Have a plan when an employee asks if they can use an AI tool.

Nick Iluzada
Dave Shanker’s employees at STS Consulting have been asking lately to use Grok Bot, an agentic assistant tool released by X on August 11.
But before his team can go Grok, Shanker, as managing partner, has to consider certain factors like governance, cost, and terms of service. Also, does a tool they’re already using accomplish the same tasks?
“I’m constantly trying new services out, new products out, and my team is constantly asking for new products and access and everything else,” Shanker told IT Brew.
From a security perspective, Shanker needs to understand where inputted data goes and whether an AI vendor uses it for training. He also reviews liability clauses to understand who is responsible for how that data is used. Before giving broad access, he wants to know what an agent can reach, what it could do without a person approving, and how much damage it could do if something goes wrong.
Shanker’s challenge is a common one today as companies brace for new AI tools and new AI adopters. Cybersecurity company Netskope’s 2026 Cloud and Threat report, released in January, found that the average number of AI apps used in an organization rose from 6 to 8 year over year. (The number of GenAI apps tracked by the organization increased from 317 to more than 1,600.)
We spoke with IT leaders about an ideal approval path when an employee wants to try out a new AI tool.
The main steps. Thiago Ferreira, founder and CEO of Elevate AI Consulting and adjunct professor at the University of Miami, broke the process down:
Before you start…Ferreira recommends companies state their AI policy—guidelines regarding approved uses and data practices—and their list of AI tools that have already been approved.
“Most requests actually end right there once employees figure out, ‘Oh, I already have access to this tool, or I have access to a tool that does something similar,’” Ferreira told IT Brew.
Who fields the request? The AI lead decision-maker may vary, based on expertise and company size. Requests at small businesses frequently go to an IT admin, Ferreira said, while enterprise requests often go from employee to manager to the IT team.
Andy Boettcher, CIO at data and AI consultancy DoubleTrack, sees mature organizations using chief data officers and AI centers of excellence to get first look at a new AI request, and companies without established governance turning to chief technology officers or chief revenue officers.
The role of “AI decider” comes down to expertise. Shanker recommends that requests route to “the most knowledgeable team around AI under the CIO.”
What you need to know. A request, whether formal or informal, requires answers that boil down to data protection and value proposition.
Ferreira recommends leaders figure out:
Regarding value, Shanker wants to know: What is the use case, and what value would you lose if you don’t have it?
A new question has entered the mix recently, according to Boettcher: How will you measure success, and how are you planning on making sure that usage remains successful in six months?
Good answers to that last question, he said, could include regularly checking a baseline of questions for accuracy or defining a measurable business outcome. “Traditional systems never had to do that because they were like the old Ron Popeil food dehydrator. You set it and forget it,” he said. “AI drifts. That’s what it does. It actually does that on purpose because of new inputs, new technologies, new engines, new everything.”
The review. Akeel Mohamed, senior product manager in the business and enablement department at global real estate investment management advisor BGO, fields automated workflow requests that often combine existing, approved AI tools along with new products: A requested agentic HR process, for example, might extract data from files placed in SharePoint. For each request, he reviews cost, value, data security, and how a business process change might impact the workforce structure.
A tool may require a separate security audit, he added, and business must agree with tech teams on the need.
“I think a proper approval process includes a strong business case as well as strong foundational IT security and IT governance,” he said.
Who decides? For Shanker, the CIO gets the final decision and the security team has a powerful veto. Boettcher sees the C-suite setting up overall governance structures, while leaving the final decision to a resident expert on the IT team, someone who’s educating themselves, and ensuring the business adheres to success metrics.
How long should it take? There is a fast-path option for some tools—for example, if a well-known vendor already used by the company has a new AI feature. Nonetheless, it’s always important to check what data that new feature can access, Ferreira said. A low-risk lane could include tools that don’t access sensitive data or connect directly to company systems or areas like SharePoint or OneDrive, he added.
When a vendor has detailed documentation regarding security and data processing, Mohamed said, his company’s security audits can take a week or two. (Be prepared for longer, depending on a vendor’s responsiveness, he also warned.) Ferreira’s rule of thumb: low-risk requests should be decided within a couple of business days—and full reviews within two weeks.
The IT pros who spoke with IT Brew all emphasized speed. Shanker likes to at least have a status update within 48 hours—a first response, not necessarily a final decision: “You want to stop anyone from going and saying, ‘Look, they’re going to take forever. I’m going to go and do this myself.’”