How an IT Director Saved Her Nonprofit $25,000 and Got Its AI Policy Approved

Where we started

My client runs IT for a small DC-based privacy nonprofit. She calls herself a “department of one,” responsible for vendor management, the help desk, and everything else that touches a plug or a password.

She was already a confident daily user. When an expired certificate broke Salesforce sign-in at 3 a.m., she pasted screenshots into Claude and fixed it herself instead of spending hours digging through online support forums.

Over three coaching sessions, with messages and follow-up materials in between, we worked through two problems together.

• The outside contractors she relied on for Salesforce kept delivering unreliable code. Leadership asked her about letting the contractors go, but she worried about losing coding support she might still need.
• She had drafted an updated AI policy, but approval had stalled after about a year of committee discussions and revisions. Staff needed clearer guidance on using AI, while leadership wanted reassurance that the tools wouldn’t expose restricted information.

Taking the work in-house

She already knew Claude could help. But could she rely on it enough to manage the coding herself?

What helped her get there

• Taking stock of the contractor relationship. I encouraged her to track recurring problems with the contractors’ work. By our second session, she wondered whether she could write the Salesforce code herself rather than spend so much time directing them.
• Seeing a clear alternative. We explored how much of the coding she could handle with Claude, which she described as “almost as good as my consultants.” Explaining tasks to contractors was taking too much time: “I could be doing this instead of explaining this.”
• Getting more from Claude. We reviewed her settings, and I encouraged her to default to the strongest available model for coding and other work.

By our third session, she had let the contractors go and was using Claude to handle the Salesforce coding herself.

When I asked her to check the IT budget, she found that the organization had paid the contractors $25,000 the previous year.

“Oh, that’s going on my yearly review!”

Getting the AI policy across the finish line

I made helping her get the policy approved a central goal of our work. We focused on showing why staff needed clearer rules and addressing leadership’s security concerns.

What helped her get there

• Showing how unclear rules held staff back. I asked her for examples of why the update mattered. Leadership encouraged AI experimentation, but the policy hadn’t caught up. Some staff missed out on AI’s benefits for fear of breaking the rules. Others used it without clear limits on what information they could share.
• Addressing security concerns with practical tests. She used Claude to write prompts that tested whether AI could reveal files users weren’t authorized to see. She provided instructions and a spreadsheet to record results. She and the CEO ran the tests themselves, giving him confidence to proceed with the policy rollout.
• Stronger sign-in protections. We reviewed the organization’s Claude setup, and I recommended requiring staff to sign in through their work Google accounts with multi-factor authentication.

About two weeks after our third session, she secured leadership’s approval for the updated AI policy.

Staff now have clear guidance on when AI is allowed, when it’s off limits, and who to ask if they’re unsure.