What CFOs should consider before rolling out internal AI tools

The AI gold rush has pushed CFOs into an even more complex leadership position than before. In addition to evaluating the internal tech’s ROI, CFOs can expect to be called on to lead internal AI rollouts, including pilot programming, change management, and data accuracy.

In a survey from Deloitte, CFOs reported feeling pressure from boards to deploy AI quickly, but to be able to give them realistic timelines, a CFO needs a few other things first. Mitchell Nakken, Accordion managing director and head of strategic finance, said his first question when someone asks him about AI is “How strong is your data environment?”

“If you started with really great data and a ready management team, you could be up and running in about a month,” Nakken explained. A successful AI deployment “hinges on the ability to have very clean data,” he added.

Try before you buy. Before a full rollout timeline is considered, Tom Derhake, co-founder and partner at business tech consultancy Cuesta Partners, emphasized the need for a pilot period where employees get semi-access to the internal model to prove the concept.

“This is not just buying the Claude or ChatGPT licenses, but this is setting up the security, ontologies, and connections across the necessary foundation to ensure the AI tooling shows compounding value, securely across the user community,” Derhake said in an email.

But getting to that point with data can mean pushing beyond simply “turning on Copilot,” Nakken said in an email, and into work like “data integration, tool use, and agent workflows where ROI is realized.”

He also said that in his experience, eight to 14 weeks is the average build- out timeline for an internal AI system, from ideation to release.

“We can start building context if we can get it to this level, that context window can be built alongside the data cleansing effort, so you’re not necessarily losing any time there,” he added. (Building context into AI means investing into the systems that connect and govern the internal AI prompts and processes.)

Go with the flow. While there isn’t necessarily a checkered flag at the end of an AI rollout encouraging speed over substance, Derhake recognized there can be a decent bit of FOMO fluttering around. Balancing the speed of a rollout with the human ingenuity and change management considerations necessary helps ensure a successful rollout, he added.

“CFOs also need to own the ‘I’ in ROI, which now means understanding and estimating usage-based costs, like token consumption, that grow with adoption. In the end, the goal isn’t to report on AI as a side project. It’s to weave AI into the metrics the CFO already owns, so it shows up in margin, growth, and cash flow,” Derhake said via email.

Nakken advised being as well-versed in the technology as possible to help counteract the many voices in their ears asking how the AI development is going.

“Sit down and do the work with the team. Our best customers are users of the tools and know when they are ‘blocked’—those leaders are actively redefining the workflows and skills in their functions,” Nakken said in an email.