AI Process Automation: The Real Cost-Savings Potential, From 20% to 45%

Every vendor claims AI process automation slashes costs, but the numbers thrown around online rarely trace back to a real study. When a claim can’t be sourced, it shouldn’t be repeated, so before making any promises to our own customers, we went looking for what credible research firms have actually measured. The picture that emerged is more useful than a single flat percentage: it turns out the size of the savings depends heavily on which process you automate, and understanding that difference is the key to building a realistic AI automation business case.

What the research actually shows

McKinsey’s widely cited analysis, “The economic potential of generative AI: The next productivity frontier,” breaks cost and productivity impact down by business function rather than offering one universal number, and the range is wide. Customer operations could see productivity gains worth 30 to 45 percent of current function costs, software engineering could see 20 to 45 percent of current annual spending, R&D could capture 10 to 15 percent of overall costs, marketing could unlock 5 to 15 percent of total spend, and sales could gain 3 to 5 percent of current global sales expenditures. That spread, roughly 3 percent at the low end and 45 percent at the high end, is a far more honest picture than any single figure, and it explains why a claim like “20 to 37 percent across the board” doesn’t hold up: the real savings cluster tightly around specific process types, not around business automation in general.

Customer operations: the biggest win

Customer-facing workflows consistently show the largest documented gains, and McKinsey’s 30 to 45 percent range for customer operations sits at the top of the scale for good reason. These processes are high-volume, repetitive, and text- or conversation-heavy, which is exactly the profile that large language models handle well: ticket triage, first-response drafting, knowledge-base retrieval, and routine query resolution can all be automated with a human reviewing edge cases rather than every single interaction. The cost reduction here comes less from replacing headcount outright and more from letting existing support and success teams handle a much larger volume without proportional staffing growth, which is precisely the kind of compounding return that makes a marketing-automation and customer-engagement platform valuable over multiple quarters.

Software engineering and IT workflows

Software and IT teams show the second-largest range, with McKinsey estimating 20 to 45 percent of current annual spending recoverable through AI-assisted coding, testing, and system design. This isn’t theoretical: a study McKinsey cites on GitHub Copilot found developers completed a coding task 56 percent faster with AI assistance than without it. For B2B companies running their own marketing or product engineering stacks, this matters directly, since much of the cost of building and maintaining automation workflows, integrations, and internal tools falls into this same category. It’s also the clearest evidence that AI process automation’s savings come from compressing the time skilled people spend on repetitive sub-tasks, not from eliminating the need for expertise.

Marketing, sales, and R&D: real but more modest gains

Marketing and sales functions show meaningfully smaller ranges in the same McKinsey analysis: 5 to 15 percent of total marketing spend and 3 to 5 percent of global sales expenditures, respectively, with R&D landing at 10 to 15 percent of overall costs. That’s still real money at enterprise scale, but it’s a fraction of what customer operations or engineering can capture, largely because marketing and sales work involves more judgment calls, brand nuance, and relationship management that AI still needs a human to steer. This is a useful corrective for teams evaluating marketing automation platforms: the honest expectation for lead scoring, campaign sequencing, and content workflows is efficiency gains in the high single digits to mid-teens, not the 30-45 percent range that gets attached to customer support.

Why the range varies so widely

The variance across functions lines up with broader market signals about how enterprises are actually approaching AI cost strategy. Gartner’s 2025 survey of infrastructure and operations leaders found that 54 percent named cost optimization as their top goal for adopting AI, which shows intent is high even where achieved results are still being measured and reported unevenly across companies. IDC’s research on enterprise AI investment has separately found organizations reporting roughly $3.5 to $3.7 in returned value for every dollar spent on AI initiatives, a figure that captures the combined effect of cost avoidance, productivity, and revenue impact rather than cost savings alone. Together, these data points support the same conclusion the function-level breakdown suggests: the processes that are high-volume, rules-based, and language-heavy see the biggest, fastest-verified savings, while judgment-heavy work sees smaller, slower-to-prove gains.

The practical takeaway for any B2B team building an automation roadmap is to sequence by process type rather than chase one headline percentage: start with the customer-facing and engineering workflows where the evidence is strongest, then extend into marketing and sales once the operational foundation is proven. That’s the approach we build into the Unomage platform, targeting automation where the research shows it actually pays off rather than automating for its own sake. If you want help mapping which of your own processes fit the high end of that range, our team in Warsaw offers a free consultation, and you can explore how the platform handles this at platform.unomage.com.

This article was created with the help of the Unomage AI platform.