agenticAI. It's ok!: Objection Handling Guide

A practical guide to anticipating and addressing common objections & concerns when implementing AI projects within the APS. This resource is designed to support project teams in navigating stakeholder discussions effectively.

We can’t use OpenAI, it’s not on our approved vendor list.

We don’t allow apps that read mailbox or calendar data.

Who approved this project or pilot?

This looks like shadow IT.

Will this expose data inappropriately?

This will just create more work for my team. We don't have time for this.

The AI will be biased and make unfair decisions.

This is just a shiny toy; it doesn't align with our strategic priorities.

We don't have the budget for this.

Our staff don't have the skills to use this technology.

How can we ensure accountability? Who is responsible if the AI makes a mistake?

This will be another failed IT project that doesn't deliver.

Our data is too fragmented and low-quality for AI to be useful.

This introduces vendor lock-in.

The AI won't understand our specific context, jargon, and acronyms.

We have strict privacy and data sovereignty requirements that this won't meet.

This will slow us down during the initial learning phase.

My team's work is too nuanced and creative for an AI to handle.

How do we manage version control and changes if an AI is generating content?

What happens if the external AI service (e.g., Azure OpenAI) has an outage?

This technology is changing so fast. Won't this be obsolete in a year?

This doesn't integrate with our key line-of-business application.

The output from generative AI can be confidently wrong ('hallucinations'). How can we trust it?

Our union (CPSU) will object to this as augmentation that threatens jobs.

How do we handle intellectual property for AI-generated content?

This is just for knowledge workers. It doesn't help our frontline or operational staff.

We tried something like this a few years ago and it failed.

The AI's responses are too generic and lack the specific tone of our department.

What's the carbon footprint of running these large AI models? How does this align with our sustainability goals?

This is a solution looking for a problem.

The procurement process for this will take forever.

How do we stop staff from putting classified or PROTECTED information into the AI?

We already have M365 Copilot. Why do we need another AI tool?

This looks great in a demo, but it will never work with the complexity of our real-world data and processes.

The results are not reproducible. I get a slightly different answer every time.

Who in our IT team is going to support this when it breaks?

This undermines the professional judgment of my expert staff.

The outputs are not formatted correctly for our official templates.

We have to keep records for 7 years. How is this system's data retained and archived?

This will never get past our architecture review board.

What if staff become over-reliant on the AI and lose their critical thinking skills?

The interface is too cluttered and confusing.

This system doesn't provide a full audit trail of who did what and when.

Change management for this will be a nightmare.

This won't work for our staff in regional offices with lower bandwidth.

The AI's knowledge cut-off is too old. It doesn't have current information.

We can't get a clear answer on the total cost of ownership (TCO).

This process requires human empathy and intuition, which an AI doesn't have.

How can we ensure this is compliant with accessibility standards (WCAG)?

This is a single point of failure. What if the whole platform goes down?