BOTTOM LINE UPFRONT
In a recent conversation with Forbes, Accordion’s Justin D’Onofrio, Managing Director in the firm’s CFO Tech practice, tackled a problem keeping finance leaders up at night: building an AI budget that won’t blow up mid-year. The guidance? Stop budgeting around a tool, and start budgeting around the workflow you’re trying to improve—with clear ownership and centralized tracking from day one.
Off the ledger:
How to make a realistic AI budget for 2027
AI can do incredible things in the enterprise—including driving incredibly high usage bills. Accordion published a report to help CFOs get a handle on AI costs and create a realistic AI budget for 2027 that will fund what your enterprise needs and shouldn’t be unexpectedly depleted before summer. I talked to Justin D’Onofrio, managing director and planning solutions team lead at Accordion’s CFO Tech practice about how to do it.
Why were so many CFOs caught off-guard this year with AI bills that were unexpectedly high?
D’Onofrio: Primarily, it’s a new muscle for people in a new type of spend category. I was with a SVP of finance about a month back, and they showed me their Claude bill. It was an invoice. There’s literally one line on the invoice: multiple hundreds of thousand dollars for the month. His question was: What’s going into this? That’s a senior finance person that doesn’t know you have to dig one level deeper to understand all of the different users attributing to that total spend, all of the engineering API keys that are going to equate to that spend.
It’s a new muscle where there’s a level of granularity finance teams now need to understand and surface, that they haven’t done in the past. They’re used to getting vendor invoices that are relatively streamlined—month-over-month growth, year-over-year growth. These have the potential to balloon cost-wise. The costs can change quickly and the invoice doesn’t share all the details, which is where finance teams are typically looking to start.
The other big change is—and this is true of all of our clients—the smaller but growing, mid-market PE-backed clients, and the enterprise clients that are publicly traded, a lot of them don’t know where to spend. And because they don’t know where to spend, it’s a lot of siloed business cases or projects that are not tracked or broken out centrally. In those cases, budgeting cohesively across the entire organization becomes a very big challenge.
You just said CFOs may not have the muscle they need to track and pay attention to AI use in their company. How can they start exercising that muscle?
D’Onofrio: The first thing we typically recommend is generally benchmarking their sector. What are good companies in their industry doing, and where are they spending? This is changing very rapidly, and you want to understand where peers are seeing value.
Once you’ve created a directional strategy on where you want to invest, when you create those business cases for an investment, [you need to make sure] you’re doing a few things. One is tagging those investments to some sort of value lever. Do I think this is going to increase revenue? Improve our internal efficiency? Is it something we have to do to stay competitive? Making sure there’s some business case around tagging it to an EBITDA or value lever.
Then, make sure that someone owns that spend and is accountable for it. That person doesn’t have to sit in finance, but these things need to be tracked constantly to make sure costs aren’t ballooning and you’re capturing the benefits of the AI program. Tracking is an ongoing exercise.
AI is always developing, and it seems potentially groundbreaking applications are launching daily. How can a CFO put together a budget that takes into consideration a potential new AI application that launches midway through the fiscal year?
D’Onofrio: It’s not easy. One of the issues we see is that folks will budget for a tool or AI capability. ‘Hey, we’re going to spend on Claude.’ That’s the wrong framework to think about budgeting for AI.
Instead, it should be: Here’s a workflow or a need we have internally at the company. We have an assumption we can improve that. It’s time savings, cost savings, revenue increase drivers. We’ll fund that workflow, and as AI improves, we can start to capture more benefit with the new features and functions.
But the business case is built around the reframing of the workflow leveraging AI, not funding a tool. You can’t build a business case around features and functions.
What advice would you give a CFO working on making their AI budget for next year?
D’Onofrio:
- Get a central place to track your AI spend. Most clients that we work with cannot easily pull out the AI spend that’s intertwined within their non-AI spend—cap ex, op ex. That’s where you’re going to see a non-traditional ballooning of costs.
- Within that spend, be able to track it well. Start to dig into that second level of granularity that’s not just the vendor invoice. As you track it, make sure you have that accountability associated with the tracking. Who’s the owner? And is the tracking giving you the benefits that you expected?
What should a CFO absolutely not do as part of this process?
D’Onofrio: Funding the tool is a common mistake. Saying we’re going to fund Claude or OpenAI usage is not really thinking about the benefits, the business case and the financial view. It’s just a platform. It’s an easy fix if you think about reframing the budgeting process.
Strategies + advice
D’Onofrio: Everyone wants to present plans that work all the time, but that’s unrealistic. Sometimes, the numbers are off, and someone needs to be “in charge” of closing the gap. Here’s how to strengthen accountability to assign responsibility, and fix inaccuracies before things go even further in the wrong direction.
Four forces in today’s workplace are combining to create a “tornado” for leaders, creating issues and expectations that even the most seasoned executives may feel like they can’t handle. Traditional leadership development training doesn’t prepare for this confluence of problems, but here are four ways to deal with it.