When your company makes its biggest decisions, whether that’s launching a new product, reorganizing teams around AI, or setting next year’s priorities, every plan rests on one question: do our people have the skills to get there?
According to our latest survey, six in 10 managers say they make workforce decisions based on assumptions rather than reliable data. And while 84% of managers say their organization increased AI investment over the past year, only one in five can measure the return.
“Before you decide who’s ready for a bigger role or where to invest in development, you need to know what people can do today. Measuring skills from real work gives leaders evidence they can act on and employees a clearer path to grow,” said Tigran Sloyan, CEO and Co-Founder of CodeSignal.
We surveyed 2,000 people managers and knowledge workers to get a view into where workforce skills data falls short, how people are building AI skills, and what leaders can do to better inform their future decisions.
Key takeaways from this report
- 62.5% of managers said they made workforce decisions based on assumptions in the past 12 months.
- 86.4% of managers say that once hired, candidates perform differently on the job than their resumes or interviews suggested.
- 64.4% of managers rely on manager observation to build their workforce skills data.
- 84.2% of managers report rising AI investment in the past 12 months. Only 20.6% can measure the ROI.
- 71.9% of employees said that access to AI training that fits their job was a barrier to building AI skills.
62.5% of managers make decisions based on assumptions
The survey asked managers about workforce decisions such as hiring, promotion, reskilling, project staffing, and restructuring. 22.4% of managers said they frequently make decisions based on assumptions rather than reliable data, and 40.1% said they did so occasionally. Just 14% said they never make decisions without data.
That puts a practical question on the table for HR leaders: what evidence can your team reach for when a people decision needs to be made?
“A resume is a story someone tells about their skills. The work is the proof,” said Sloyan. “Right now most companies are making their biggest decisions about people based on the story.”
And the data supports this, according to people managers we surveyed, 86.4% say that once hired, candidates perform differently on the job than their resumes or interviews suggested.
This carries through with upskilling and reskilling too. A strong quarterly review can tell you if a manager thinks this employee delivered. It takes a closer look at the work to understand the skills behind that result and whether those skills fit the next assignment.
Consider a sales rep who consistently hits quota. Before putting that rep in charge of launching a new product, you’d want evidence of how they explain unfamiliar ideas and respond to objections. The quota alone can’t answer those questions.
64.4% use manager judgement to build their workforce skills data
When asked about the biggest barrier to determining what skills their employees have, 16.9% of managers think skills are changing too quickly, 13.9% say that data is scattered across systems, and 12.7% think it’s a lack of the right tools.
So if managers don’t have the resources to make decisions based on reliable data, what sources are they using?
Manager observation or judgment was the most selected way to determine employee skills, at 64.4%. Employee self-assessments or surveys drew 47.1%, and certifications or credentials drew 45.7%.
Manager observations can be valuable, but highly subjective. The challenge is making those observations specific, consistent, and up-to-date enough to compare. Which skill did someone demonstrate? What work supports that judgment? How recent is the evidence?
84.2% report rising AI investment, but only 20.6% can measure the ROI
Based on the data above, we can see a clear trend: proxies are often being used to make some of the biggest people decisions. AI puts this proxies problem in the budget conversation.
84.2% of managers said their organization’s investment in AI tools and capabilities had increased over the past 12 months. Leaders need to know how people’s work is improving thanks to that investment.
Managers often feel they have a view of readiness: 41.7% described their visibility into their teams’ AI skills as very clear, and 42.7% as somewhat clear. Yet on the employee side, 71.6% said their AI-related skills had never been formally assessed or verified.
And when asked what their organization could measure effectively today, only 20.6% of managers selected ROI from AI spending.
Considering that so many of these leaders are increasing investment in AI, including increasing the salaries they are willing to pay for AI skills, this is where we see a big potential for skills validation.
“Most managers think they know their team’s AI skills. Most employees say no one has ever checked,” said Sloyan. “That gap is where bad decisions and overlooked talent live.”
Training has to make it into the work
In the past year, most people teams have realized the importance of incorporating AI training into their L&D plans. What most teams haven’t been able to master is ensuring that training is applicable to real jobs and ends up showing up later in the work.
62% of workers received some kind of AI training from their employer.
Among the 616 employees who had received employer-provided AI training, only 26% said they could apply most or all of it in their daily work.
Employees also pointed out that this has become a barrier to building AI skills. Of all 1,000 employees surveyed, 71.9% said that AI training that fits their job or role was a major or minor barrier.
“Companies are buying AI courses or building their own, but if most workers are going elsewhere to learn, you have to wonder if that training is actually helping,” said Sloyan. “Training only works when it’s built around the job someone actually does. Right now most training options, whether purchased through outside sources or created internally, are built for entire teams or departments, not specific enough to see the development in someone’s work.”
Thinking back to our sales rep example, a course for the entire sales department doesn’t help a salesperson handle the objection they struggled with on a call. We need to base training around the work someone actually does to truly give them a chance at improving their skills.
This is where we see a greater opportunity for L&D: being able to not only measure if the training is being done, but if employees can actually apply it to their day-to-day work.
What this means for your talent decisions
For your next promotion, project assignment, or development plan, start with evidence of what people can do today.
Here are the actions you can take right now:
- Audit your evidence. Check whether titles, self-ratings, and course completions show the skills a role requires or leave you guessing about readiness.
- Measure demonstrated capability. Evaluate recent work against clear, job-relevant criteria so employees and managers have a shared basis for discussing readiness.
- Pair learning with practice. Target coaching and feedback to the gaps the work reveals, then reassess to see whether skills improve.
- Track business outcomes separately. Check whether stronger skills translate into the quality, productivity, or cost improvements you set out to achieve.
CodeSignal’s agentic skills intelligence measures employees’ skills from work they already produce in tools such as Salesforce, GitHub, Zoom, Claude, and more. Agents evaluate that work using rubrics grounded in I-O psychology and tailored to an organization’s competencies. Cosmo, our AI coach, helps employees build the skills they need through practice drawn from their own work.
Not sure where to start? Check in on your current approach. Take CodeSignal’s free skills validation quiz to get a scorecard on how your organization validates new hires, develops existing employees, and trusts its skills evidence.
Methodology
CodeSignal’s workforce decisions study was conducted with two separate surveys: one with 1,000 people managers and one with 1,000 employees.
The study was limited to knowledge workers who use digital tools in their day-to-day jobs.
Unless otherwise stated, each survey finding uses a base of 1,000 respondents in its respective group. The training-application question uses the 616 employees who reported receiving employer-provided AI training. Multi-select questions allowed respondents to choose more than one answer, so those response percentages can overlap.