In Issues 23 and 25, I wrote about outcome based contracting and the measurement discipline it requires from GBS. The attribution gap, the baseline problem, and the governance work that goes into verifying what a platform actually delivered versus everything else happening in the business at the same time. Those issues were built on an assumption that most current GBS technology investments operates within a defined function and delivers a measurable result within that function.
AI orchestration is beginning to change that assumption by not not just just adding a new layer to the ROI calculation but changing the underlying logic the calculation depends on.
Issue 26 is about the ROI on this new way of tying together the work.
AI Orchestration in a GBS Context
In a GBS environment, AI orchestration means multiple AI agents coordinating work across functions using shared data and shared context rather than operating independently within a single process. The O2C agent managing collections prioritization shares cash position data with the P2P agent optimizing payment timing, so a stronger than expected collections run can trigger early payment discount capture that neither agent could identify on its own. Dispute patterns identified by the O2C agent flow back through the orchestration layer to inform upstream contracting and purchasing behavior rather than staying trapped inside the collections workflow. Accrual anomalies the R2R agent flags during the close cycle inform the O2C agent’s collections prioritization for the same customer or product line rather than being handled as a separate and unrelated problem. Working capital forecasting spans all three functions simultaneously rather than being assembled manually from three separate function reports after the fact.
The orchestration layer is not passing generic signals between the agents. It is passing specific data that changes the decisions each agent makes in ways that produce outcomes no single agent, operating within its own functional lane, could produce on its own.
According to Peakflo’s May 2026 analysis of AI agent orchestration in finance, coordinated multi-agent systems deliver 60 to 85 percent less manual work, 40 to 67 percent fewer errors, and 25 to 50 percent faster close cycles compared to sequential single-function automation. According to Deloitte’s 2026 AI Agent Orchestration research, the orchestration market will grow from 8.5 billion dollars to 45 billion dollars by 2030 as enterprises move from single-purpose AI tools to coordinated multi-agent systems.
EY’s April 2026 research on AI in GBS noted, AI reveals the interdependence of end-to-end workflows more clearly than any tool before it, positioning GBS as the orchestrator of an interconnected enterprise rather than a manager of separate process towers.”
The shift from individual AI tools to an orchestrated AI system Should be something more than an incremental upgrade. If the coordination between agents produces outcomes that no single tool could produce independently, the way GBS measures and reports the value of its technology investments should change alongside it.
The Current ROI Framework
The ROI framework most functions applies to technology investments, the one Issues 23 and 25 described in detail, was built for a world where each tool operates in its own lane. DSO improvement gets attributed to the O2C platform. DPO improvement gets attributed to the P2P platform. Days to close improvement gets attributed to the R2R platform. Each calculation is independent and the attribution methodology, however imperfect, at least has a clear starting point.
AI orchestration removes beginning and end points that we’re used to and creates an entirely new way of realizing value.
When an orchestration layer is coordinating cash application timing in O2C with payment term optimization in P2P and accrual accuracy in R2R simultaneously, the working capital improvement that can result is a product of all three agents acting together, not the output of any single platform. The attribution methodology that worked when the tools were independent does not tell you how to split that improvement across three systems making coordinated decisions.
This is not a hypothetical problem as noted in many studies. According to IBM’s Q4 2025 Think Circle, only 29 percent of executives say they can measure ROI from AI confidently today, while 79 percent report seeing productivity gains. The gap between seeing the value and being able to measure it is exactly the gap that orchestration widens, because the value is increasingly produced by the interaction between systems rather than by any individual system on its own.
BCG’s survey of over 280 finance executives, published in early 2025, found that median reported ROI from AI in finance was just 10 percent, well below the 20 percent many were targeting, with nearly a third of finance leaders reporting only limited gains. BCG’s more recent AI Radar 2026 research, drawing on 2,360 executives across 16 markets, suggests the pattern is holding. Four out of five CEOs are more optimistic about AI ROI than they were a year ago, yet 94 percent of organizations plan to continue investing even if current initiatives do not pay off in the next year, which implies the measurement problem has not been solved even as confidence and investment both rise. BCG’s 2025 finance analysis found that the teams generating strong ROI were taking a broad transformation view rather than focusing on single use cases. That finding points directly at the orchestration question. The value appears to be in the coordination across tools rather than in any individual tool on its own.
What Changes in the Orchestration ROI Calculation
Three specific things change when AI orchestration enters the picture, and each one requires a different approach than the methodology Issues 23 and 25 described.
The attribution unit changes. In a single-platform evaluation, the attribution unit is the platform. Did this O2C tool improve DSO? In an orchestrated environment, the attribution unit is the outcome, not the tool. Working capital improved by a specific dollar amount. The question is not which platform caused it but what combination of coordinated decisions across O2C, P2P, and R2R produced it, and whether that combination is reproducible and improvable.
The baseline becomes a system baseline rather than a function baseline. In Issue 25, the baseline work I described was function-specific. Build a baseline for DSO before the O2C platform goes live. Build a baseline for first time match rate before the P2P platform goes live. In an orchestrated environment, the relevant baseline is the system’s overall performance across the enterprise outcome the orchestration layer is optimizing for. In an orchestrated environment, the relevant baseline is the system’s overall performance across the enterprise outcome the orchestration layer is optimizing for, such as working capital position, revenue cycle velocity, or close cycle time. These are enterprise level measures that require a different kind of measurement infrastructure to establish and maintain than the function-specific baselines described in Issue 25.
The governance model has to operate at the system level rather than the vendor level. In a traditional multi-vendor GBS technology stack, governance is organized around individual vendor relationships. Each vendor is measured against its own contract, its own SLAs, its own outcome commitments. When an orchestration layer is coordinating the activity of multiple vendors’ agents simultaneously, vendor-level governance no longer captures whether the system as a whole is delivering the enterprise outcome it was designed to produce.
What GBS Needs to Build
The measurement infrastructure for an orchestrated AI environment has to be designed at a different level than the function-specific baseline and attribution methodology described in Issue 25.
Here is how I think of this in practice.
Define the enterprise outcomes the orchestration layer is supposed to optimize before any agents are deployed. Before any agents are deployed, the GBS function needs to define the enterprise financial outcomes the orchestration layer is supposed to optimize. Think of things like freed up working capital , revenue cycle days reduction, close cycle time improvement, and cost per unit of output across the combined process stack. Establish baselines, targets, and attribution methodology for each of them at the system level. The logic is the same as the function-level baseline discipline described in Issue 25, applied one level up
Build a system level measurement architecture that tracks outcomes across the full process chain rather than within individual functions. According to a joint study from MIT Sloan and Stanford GSB published in August 2025, AI cuts the monthly financial close by 7.5 days on average across the organizations studied. That is a system level outcome measure. Getting to that number required tracking close cycle performance across the full R2R process, not within any single tool’s reporting dashboard. GBS needs the equivalent measurement capability for every enterprise outcome the orchestration layer is supposed to produce.
Design governance around the orchestration layer itself, not just around the individual agents it coordinates. High Radius’ Algo Quotient concept I referenced in Issue 23, is about managing AI agents with the same performance accountability applied to human performers. It becomes more complex in an orchestrated environment because the performance in question is increasingly collective rather than individual. GBS needs a governance model that can evaluate whether the system of agents is producing the outcomes it was designed to produce, identify which agent or interaction between agents is underperforming when outcomes fall short, and make adjustments at the system level rather than managing each vendor relationship in isolation.
When negotiating with vendors whose platforms will operate as part of an orchestrated environment, the contract conversation needs to go beyond what that vendor’s individual tool is accountable for delivering. If an O2C platform, a P2P platform, and an R2R platform are all operating under an orchestration layer that coordinates their activity toward a shared enterprise outcome, each vendor’s contract should reflect their contribution to that shared outcome, not just the performance of their tool in isolation. That requires GBS to have defined the enterprise level outcomes and the system level attribution methodology before any individual vendor negotiation begins, so each contract is written against a consistent reference point rather than against the vendor’s own definition of success.
The Connection Back to Outcome Based Contracting (what we all should want)
AI orchestration makes outcome based contracting more compelling but also more demanding.
More compelling because outcomes become measurable at the enterprise level rather than within individual functions. When an orchestration layer is coordinating decisions across O2C, P2P, and R2R, the working capital impact of that coordination is visible in a way that individual function metrics cannot fully capture. That visibility is exactly what outcome based contracting requires to work as intended.
More demanding because the attribution question becomes harder to answer when multiple agents across multiple vendors are all contributing to the same result. The vendor-level attribution methodology from Issue 25 is a necessary but insufficient foundation for an orchestrated environment. GBS needs the system-level measurement architecture described above before outcome based contracting at the orchestration layer can be negotiated and governed with enough precision to be meaningful. This will be no easy task.
The sequence of this work matters more than any individual component of it. System level baselines and attribution methodology need to be in place before the orchestration layer is deployed, vendor contracts need to reflect system level outcome commitments before they are signed, and the governance model for the orchestration layer needs to be designed before the first agent goes into production. That order is consistently harder to maintain under the schedule pressure that accompanies any significant technology deployment, and it is consistently the order that determines whether the ROI number eventually reported to the CFO reflects what the orchestration layer actually produced across the enterprise or what each individual vendor calculated within their own reporting boundary.
Where This Leaves GBS
According to EY’s June 2026 research on GBS and agentic AI, the opportunity for GBS is to evolve beyond a service delivery model into an enterprise intelligence hub, orchestrating people, processes, and AI agents at scale. There is the value! The measurement infrastructure to verify whether it is happening, and to calculate the ROI on the investment required to make it happen, is the work that has to be built before the orchestration layer goes live rather than after the first board update requests a return figure.
GBS has the cross-process visibility to define what the right enterprise outcomes are and build the baselines to measure them. It has the governance experience to design the accountability framework that an orchestrated AI system requires. GBS has the most to gain from getting this right because it further clarifies the value.
Ian P. Thompson is a finance operations and GBS executive with over 30 years of experience. He writes the Clear Cycle Dispatch for finance operations and GBS leaders who want to understand how the work is actually changing. If someone forwarded this to you and you want to receive future issues, subscribe at https://lnkd.in/eMfv4FMv