What Causes AI Efforts to Fall Short?
MIT researchers recently found that 95% of enterprise AI implementations yielded no return on investment. ‘That is a sobering statistic’, Hannah Calhoon, VP of AI for Indeed, said at the FutureWorks event.
One big problem, she argued, is that companies rush to adopt AI without a coherent plan. Too often, she said, they take a one-size-fits-all approach to rolling out AI, without first mapping out how the tools relate to specific jobs.
At Indeed, Calhoon said the company has taken a function-specific approach. ‘We do not go in and say AI is generically good everywhere’, Calhoon said. Instead, her team starts with a defined role, like sales or marketing, and asks for specific pain points AI might fix.
‘You get business impact faster because you have gone in to fix an actual problem that you understand’, she said. ‘It is tailored to exactly the task that they are doing on a daily basis’.
How Do You Define AI Fluency for Different Roles?
AI fluency is not one-size-fits-all. What it looks like depends on the role and the outcomes teams are trying to improve.
Zapier, for example, has created a hiring framework that breaks down AI skills and mindsets by job function. It groups candidates into four categories: unacceptable, capable, adoptive and transformative.
A job candidate for Zapier’s people team, for instance, may be assessed as ‘unacceptable’ (meaning they would not get the job) if they dogmatically distrust all AI hiring tools. They would be considered ‘capable’ if they save two hours a week using ChatGPT to draft interview guides. But ‘transformative’ candidates might use AI to streamline the recruiting process and reduce time to hire by 50% while boosting offer-to-accept rates by 10%.
How Can Interviewers Assess Job Candidates’ AI Skills?
Zapier now uses its framework to assess AI skills in 100% of candidates. That does not mean giving every potential hire a technical assessment. But it does mean, say, asking a marketer how they use AI to generate content, or asking a prospective talent leader about hiring processes they have rebuilt using AI.
Zapier is looking for candidates with some degree of AI fluency, Sammut said. ‘If we are hiring someone who is coming to Zapier to help us reimagine how a certain type of work gets done with AI at the core, I really hope we are hiring them at that transformational level’, he said. ‘Yet for some jobs, we will also hire folks with less AI experience who have self-taught to some degree and have a deep understanding of where AI can help – even if they can not rebuild those practices with AI by themselves quite yet’.
Should AI Abilities Factor Into Performance Reviews?
Evaluating AI fluency among current employees can be tricky, since managers often see skill development as their responsibility. In Indeed’s 2026 report ‘Smarter Hiring Insights: Solving the 7 Disconnects that Make Hiring Harder’, roughly half of employers said managers are responsible for reskilling and upskilling their teams. Given this, experts say it would be premature to penalise AI-wary employees.
Ruth Hickin, VP of Workforce Innovation at Salesforce, said it is too early to tie AI skills to performance reviews, though the company uses self-assessments and manager reviews to gauge fluency. As AI increasingly boosts productivity, she added, that may change.
If an employee opts out of using Cursor, the AI-powered code editor, ‘it is probably impossible for them to code at the same speed as a colleague’, Hickin said at FutureWorks. ‘We are going to have to make decisions around whether or not that becomes a performance issue, but we are not there yet’.
At Zapier, Sammut said the company sets goals around AI fluency and adjusts performance expectations if those goals prove feasible. On the customer support team, Zapier tested whether AI could cut response times in half over a year.
When the team met that goal without sacrificing customer satisfaction or employee engagement, they incorporated this type of AI usage into performance expectations. ‘It really invited the team into experimentation’, Sammut said, noting that employee engagement scores during this period actually ticked up by 20-30 points on a scale of 100 during the implementation period.
How Can Companies Keep AI Skills Up to Date Over Time?
AI fluency, of course, is not a destination. As technology evolves, so will AI skills.
Indeed has taken a layered approach to training, says Calhoon. The company offers AI 101 training to all employees, as well as group-based training for specific roles and peer-led training taught by ‘power users’ at the company.
“No matter how many times I or any other leader at Indeed tells folks to use AI tools, the people they’re actually going to believe are their colleagues,” she said.
At Salesforce, Hickin said the company requires all employees to take a certain level of AI training. Those courses can double as benchmarks, measuring where employees’ AI skills stand across the company.
For instance, on Trailhead, Salesforce’s online learning platform, employees can take a course on building AI agents and progress through levels from ‘champion’ to ‘innovator’ to ‘legend’. In 2025, Hickin said every employee was required to reach at least champion status.
As AI reshapes work, the challenge for many companies is not merely encouraging adoption, but encouraging AI fluency. Using AI occasionally is not the same as knowing how to use it to redesign processes and set new expectations. The organisations that figure out that distinction early will be better positioned to keep pace as the technology evolves.
