Technology & Innovation

How to budget for AI adoption: Tips for CFOs

Accurate budgeting for artificial intelligence (AI) adoption can be tricky but it's possible. Read these tips from CFOs and other experts.

Budgeting for artificial intelligence (AI) adoption can be a tricky business.

Rapid technological evolution and vast numbers of emerging applications create deep uncertainties in cost-benefit analyses and return on investment (ROI) forecasts.

CFOs say it can feel like trying to hit a moving target while blindfolded.

Research into ROI from AI hasn’t always been helpful either, with results varying enormously, from 5% to 350%, according to two recent studies.

Lack of clarity around the realistic calculations involved is becoming a pressing concern for finance leaders, especially given the pressure to deliver AI innovations with 59% of companies planning to increase rollout or investment in AI.

Let’s delve into the challenges of budgeting for AI and tips for solving them from CFOs, tech heads, and other experts who’ve learnt from experience.

Here’s what we cover:

Sage Ai

Discover how you can use Sage Ai to make smarter, faster decisions, get a dynamic, real-time view of your business performance, and gain access to actionable insights.

Learn more about Sage Ai

Laser focus on business priorities

Start your budgeting process by aligning the AI project as clearly as possible with business goals such as efficiency, growth, and increased competitiveness.

This helps avoid spending on unnecessary bells and whistles.

Leo Smigel, founder of algorithmic trading firm Analyzing Alpha, has navigated the complexities of AI adoption and says: “I’ve seen unclear objectives lead to resources getting wasted. Not all AI delivers equal value or will impact your specific goals.

“So prioritise tools based on your expected ROI.”

Take small, agile steps…

Christoph Cemper, founder and CEO of prompt management platform AIPRM, says: “The field is evolving incredibly fast, so it’s hard to predict where things will be even a year from now.

“But in my experience, an exploratory, iterative approach is key. Take it step-by-step rather than trying to forecast and budget for every possible application from day one.

“Starting with a small pilot or proof-of-concept focused on a single use provides a low-risk way to gather data and build the business case for larger investments.”

As AI is so potentially disruptive, Christoph recommends benchmarking potential value against competitors who are already using it, rather than against your historical performance.

Leo recommends an agile approach that includes plans for regular check-ins and adjustments, and using cheap or free resources where possible to keep costs down.

“The open-source ecosystem has awesome free solutions, so use them to avoid licensing fees,” he says. “I remember one stock-trading AI project where we underestimated data needs and training.

“However, taking an agile approach, we corrected the course with open-source tools and got a high-value solution.”

Joe Scarboro, CFO at Replan Technology, says if you are starting to explore AI, the best way to start budgeting for it is to allocate a general fund for experimentation.

“If you embark on a longer cost/benefit exercise for each application, the potential ROI will likely be out of date by the time you finish,” he says.

“Instead, build a fundamental framework for AI experimentation, with a budget attached. Aim to prove how AI can benefit a process or department conceptually.

“Once you have that, you can see what a larger roll-out might look like, and get a better picture of overall benefits.”

…But build a longer-term view

Initially, your budget will likely focus on short-term goals. But don’t ignore AI’s longer-term transformative potential.

For example, how could this project:

  • Amplify human capabilities across the organisation
  • Open new revenue opportunities
  • Enhance competitiveness in future?

Alistair Brisbourne, head of technology at the Association of Chartered and Certified Accountants (ACCA), says: “To avoid opportunity costs, especially around efficiency gains, CFOs can take a long-term perspective.

“This might include a range of additional considerations, such as solution scalability, employee morale, performance management, and growth opportunities versus cost reduction.

“With the pace of development, it is important that any tools you implement can be adapted for future advancements and challenges.”

Branson Knowles, head of US digital banking at Top Mobile Banks, says: “I’ve grappled with forecasting costs and returns for AI initiatives time and again. Cracking the code is possible if you take a holistic approach.

“Once you’ve aligned investments with strategic priorities, you need a realistic cost-benefit analysis examining everything from solution costs to computing resources and employee training.

“Be pragmatic about immediate costs and returns, but keep a long-term lens on how AI can shape your future.

“Go beyond ROI projections to convey AI’s powerful transformational potential. It requires foresight as much as spreadsheet mastery.

“Once, when proposing an AI virtual assistant, the numbers alone weren’t an immediate slam dunk. However, I sold the idea by showing how it could revolutionise our customer experience and position us as true innovators.

“Years later, that AI assistant has delivered on its promise many times over.”

2 big challenges: Data and refinement

Factoring data management into your AI cost-benefit analysis is critical.

Data quality is essential for AI efficacy, so you’ll need to budget for cleaning, normalising, and cataloguing it ready for AI processing, and for continuous data maintenance and storage.

You’ll need to budget for refinements to the overall model too.

Steven Kibbel, financial planner and editor at investment website Day Tradingz, says: “I’ve grappled with the unique challenges of budgeting for AI projects.

“Early in my career, I underestimated needs like data labelling, and testing and refinement. This led to budget overruns.

“Thankfully, I’ve learned much since.

“Now, I recommend CFOs account for data management, testing, refinement, feedback, and reworks. Gathering, cleaning, and annotating datasets for AI model training requires massive resources that elongate timelines.

“Build generous budgets and set realistic expectations around making data operational.”

Steven also recommends allocating funds for user feedback loops and reworks over at least 12 to 24 months.

“Numerous iterations are typically required as AI systems interact with real users and conditions,” he says.

Soft costs are essential

It’s critical not to overlook ‘soft’ costs, such as training, change management and user adoption, in your budget as these can make or break the success of an AI rollout.

Christoph says he always advocates for building these costs into budgets explicitly, even if they don’t show up on traditional ROI spreadsheets.

“Failing to plan for enabling people and processes can derail even the most promising AI technology,” he says. “Soft costs are the secret sauce for long-term AI success, and with the right process, the payoff can be immense.”

Steven says user adoption, retraining personnel, and workflow adjustments create significant change management costs that are often underestimated.

“Proper change planning is paramount for ongoing budgets,” he says.

Joe adds that people will likely perceive new AI as a potential threat to their role. So if your innovation culture does not consider the human element, even the best budgeting processes and ROI analyses will fail.

Also, AI moves fast, but organisational change generally does not.

So Joe suggests phased rollouts to give people time to acclimatise, rather than reject the new solution outright – even though this can delay the benefits.

In addition, finding skilled AI engineers and data scientists capable of overseeing projects takes time and money. Competitive compensation is key for project success and budget predictability.

But factor in that skilled scientists and well-trained users will accelerate adoption.

Leo says: “It takes a village to do AI adoption right, so mix cross-functional skills – including areas like data, user experience, and compliance – and diverse perspectives in your team.”

Plan ongoing governance

Many companies don’t yet have adequate measures in place to handle the potential risks of AI, such as around accuracy, privacy, and copyright.

It’s essential for CFOs to understand these risks and allocate resources to address them.

Dedicate part of your budget to ongoing monitoring, assurance, and governance of AI systems to support compliance with current or future potential regulations, ethical standards, and alignment with organisational values.

Flexibility is key

Michael Dinich, founder of personal finance website Wealth of Geeks, says in his experience of budgeting for AI projects, the most important thing is to stay flexible.

“I’ve learned not to get too attached to projections because requirements and capabilities tend to shift faster than you expect,” he says.

“I try to make conservative initial cost and benefit estimates. Then I revisit those estimates often as we move through pilot projects and refine our approach.”

It’s also important to involve stakeholders from across the business in the budgeting process. Getting input from operations, IT, and users helps build certainty by spotting potential risks and unseen costs.

Plus it grows support for any changes to come.

“Even with the best analysis, there will be uncertainties when adopting new technologies like AI,” says Michael.

“But by taking a flexible, collaborative approach to budgeting, and focusing on enabling resources, I’ve been able to move forward strategically.”

Final thoughts on budgeting for AI

Setting accurate budgets for AI projects can feel like navigating a dense fog – the path seems to keep shifting as the technology evolves.

But the points covered in this article should help you move towards de-risking budgets and forecasting potential returns on investment.

While uncertainties remain, a thoughtful approach to data, skills, testing, governance and change management can help you make forecasting more reliable, so your business can reap the transformational benefits of AI technologies with minimal waste.