The Richmond Fed and Duke University surveyed 548 business owners and finance leaders in December 2025 on their AI investment plans. About 80% of small firms plan to invest in AI in 2026, up from 48% in 2025. That's a significant acceleration.
The same survey found that AI investment is not expected to produce measurable cost savings in 2026.
That's the part most vendors won't tell you. Most AI implementations take 12 to 18 months to generate the efficiency gains that show up in a financial statement. Year one is setup, learning, and process adjustment. Year two is where the returns accumulate.
This matters for how you structure your investment, what you measure in the first 12 months, and how you report AI ROI to stakeholders who are expecting a P&L impact.
Why CFOs Are Struggling With AI ROI
Gartner's 2026 CFO research found that only 36% of finance chiefs feel confident they can drive AI impact. Less than half, 44%, believe they can accelerate adoption even within the finance function. The technology isn't the problem. The problem is accountability and measurement.
Most AI projects fail the ROI test not because they don't deliver value, but because the value they deliver isn't captured in the metrics the CFO is watching. Productivity gains are real but diffuse. Decision speed improvements are hard to attribute to a single tool. Customer satisfaction lift is a lagging indicator that takes months to appear in retention data.
Deloitte's Q4 2025 CFO Signals survey found that 87% of CFOs predict AI will be extremely or very important to finance operations in 2026. The confidence in the category is high. The confidence in measuring the return is not.
What Year One Actually Looks Like
McKinsey's 2025 research found that 92% of companies planned to increase generative AI investment over the next three years, but only 1% considered their AI investments fully mature. Almost everyone is early. The businesses that outperform in years two and three are the ones that get the foundations right in year one.
Here's what realistic year-one outcomes look like for different types of AI investment:
Process automation (document processing, data entry, invoice handling): ROI timeline of 3 to 6 months. This is the fastest-return category because you're replacing a manual, time-measurable task with an automated one. The time saving is quantifiable from day one. A business processing 500 invoices per month at 4 minutes per invoice saves 33 hours per month when automation handles it in 30 seconds each. At $50/hour fully loaded cost, that's $1,650/month, $19,800/year, from a tool that typically costs $100 to $300/month.
Customer service automation (chatbots, automated responses, triage): ROI timeline of 3 to 9 months. The saving is in contact volume handled without staff. Businesses that can resolve 30 to 40% of customer queries without human intervention reduce response times and free staff for complex interactions. The ROI shows up in customer satisfaction scores and staff time metrics before it shows up in headcount.
Analytics and reporting automation: ROI timeline of 6 to 12 months. The first phase is getting dashboards and reports running automatically. The second phase, which takes longer, is actually changing decisions based on the data those reports surface. The return on analytics investment is usually decision quality, not time saving, which is harder to attribute but often more valuable.
AI assistants for knowledge workers: ROI timeline of 6 to 18 months. This is the most variable category. Productivity gains are real but depend heavily on how consistently the tool is used. GitHub Copilot data shows developers using AI coding assistants complete tasks 55% faster on average. Similar productivity lift is plausible for knowledge workers using AI writing and research tools, but actual uptake in a 50-person SMB is typically 20 to 40% of the workforce initially.
The Right Metrics for Year One
If you're not going to see meaningful cost savings in year one, what should you be measuring? Here's a practical framework:
At 3 months:
- Is the tool being used? Track active users and frequency of use. A tool nobody uses is a failed investment regardless of its potential.
- What is the time per task for the automated process vs the manual baseline? This is your future ROI foundation. Measure it now, before the manual baseline is forgotten.
- Are there integration problems slowing adoption? Most early adoption issues are technical friction, not resistance.
At 6 months:
- Has error rate changed on automated tasks? Automation should reduce errors in data entry and processing tasks. Measure this.
- Are staff spending time on different tasks now, or doing the same tasks faster? "Faster" is ROI. "Different" is capability expansion, which is longer-term ROI.
- What do users say about the tool? Qualitative feedback at 6 months often surfaces the limitations that will constrain ROI if not addressed.
At 12 months:
- What is the cumulative time saving in hours? Convert to dollar value using fully loaded staff cost.
- Has the tool enabled any business outcomes that were previously impractical? New service capabilities, expanded capacity, improved response times?
- What did it actually cost, including implementation, training, and the time your team spent adapting processes?
The honest ROI calculation at 12 months for most SMBs will show positive returns on process automation and mixed returns on more speculative AI initiatives. That's normal. The compounding comes in years two and three.
Where to Start: The Anti-Moonshot Principle
The pattern we see most often in failed AI projects is starting with the most ambitious use case. Building a custom AI model. Deploying an autonomous agent across the whole business. Attempting to replace a core system with AI before the existing system is even well-understood.
The businesses that get good ROI start with process automation on one well-defined, measurable task. The criteria:
- The task is currently done manually, by a person, taking measurable time.
- The task is repetitive and rule-based enough for automation to handle reliably.
- The output quality of the automated task is measurable (error rate, completion time, etc.).
- The task is not customer-facing initially. Start with internal processes where errors are catchable before they cause damage.
For most SMBs, this means starting with one of: invoice processing, data extraction from documents, scheduling and calendar management, or internal report generation.
Nail one of those. Measure the ROI. Use that success to build internal confidence and a business case for the next project.
The Accountability Problem Gartner Identified
Gartner's finding that less than half of CFOs believe they can accelerate AI adoption even within their own department points to a structural problem: nobody owns the AI investment.
In a large enterprise, there's a Chief AI Officer or at minimum a dedicated AI programme team. In a 50-person SMB, AI adoption usually falls to whoever championed the first tool, which is often someone in operations or IT who doesn't have a budget line for it.
The fix is simple: assign clear ownership before you start. Pick one person responsible for the AI programme. That person doesn't need to be technical. They need to be able to run the measurement framework, communicate progress to the CFO, and make decisions about what to automate next.
A Realistic Budget Framework
For year one AI investment at an SMB level, a practical allocation:
- Process automation tools: $200 to $600/month for 2 to 3 well-chosen tools
- AI productivity tools for staff: $20 to $50 per user per month (not everyone needs them in year one)
- Implementation and integration: $5,000 to $20,000 depending on complexity. Don't skip this. Tools that aren't properly integrated into existing workflows don't get used.
- Training: $2,000 to $5,000 for structured staff training in year one
- Total year one: $15,000 to $50,000 for a 20 to 50 person business
At that investment level, break-even in year one is achievable for process automation, unlikely for more ambitious initiatives. That's fine, provided you're building the foundations that make years two and three significantly more productive.
Over 50% of finance leaders are planning higher spending on IT and AI in 2026, according to Gartner. The businesses that invest thoughtfully in year one, measure honestly, and build capability incrementally will be materially ahead by 2027. The ones that buy expensive tools without clear use cases and measurement frameworks will cancel them at renewal.
