Take a calming deep breath. Pause for a moment. Meditate on generative AI and the plans you and your peers have laid for its deployment in your enterprise.
Were the plans thoughtful or slapdash? Are your initiatives proactive for the business and the front-line employees or hype-reactionary FOMO, perhaps driven by a panicky CEO or board’s edict? Do you have something measurable for the CFO — the reaper — when she soon comes knocking at the door wanting justification for further funding of your team’s GenAI vision?
If the answer is “no,” you’re not alone.
Analysts report that few of their enterprise clients, if any, have figured out how to monetize or even measure generative AI technologies’ returns on investment.
Now it’s time to panic and develop more granularity for your generative AI strategy.
Endless vendor hype around generative AI would have us believe it’s a panacea to cure all enterprise IT ills or, perhaps, a portal to a future of undreamt-of revenues. But it’s just a tool, Constellation Research analyst Liz Miller reminds us.
As one wouldn’t pick up a hammer, examine it and buy it in hopes it does something good. companies shouldn’t just bolt gen AI tools to their application stacks because they can. You need a nail, or a workflow problem, to fix.
“The hyperbole around GenAI is the show — the circus act that gets you into the tent,” Miller says. “The real show is the tasks, processes and operations that generative AI models, combined with dat … can do something differently and more effectively [with]. Any AI needs data, governance, rules and restrictions to be an effective addition to any use case. But none of that sounds sexy enough to be included in the hype machine.”
GenAI’s business value amorphous for most
The reason few enterprises have figured out what to do with GenAI is because there are as yet few clear-cut examples where GenAI creates measurable efficiencies, said Alan Pelz-Sharpe, Deep Analysis founder.
One is customer service. Managers already measure time for tasks performed by individuals and track metrics for the contact center, such as time-to-answer and hold times. Contact center leaders typically can draw from years of pre-GenAI baseline data at their disposal to show improvement, or lack thereof, a new tool might bring to their workflows.
In settings such as these, there are long-established, standardized ways to determine performance. GenAI will either help agents more quickly to find answers to customer issues and then help them write answers faster or not. Where an enterprise has already been measuring something, GenAI strategy is more straightforward.
Customer satisfaction should be factored into a GenAI strategy as well. For example, most contact centers constantly measure customer satisfaction with automated surveys. “Before and after” comparison studies will quickly prove or disprove whether a GenAI tool pays for itself by saving time and whether customers approve of its “help.”
Ultimately, because the contact center use case is so obvious, AI might replace some agents, Pelz-Sharpe said — even more than some.
Beyond the contact center, though, GenAI’s measurables turn a bit opaque. In theory, rolling out efficiency tools such as Microsoft Copilot to every employee in an organization should save a lot of time, increase productivity and bring a cornucopia of benefits, as Microsoft promises. But it’s hard to quantify, unless an organization has a long data record of measuring common business processes that Copilot seems like it should speed up, such as writing requests for information.
On top of that, before turning on GenAI tools, companies must assess the hidden costs beyond the subscription or licenses.
The first comprises identifying data to train the AI system and then the access and normalization it requires. Next, AI tools must be studied to determine how they fit into a company’s data-sharing protocols as well as its security policies, corporate compliance and software. After all that has been done, there’s educating employees on the potential of a particular GenAI tool and getting them to use it to realize all the great, yet theoretical, potential.
When prospective GenAI users add all that up, they still might not arrive at a number.
“There are specific areas in which you can say that time savings actually [lead] to return on investment in the form of reallocation of resources. But right now there are not a lot of hard ROI metrics on, say, revenue or hard-dollar savings at this time,” Wong said.’
Fix your GenAI business plan
Copilot is unquestionably the most fully formed GenAI product available now. As the old IT aphorism goes, no one ever got fired for buying Microsoft.
Wong suggested that companies might start sharpening their generative AI strategy by identifying resource-constrained areas of the enterprise and evaluate how GenAI might relieve some of those constraints. Those are the first targets.
Improved employee satisfaction is another benefit that can’t be measured in dollars but is definitely a positive.
“Think about the employee value if you are providing them with better tools and you’re making the case for new hires to come work with you because you are providing more advanced tools,” Wong said. “The people who use them say they are satisfied; they feel more like it’s helping them reduce some of the manual work. It’s ‘return on employee’ instead of return on investment.”
But overall, according to Gartner’s recent “Establishing a 2024 Microsoft 365 Copilot Strategy” guidance, it could take years to determine the productivity gains Microsoft Copilot could yield. Enterprises, Gartner said, should negotiate the lowest possible discounted license price for Copilot, lock it in for three years and make sure the agreement covers new seat licenses to be added at the same discount.
Furthermore, Gartner advises companies that use third-party Microsoft support to ” assess the vendor’s ability to support Copilot and the thresholds necessitating escalations to Microsoft.” That sound advice will probably work for negotiating with many vendors hawking GenAI wares under the circus tent.
Even though calculating the business value of this technology might be vague and imprecise at the moment, Miller is bullish on its long-term promise.
“Take the example of Adobe Firefly,” Miller said. “While we focus on the amazing images and assets created when the right prompt hits the right model, we lose sight of the fact that a human could have created the same exact asset, but it would have taken hours, if not days.
“Firefly hasn’t done the impossible. It has done the improbable by creating in seconds, not hours. This gift of solving the improbable gives our human artist back the capacity to think and dream bigger. … These copilots are enabling people to achieve the impossible in both scope, scale and time.”
Don Fluckinger covers digital experience management, end-user computing, CPUs and assorted other topics for TechTarget Editorial. Got a tip? Email him.
