This wasn’t the article we wanted to write.
We wanted to write an article with case studies from mid to small sized businesses who are seeing real, quantifiable ROI from AI. So, we talked to several sources, messaged some of our LinkedIn followers and asked a few consultants if they knew of anybody or anything — nothing. Heck, we even asked AI! See response below:
Most “AI ROI case study” content right now comes from AI implementation vendors and consultancies (Mindpath, AI Monk, Titani, iApp, Alice Labs), publishing SEO listicles with numbers like “260% ROI over 18 months.”
Well, that sounds fishy. So, instead, we turned this into a how-to piece about the steps you can take to achieve ROI with your AI investment. And if six months to a year from now you’re able to do that…send us an email or LinkedIn DM and tell us about it. We’d love to write about it.
In the meantime, we think the reason why ROI with AI is so hard to find is because…maybe it just doesn’t exist yet? A study by MIT’s Project NANDA (Networked Agents and Decentralized AI) found that 95% of enterprise GenAI initiatives deliver no measurable return.
Kristin Ginn, founder of trnsfrmAItn, who works on the human side of AI adoption inside large organizations, isn’t too surprised by the study’s results. When her clients tell her, “we need AI,” her next question is: What does ROI mean to you?
Most of the time, she’s met with blank stares.
“Most companies don’t understand what ROI means for AI and what they’re looking to get out of AI,” Ginn says. “Is it for automated processes? Is it to upskill workers? Is it for efficiency? You have to define what you want the win to be, before you invest in AI.”
Josh Streets, an AI consultant and trainer for HDI and ICMI, says the ROI question gets skipped, largely because most AI purchases bypass the same scrutiny that would apply to any other technology investment.
“With most tech investments, there’s usually a business case analysis to get funding and to get approval to move forward, at least even in a proof of concept,” Streets says. “Why that’s not happening with AI, I don’t know. I almost feel like people are moving so fast trying to compete or catch up with other companies, that they throw all caution to the wind.”
But if you’re reading this article, we know you’re the kind of person who is thinking cautiously and strategically before you sign off on an AI tool. We hope you’ll skip the vendor listicles about ROI and for now, start building the framework on how to get there.
Explore automated call summary and agent assist
One of the initiatives that Streets sees working is the automated call summary. The technology transcribes a call in real-time and generates a short recap for the agent’s notes, tailored to whatever the business needs, whether it’s clinical details for healthcare or incident specifics for retail.
Based on current industry data, Streets puts the first-year ROI of an automated call summary between 150% and 500%, with a typical payback period of three to nine months. He says that’s faster than almost any other AI initiative in the contact center space right now. Streets says the reason why it works is that the value is easy to isolate. It replaces a specific, measurable task with something automated.
Streets recommends putting agent assist on your radar, too. This is an AI-powered software tool that can help your agents during live phone calls or chats. However, Streets says agent assist is one of the easiest AI use cases to get wrong because agent assist means different things to different AI vendors.
A shallow version requires someone to manually program every question an agent might get asked and the corresponding answer. This limits its usefulness and requires constant maintenance. A properly built version listens across voice, chat and email in real-time, ties directly into the company’s knowledge base and surfaces the right answer without anyone pre-programming the question.
“A shallow implementation might show usage numbers without meaningful impact,” Streets says. “A properly built one reduces AHT (average handle time) and improves first call resolution, two metrics that affect your company’s budget.”
Speaking of budgeting…
When it comes to AI, you’ve got to budget for the people, not just the license.
AI initiatives only pay off if the organization budgets correctly, and Streets says the biggest miscalculation companies make is treating AI like a typical software purchase. He says you need to bring in product owners, conversational designers and subject matter experts to the initiative. Squeezing AI oversight into someone’s full-time day job isn’t going to work.
The second hidden cost is dependency work that gets discovered too late. Streets describes a recent client who wanted to replace a tool that wasn’t performing, only to realize their knowledge management articles had never been updated in the first place.
Encourage your employees to build AI habits
Even when the technology and the budget are right, Ginn says most AI rollouts still fail because companies treat AI adoption like flipping a switch instead of building a new habit. She compares it to a company-wide software migration, like moving from Gmail to Outlook, where the old tool eventually gets shut off and employees have no choice but to adapt.
“That is not the case with AI,” Ginn says. “Nothing is preventing you from continuing to work the same way you always have for years or decades on the job. A single prompt-writing demo from your IT department won’t change how people work.”
Ginn offers her clients a structured 30-day immersion process where employees are given time and space to experiment with AI inside their role. She says this approach works because AI adoption isn’t uniform across an entire company. Another thing she does with all her training is to make it fun. For example, she’ll have people use this prompt, “Summarize my entire day in pirate-speak.” If people have fun and experiment with AI, she says they’re less likely to resist trying it out.
Set a realistic timeline for when you can expect to see results
Ginn says leadership expects results on a timeline that has nothing to do with how AI adoption works. Ginn says the confusion often starts with the assumption that turning on a tool produces immediate impact.
“If you’re bringing in AI to automate entire workflows, you might be able to measure that faster,” Ginn says. “But if you’re talking about generative AI tools that people are using for their day-to-day, it can take months or even a year before you start to see impact.”
Ginn says the first milestone worth tracking isn’t ROI at all, but usage. Next, look for adoption to see if the tool is woven into daily work rather than opened once and abandoned. Then, ROI starts to show up, typically six to eight months after adoption takes hold, sometimes longer.
But Streets says that six-month timeframe isn’t realistic for every company.
“You have to take into account the organization’s AI literacy,” Streets says. “Not every company has the knowledge and teams available to get them to be able to accurately measure ROI after six months.”
And Streets warns about being too obsessive about calculating ROI with AI. If you move too fast and skip the planning, you’ll end up without any measurable results. But if you plan too cautiously and try to account for every dollar before getting started, the initiative can stall out entirely.
“I’ve seen organizations stuck for a year trying to measure the potential ROI for certain use cases and they haven’t moved forward because they were trying to plan every single penny that would come back from it to get it approved, and they missed the boat on future AI investments,” Streets says. “You’ve got to have the right person to own the process. The sweet spot is having a really savvy internal leader that knows how to balance governance in the right area and leveraging a financial team to help run the business case analysis.”