Employers’ neurodivergence inclusion efforts need to include AI
- Key insight: Discover how AI-trained performance systems can systematically disadvantage neurodivergent workers.
- Expert quote: Ramakrishnan: AI trained on neurotypical inputs runs the risk of misclassifying neurodivergent performance.
- Looking ahead: Prioritize inclusive training data before wider rollout.
Source: AI generated bullets with editorial review
As more adults are being diagnosed with autism, and 2.21% of US adults report having an autism spectrum disorder according to the CDC’s most recent data, employers are increasingly responding to the needs of this population, even as they may overlook the role of AI in their inclusion efforts.
Content processing
A study published in the Journal of the American Medical Association Psychiatry found that the overall rate of adults diagnosed with autism doubled between 2011 and 2019, although a 2025 Gallup report, Neurodiversity in the Workplace, found that 37% of neurodiverse employees do not share their condition with colleagues for fear of stigma.
Meanwhile, 91% of organizations are increasing their investments in AI, according to the 2026 Data and AI Leadership survey, 99% consider AI investment to be an organizational priority, and it is included in performance management systems.
Rita Ramakrishnan, founder of Iksana Consulting, an executive coach and current master’s candidate researching coaching modalities for neurodivergent leaders, cautions that these AI-powered performance tools are structurally biased.
Read more:
Ramakrishnan is also familiar with performance review processes, having most recently served as CEO of startup Cadre, and has continued to partner with growth-stage startups as a part-time CEO.
He advised companies to assess their performance systems and where they may be lacking in accounting for the neurodiversity of their workforce, and a critical starting point is how AI is trained.
“The number of performance management or performance management platforms that are centered around AI platforms now with built-in AI capabilities is staggering,” he observed, noting that the technology is often being measured against a narrow behavioral baseline that doesn’t account for traits associated with neurodivergence, and thus reads as poor performance.
Read more:
“The other part is that more and more people are getting diagnosed, and they’re trying to figure out what that means to me, and the community is not a monolith, so you’re seeing multiple characteristics. If we start to demonize those characteristics, we’re going to stop ourselves from being able to design for them,” he continued. “I think there’s a huge opportunity here around performance management, where we can set new standards for what impact means in the context of an organization, and we need to focus solely on outcomes rather than outputs, and I think one of the biggest problems here is oversimplifying that definition of performance.”
Recommendations for companies to evaluate where their performance systems are lacking is to first evaluate the training methodology.
Apply appropriate training
“The challenge is that AI models have to be trained, so you have to give them data on what to look for,” Ramakrishnan explains. “As I was having in-depth conversations with many of these providers, one in particular stands out, where … (he said) the inputs are largely neurotypical, so the expectations of what looks good, of good performance are very neurotypical, and when we think about the inputs that are going into the process and the outputs that are coming out of the process, the ones that match the process don’t match. The brain works.”
One example Ramakrishnan shared was eye contact, which “can be very challenging for those with autism.”
While this isn’t a characteristic of everyone with autism, those who show it “are still able to make needs, ideas, repeat, ideas, and they’re in their roles for a specific reason,” he continued, “but what we’re saying is that we’re making the threshold to become an executive much more difficult. activated?'”
Diversity of data
When discussing AI training methods with a beneficial professional, Ramakrishnan found that they include the company’s values and the manual as a single resource.
“That’s not exactly a rich source of data,” he commented.
Against this example, Ramakrishnan shared an experiment he conducted when he was designing performance management at Square, now known as Block, “where we look at several years of performance data, and…we identified a number of high performers who had been with the organization for some time, and then we took a control sample, and we tested it to identify specific characteristics or skills or attributes that were differentiating between the reviews of the differences in performance between those two samples.”
“So I think that’s a great way to identify, well, these are some of the differentiators that drive high-quality performance in the organization,” he continued. “So being able to identify rich qualitative data sources that can feed the model, so that they’re working from real information that’s not neuronormative and not oversimplified, allows us to think about another important piece of data in business strategy.”
Read more:
For this data to be helpful, there must be alignment around an organization’s key goals, Ramakrishnan emphasized. When asked during his coaching sessions, he shared that five members of an executive team can often recite five different goals.
Once goals and objectives are effectively communicated, the appropriate skill sets can be identified and used to train the AI.
“Putting in additional research (skill sets) is also very helpful, I think,” he said. “You just have a wealth of information available in this day and age that you can feed into a model to make it more efficient, not just in this process, but in many different processes. I don’t know that people are putting in that legwork.”
The business case
This work could be a hindrance to companies seeing that any additional financial investment in their AI systems requires too much time. But Ramakrishnan cautioned against such thinking.
“You can use AI as a very expensive tchotchke or you can use it to get extraordinary value for your business,” he said. “If you’re not going to put in the legwork to set it up right—make sure your human capital is actually getting human responses and setting the right parameters—then you’ve paid a lot of money for very little value, because the data they’re going to give you is unlikely to be robust, and unlikely to be useful in the way you think it is.”
Start small, and be specific
To get the most value from AI in performance management systems, employers need to be as inclusive and specific as possible when defining the purpose of AI.
“Let’s identify the tasks to be done where you’re hiring this AI, this agent for this product, and narrow down the scope there,” said Ramakrishnan, “because an AI agent or AI model will send you what it wants if you’re not specific about the parameters and the recommendations you’re giving it, so be specific, think about it, and give that prompt. solving a task to be done, that’s a critical job.”
Additional changes are also helpful.
Ramakrishnan recommended that you first identify “small use cases” for ways that AI can do your job better, or ways you can do your job faster, or both. Small little use cases, look at your business processes and see what things can be automated. See what things shouldn’t be automated… Start small, then ask bigger questions, then ask where legwork is needed, think and make judgments about human work. is critical, and identifying (where) human judgment is less important.”
Make your people part of the change
As employers look to equip their performance management systems more effectively to include a neurodiverse workforce, these lessons apply to evaluating the AI that all companies are implementing across multiple departments and goals.
“If you do one thing well, if you’re dealing with a neurodivergent workforce, if you’re doing anything with neurotypical populations or mixed populations, AI is not infallible, it’s a very flawed system, but it can add value,” Ramakrishnan advised. “And two, I’ve been a practitioner of change for 17 years, and we’ve known for a long time, you should involve people in change first. You should involve the affected people and support their needs, and yet with AI, there are so many organizations that are just, I’ll throw this out to you, you will think, and we have to think big change.”
