Coordination Edges Inventory: The First Practical Step
You don't need a new dashboard, new software, or extra budget to start measuring human-agent coordination costs. You need 3 real decisions, a blank sheet of paper, and 30 days. The inventory is the step zero.
90-Second Summary
Governing human-agent coordination in financial terms begins with the same sequence CFOs used to govern cloud spend in 2017: a paper-based inventory prior to instrumentation. In 30 days, five steps cover the essentials. Select 3 to 5 recent completed decisions, reconstruct the workflow graph for each, classify every edge as H2H, A2A, H2A, or A2H, estimate the senior human time consumed, and consolidate the results into a coordination edge radar. The deliverable is a single page that fits on the COO's and CFO's desks without requiring new software. This radar serves as the input for fully loaded cost allocation and qualifies subsequent vendor discussions. Without this initial exercise, any software platform purchased will measure the wrong operational category.
You are a COO at a 500-FTE SaaS company. At the last quarterly board meeting, the directors demanded a granular explanation of why operating margins have not matched the individual AI productivity gains reported by the team. The CFO approached you the following week seeking the same data. Meanwhile, the CTO sent a Slack message confirming that raw inference costs are under control, but the coordination around the models has expanded without a map.
You do not have a dashboard to answer them. You have no budget for new enterprise software this quarter. You have 30 days until the next board prep cycle. Where do you start?
The operational answer is the same one CFOs used in 2017 when cloud spend began appearing on the P&L without explanation. Before buying tools, take inventory. Before instrumenting, map the workflow. Do it on a sheet of paper, focus on 3 to 5 real decisions, and complete it in 30 days.
Why Inventory Before Instrumentation
The typical response to a new cost category is to buy a software platform first. A vendor demonstrates a polished dashboard, the team approves a proof of concept, integration begins, and three months later you discover that the tool measures raw API calls instead of completed decisions. You have spent your instrumentation budget on the wrong category and missed the quarter's planning cycle. This occurred in Cloud FinOps between 2015 and 2017, and it is repeating in human-agent coordination in 2026.
The economic dimension of this problem is straightforward. Human-agent coordination is the unnamed third layer of FinOps. In a typical mid-market company with moderate AI adoption, coordination carries 76% to 88% of the real cost of hybrid operations. Existing tools cover only 8% to 12% (raw inference) or bundle the rest into general payroll without distinct categorization. Purchasing software today without knowing which coordination edge leaks the most cash means measuring a minor fraction of the invoice with high precision and leaving the largest portion unmapped.
A paper-based inventory establishes the necessary baseline. In 30 days, it delivers three concrete assets for the boardroom and vendor evaluations. First, a map of the decisions that truly drive value. Second, an initial reading of where coordination efficiency leaks. Third, a shared vocabulary for the COO, CFO, and CTO to discuss operational costs without speaking past one another. With these three assets, subsequent instrumentation becomes defensible.
Operational Principle: 3 to 5 Real Decisions Are Sufficient
How many decisions should you map? Pragmatic discovery in mid-market organizations shows that 3 to 5 representative, recent decisions are sufficient. This is not a classic statistical sample; it is a pattern analysis. In a 500-FTE enterprise, critical decisions exhibit recurring workflows: the same senior executives, the same departments, and the same points of iteration repeat. The pattern emerges within three decisions. With five, you gain the confidence needed to present at the executive level. Beyond five, the marginal learning per extra decision decreases while administrative effort increases.
| Criterion | Strategic Importance | Typical Mid-Market Example |
|---|---|---|
| Crossed at least 3 departments | Siloed decisions do not reveal organizational coordination tax | A major client renewal involving Sales, Finance, Product, and CS |
| Involved AI in the workflow | Without agentic touchpoints, there is no hybrid coordination to evaluate | A churn analysis where an agent retrieves data before a head decides |
| Carried material business impact | Small decisions generate minimal graphs without indicating structural costs | Quarterly budget approval exceeding $100,000 |
| Occurred in the last 60 days | Recent memory allows for accurate workflow reconstruction | An enterprise pricing adjustment approved last month |
| Varied by type | Different categories of decisions reveal distinct coordination signatures | A balanced mix of commercial, product, financial, and talent decisions |
The "varied by type" criterion is essential. Inventorying five similar renewal decisions yields less signal than mapping one renewal, one pricing adjustment, one senior hire, one budget approval, and one customer churn analysis. Coordination signatures differ by decision type, and this variety highlights where operational leakage is most concentrated.
Step 1: Select 3 to 5 Recent Cross-Functional Decisions
Bring together the COO, BizOps, and one dedicated senior analyst. In a 90-minute session, list major decisions that crossed departments in the last 60 days and involved AI. Avoid political filters or premature screening. Capture all candidates, then filter them using the five criteria in the table above.
In a typical mid-market company, the initial brainstorming yield is 8 to 15 candidates, easily filtered down to 3 to 5. The most common types that survive this filter are enterprise client renewals, cohort pricing adjustments, quarterly budget approvals, senior hires, and executive responses to operational incidents. Document the final selection in a one-page index containing the decision name, date, primary stakeholder, and type. This index forms the cover of your inventory.
Step 2: Reconstruct the Workflow Graph for Each Decision
For each selected decision, schedule a 30-minute reconstruction session with its primary stakeholder. Begin with a simple question: once the decision was initiated, who was brought in, in what order, and what was the deliverable at each stage? Record this chronologically. Mark each node as either H (Human) or A (Agent). The connections (edges) between consecutive nodes represent the units of coordination that will populate your radar.
| Edge # | Origin Node | Destination Node | Edge Type | Edge Content / Deliverable |
|---|---|---|---|---|
| 1 | Head of Pricing (H) | Analysis Agent (A) | H2A | Initial query on cohort price elasticity |
| 2 | Analysis Agent (A) | Head of Pricing (H) | A2H | Initial analysis output with hypothesis |
| 3 | Head of Pricing (H) | Analysis Agent (A) | H2A | Context calibration and query refinement |
| 4 | Analysis Agent (A) | Forecast Agent (A) | A2A | Direct transfer of scenarios for modeling |
| 5 | Forecast Agent (A) | Head of Pricing (H) | A2H | Modeled scenarios delivered for review |
| 6 | Head of Pricing (H) | Commercial Director (H) | H2H | Initial review of proposed models |
| 7 | Commercial Director (H) | CFO (H) | H2H | Financial escalation for formal approval |
| 8 | CFO (H) | Forecast Agent (A) | H2A | Query for complementary margin impact analysis |
| 9 | Forecast Agent (A) | CFO (H) | A2H | Margin impact delivery for final review |
| 10 | CFO (H) | Commercial Director (H) | H2H | Decision approved with two minor caveats |
| 11 | Commercial Director (H) | Sales Team (H) | H2H | Operational communication of new pricing schedule |
The example above spans 11 edges. In real business scenarios, graphs typically contain 8 to 20 edges. Simpler decisions fall below this range; decisions involving political escalation extend beyond it. Do not force an artificial structure; reconstruct the workflow exactly as it happened. The critical outcome is a faithful sequence, not an arbitrary edge count.
Step 3: Classify Each Edge Using H2H, A2A, H2A, and A2H Taxonomy
While the table above maps these classifications inline, in practice this taxonomy is applied in a second pass over the raw graph. Use the H2H, A2A, H2A, and A2H framework as your baseline. Each edge type carries a distinct cost signature and specific growth drivers.
| Classification Question | Trigger Condition | Edge Classification |
|---|---|---|
| Is it a human communicating with another human (meeting, Slack, email)? | Yes | H2H |
| Is a human prompting, calibrating, or instructing an agent? | Yes | H2A |
| Is an agent delivering output for human review or approval? | Yes | A2H |
| Is an agent passing context directly to another agent without human review? | Yes | A2A |
| Is an agent executing an action autonomously in production? | Yes | A2A (Subtype) |
A common point of confusion is distinguishing H2A from A2H. The direct rule is to track the initiator: if the human queries the agent, it is H2A; if the agent delivers output to the human, it is A2H. In real hybrid workflows, these two alternate frequently, and each incurs a different type of cost.
Step 4: Estimate the Human Time Consumed by Each Edge
Stopwatches are unnecessary. Order-of-magnitude estimates are sufficient. For human edges (H2H), record the total person-hours consumed (a 90-minute meeting with 3 senior executives represents 4.5 person-hours of senior payroll, regardless of the absolute calendar slot). For agentic edges (H2A and A2H), record both the technical execution and the human time spent calibrating or reviewing.
This distinction is vital. A 60-minute meeting with 5 senior leaders consumes 5 senior person-hours, not 1. An A2H edge that appears instant (the model responded in 2 seconds) may have consumed 25 minutes of human prompt engineering and 15 minutes of subsequent review before the output could be used. Document the realistic coordination effort, not just the obvious system timestamps. A variance of 15% to 25% is acceptable; the goal is a defensible order of magnitude, not decimal precision.
| Edge Type | Measurement Unit | Typical Mid-Market Range |
|---|---|---|
| H2H Senior Meeting | Senior person-hours × N participants | 60 to 90 minutes × 3 to 6 participants |
| H2H Asynchronous (Slack, Email) | Consolidated senior person-hours | 20 to 90 minutes per decision cycle |
| H2A Calibration | Senior person-hours + API calls | 15 to 45 senior minutes + 3 to 8 queries |
| A2H Ratification | Senior person-hours spent reviewing | 10 to 30 minutes per output reviewed |
| A2A Handoff | API calls + human remediation time if required | 2 to 5 queries + 0 to 20 human minutes |
Step 5: Consolidate Into an Edge Radar Across Selected Decisions
The final deliverable of the inventory is a single-page edge radar. Each row represents an edge category (H2H, H2A, A2H, A2A), each column represents one of the mapped decisions, and the final column aggregates the total senior time consumed by edge type. This visualization highlights your organization's dominant coordination patterns.
| Edge Type | Decision A: Renewal | Decision B: Pricing | Decision C: Budget Approval | Decision D: Churn Analysis | Consolidated Senior Time |
|---|---|---|---|---|---|
| H2H (Meetings + Asynchronous) | 18h | 22h | 16h | 12h | 68 senior person-hours |
| H2A (Calibration) | 4h | 6h | 3h | 8h | 21 senior person-hours |
| A2H (Ratification) | 3h | 5h | 2h | 4h | 14 senior person-hours |
| A2A (Handoff) | 1h | 2h | 0h | 1h | 4 senior person-hours |
| Total per Decision | 26h | 35h | 21h | 25h | 107 senior person-hours |
The radar above tells a clear story. H2H dominates the aggregate (68 out of 107 hours, or approximately 64%). H2A follows with 21 hours (20%), and A2H consumes 14 hours (13%). A2A represents 4 hours (4%). In your organization, these proportions will differ. Companies with advanced AI adoption show heavier weights in H2A and A2A; initial adopters remain dominated by H2H.
Regardless of the final distribution, the critical gain is visibility. Before the radar, coordination costs were invisible, diluted across general payroll. With the radar in hand, you know exactly which edge leaks efficiency and possess the quantitative evidence required to defend your operational plan before the board.
What the Inventory Delivers in the First Month
Three concrete deliverables are achievable within 30 days using a lean team of the COO, BizOps, and one dedicated senior analyst: the index of selected decisions, the reconstructed workflow graphs for each, and the consolidated edge radar showing senior person-hours by edge type.
The operational cost is minimal. In a typical mid-market company, it requires 80 to 140 consolidated person-hours over 4 to 6 weeks. Compared to a standard software proof-of-concept (which consumes at least 3 months, $20,000 in licensing/integration, and 200 engineering hours), a paper-based inventory is a fast, cost-effective baseline that qualifies your subsequent discussions with vendors.
The gain in leadership authority is substantial. You move from saying "we do not yet measure human-agent coordination" to "we mapped it across 5 critical decisions, and this is our structural cost distribution." In the boardroom, this is the difference between defending a budget request and driving the operational agenda. The AI Multiplier paradox is explained in clear economic terms, not technical jargon.
What Follows the Inventory (Steps 6 to 8)
Once the radar is established, three subsequent movements become accessible in 30 to 60-day windows: Step 6: Allocating fully loaded payroll costs to each edge type. Step 7: Comparing consolidated coordination costs with cloud spend and general payroll on the same dashboard for the next QBR. Step 8: Selecting the 1 or 2 edges with the highest leakage and proposing targeted interventions (e.g., separating AI evaluation from standard management workflows, reducing prompting loops, or optimizing alignment meetings).
These steps lie beyond the scope of the initial inventory because they depend on your long-term instrumentation choices. You can continue to execute them manually on spreadsheets for 3 to 5 additional decisions each quarter, or use the radar to evaluate and select specialized software vendors. Under either path, the paper inventory remains the indispensable prerequisite. The formal dashboard that consumes this radar as its core input is detailed in five metrics that measure economic governance in cash terms.
Frequently Asked Questions
Why start with a paper-based inventory before buying software?
Because purchasing software without an operational map leads to collecting data that fails to drive decisions. The classic mistake is to buy a platform first, only to discover that it tracks raw API metrics rather than cross-functional workflows. A paper inventory answers three essential questions in 30 days: which decisions actually impact business margin, who is involved in them, and how much senior human time they consume. This map makes your subsequent software choice defensible and makes the right vendor obvious.
How many decisions must be mapped to gain statistical relevance?
Between 3 and 5 cross-functional decisions are sufficient for your first inventory. The goal is qualitative pattern analysis, not a formal statistical sample. Critical decisions in a mid-market company show highly consistent workflows, involving the same senior players and similar steps. The pattern is clear within three decisions; five provide the structural confidence needed for C-level presentation. Mending the workflow occurs after this initial inventory, not before.
Who within the organization should lead this inventory?
The COO or BizOps Director, supported by a senior analyst. The COO carries the organizational authority to cross departments and request reconstructions of recent decisions without political friction. BizOps provides the process-mapping expertise. If BizOps is not a formal function, the Chief of Staff or the CFO's executive assistant can lead it. Human Resources or People Ops is not the natural owner; this is an economic and operational exercise, not a behavioral study.
What if my company is in the early stages of AI adoption?
The inventory remains highly valuable. If AI adoption is low (15% to 30% of the team using agents), agentic edges (H2A and A2H) will appear less frequently but with high intensity when they do. Human-to-human edges (H2H) will dominate the map. This pre-AI baseline is exactly what you need to calculate the margin delta as adoption scales. Without a defensible pre-AI baseline, calculating AI ROI in 12 months remains a guess. Starting now protects your future financial analysis.
Does a 30-day paper inventory replace a measurement platform?
No. It replaces a poor starting point. The inventory delivers three essential prerequisites for any serious measurement initiative: a map of critical decisions, an understanding of which edges leak efficiency, and a shared executive language. With these assets, your conversation with software vendors changes completely. You move from a passive buyer to an informed client with internal benchmarks. The software platform becomes an accelerator of what you already understand, rather than a substitute for understanding.
The Bottom Line
The cost category that carries the largest portion of your AI invoice remains the invisible vector of AI governance. The economic framework is available, transaction cost theory has supported it for decades, and efficiency leaks are reproducible. What is missing is the first practical step: inventorying before instrumenting.
In 30 days, the COO, BizOps, and a dedicated senior analyst can map the initial inventory with zero capital expenditure. The deliverable is a single sheet of paper that fits on any boardroom table and initiates a clear economic conversation. The next board that demands a granular explanation of operational margins will find a defensible answer from leaders who took step zero, and educated guesses from those who did not.