On this page
When we help clients build the case for agentic AI transformation, we account for two distinct types of return. Most companies see only the first. That gap is the difference between a narrow cost-optimisation project and a transformation that compounds, and it is why the agentic AI business case so often understates what the investment is worth. Getting agentic AI ROI right starts with naming both returns before scope is fixed.
Most agentic AI business cases measure only one kind of return #
Today most organisations evaluate agentic transformation, and their artificial intelligence investment more broadly, with the same model they apply to any technology purchase: a spreadsheet of hard cost savings. Hours removed, cost per transaction, headcount reallocated. It is the model the finance team trusts, and it is the model the CFO signs off.
That model is built to see one kind of value, so it misses the rest. The consequence is visible in the aggregate figures. MIT's State of AI in Business 2025 found that 95% of generative AI pilots delivered no measurable impact on profit and loss, while more than half of budgets went to sales and marketing even though back-office automation produced the highest returns ( Fortune, on the MIT NANDA report ). Across these AI deployments, the problem is rarely the technology. It is that companies measure the wrong return in the wrong place.
Type 1: economic ROI, the return that fits in a spreadsheet #
Type 1 is the familiar territory: directly quantifiable, and straightforward to defend in a budget review. It is the return that appears when agents take over routine tasks. It shows up as hours saved, cost per transaction reduced, headcount reallocated, and throughput gained. An agent that handles 400 invoice matches a day that previously required 2.5 full-time employees is Type 1. So is an average resolution time that falls from 14 minutes to 90 seconds.
We at Vstorm see this return in production. Before we automated device activation for a US telecommunications provider, every field installation depended on a manual, call-centre process: a technician phoned a support centre, where three agents activated the device in real time across multiple systems. The manual workflow carried a high labour cost per installation, limited service to office hours, and blocked expansion into new states. The multi-agent system we built reduces manual effort to a fraction: it achieved 98% automation of device activation and a tenfold efficiency improvement, and it removed the time-of-day limit entirely ( case study ).
Type 2: non-economic ROI, the return that compounds #
Type 2 is where the compounding value lives, and it is harder to put a number on. It shows up as faster decisions, because information flows without bottlenecks and agents improve decision quality; as higher employee satisfaction, because people stop doing repetitive, low-value work; as institutional knowledge captured in agent logic rather than held in one person's head; and as organisational agility, the ability to respond to market shifts in days rather than quarters. These outcomes reach the profit and loss eventually, but the causal chain is indirect and the timeline is longer.
The data shows how easily this return is missed. In McKinsey's State of AI in 2025, 64% of organisations said AI was already enabling their innovation, yet only 39% reported enterprise-wide earnings impact ( McKinsey ). McKinsey also notes that the productivity gains from broad AI tools, such as copilots and AI assistants, tend to be distributed thinly across employees, which makes them hard to see in headline financial results.
This is the return we design for alongside the economic one. When we cut engineers' most tedious tasks to seconds for an engineering-automation platform, the measurable time saved was only part of the value; the larger return was freeing skilled engineers for work that machines cannot do ( case study ). The multi-channel pre-appointment agent we built in healthcare changed the experience for patients and care teams, a return that a per-transaction figure does not capture ( case study ).
| Dimension | Type 1: economic ROI | Type 2: non-economic ROI |
| Nature | Directly quantifiable | Real, but hard to quantify |
| Typical measures | Hours saved, cost per transaction, headcount reallocated, throughput | Decision speed, employee satisfaction, captured knowledge, organisational agility |
| Timeline | Immediate, visible within the year | Longer, compounds over time |
| Where it appears | In the spreadsheet; signed off by finance | Reaches profit and loss indirectly |
| Risk if overlooked | Low, because it is always counted | Under-scoping; treating transformation as a cost play |
Why a partial agentic AI ROI picture leads to underinvestment #
A business case that counts only Type 1 will consistently understate the return, and that has a predictable effect on behaviour. Clients who evaluate agentic AI transformation through economic ROI alone tend to underinvest, scope too narrowly, and deploy agents only where the spreadsheet already justifies the cost. The work becomes a cost-optimisation play rather than a change to how the organisation runs its core business processes.
The pattern is measurable. McKinsey reports that 80% of companies set efficiency as their AI objective, but the organisations seeing the most value set goals around growth and innovation instead ( McKinsey ). Separately, McKinsey found that 94% of companies were not yet seeing significant value from their AI investments, an echo of the old observation that a new technology can appear everywhere except in the productivity figures ( McKinsey ). A return that is never named is a return that never gets funded.
Making both returns visible in the Strategizing phase #
Our task in the Strategizing phase, the first stage of our TriStorm methodology, is to make both returns visible before scope is set ( TriStorm ). We quantify the economic return for each high-impact use case, and we name the non-economic return explicitly, so that decision speed, employee experience, retained knowledge, and agility enter the business case as stated outcomes rather than pleasant side effects.
This discipline shapes every engagement. Whether we are deploying AI agents in the complex workflows of finance, customer service, fraud detection, or the supply chain, the economic return is only ever half the picture. When AI powers a business process end to end, the compounding return is often larger than the savings that first justified it.
A business case built on economic ROI alone will always under-scope the agentic AI systems worth building. Counting both returns is how a cost project becomes a durable, long-term advantage. That is the work we do before the building begins.
