information into words to enable the data to be more quickly interpreted. “It tells you that a revenue increase is tied to this and that metric,” says Raiswell. “But these solutions can be incredibly difficult to deploy, because they’re very sensitive to the underlying data. Once executives get them inside their organization, they find they need a ton more work than the vendor had let on.”
Compounding that problem is another one that’s definitional in nature. If you ask 20 technology leaders what they think big data means, Raiswell says, you’ll get 20 different answers — and it’s exactly the same with machine learning and AI.
Indeed, for many vendors “AI” is coming to mean “anything that seems to mimic or replace something a human can do,” he points out. That’s a problem for a technology buyer, because RPA fits under that umbrella — and in fact, some vendors are positioning RPA solutions as machine-learning or AI solutions. “Who benefits from the muddy waters, the lack of clarity? It’s the ones who are selling these solutions.”
Financial services firms lead the pack in effectively vetting and allocating costs to digital investments, according to Raiswell. They’ve been doing it for longer and tend to be more technology-intensive than other businesses.
A common problem is the legacy budgeting practice of putting big-ticket IT spending into the capital budget.
“The process for getting that approved is very slow in most organizations, because the investments are allocated and tracked at the business level, where budgets get revisited only once a year,” says Raiswell. “The business teams are knocking down the CFO’s door, saying, ‘We’ve got to move faster on this — our competitors are moving, and you’re telling me these budget dollars can’t be approved for another 4 months?’ ”
It’s better for a company’s individual businesses to approve and track IT spend at the operational level, identifying how it’s going to help improve sales product by product. “That’s what the smarter companies are doing,” Raiswell says.
Of course, there can’t be a mentality of “give this one a pass because it’s magical AI, because everyone would be checking that box,” he adds.
For promising new projects with machine learning or AI components, a company should take a conservative approach, the analyst says. Start with a prototype, perhaps funded by seed investments or a small amount of internal capital that will grow quickly if the project looks like it’s going to start succeeding.
“That takes a certain amount of preparation and flexibility so that a CFO doesn’t trade away all of the rigor and discipline,” Raiswell says.
Or the finance team could come up with a set of hypotheses: What would success look like? If we deploy this new AI thing to customers in three months, how will we know when we’re winning?
“That’s about establishing a series of nonfinancial but still economic milestones that help you understand, is this thing winning or losing? Do we want to put more money into it, or pull the plug now because it’s going to become a money pit?”
As an aside, Raiswell suggests the term “digital transformation” itself may be misleading to investors.
Gartner had some requests lately from manufacturers about how to transition to service-based business models. “Many of these businesses can’t sell enough of their products, so they’re moving moving to a leased-based model, where you can access the product through an app. That business is digitally enabled, but is it a digital transformation? Probably not.”