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
To start measuring what it costs to coordinate people and machines, you do not need software, extra budget, or anyone's permission. You need the same move finance made in 2017 when the cloud bill landed on the P&L with no explanation: take inventory on paper before you buy the instrument. Five steps that fit inside 30 days. Pick three to five real, recent decisions, redraw the path each one took, tag every link as human-to-human (H2H), agent-to-agent (A2A), human-to-agent (H2A), or agent-to-human (A2H), note how much time each one charges you, and pull it all into one radar. At the end, a single sheet that fits on the COO's desk and the CFO's, with no new dashboard. That sheet is the base for pinning a loaded cost to each edge, and for sitting down with a vendor on your own terms later. Skip this step and any platform you buy will measure the wrong slice of the bill, to the decimal.
The scene is familiar to anyone who runs operations. At the last close, the board asked for a fine-grained explanation of why margin did not follow the AI gains the whole team swears it delivered. The next week the CFO knocked on your door wanting the same number. The CTO sent a Slack saying inference is under control, but the coordination around it has grown without a map. Three people after the same answer, and it lives nowhere.
You have no dashboard to pull. You have no budget for new software this quarter. You have 30 days until the next board prep. The question is an honest one: where do you start when you cannot buy anything?
Where finance started in 2017, when the cloud invoice landed on the P&L with no owner. Before buying a tool, take inventory. Before instrumenting, draw the map. On paper, with three to five real decisions, in 30 days. The category changed; the first move is the same.
Why Inventory Before Instrumentation
The reflex in front of a new cost category is to buy the platform first. The vendor shows a clean dashboard, the team approves a proof of concept, the integration starts, and three months later someone notices the tool counts API calls and not cross-functional decisions. The instrumentation budget went to the wrong target, and the quarter is gone. It happened to the cloud bill between 2015 and 2017. It is happening again with human-agent coordination right now.
The economic version of the problem is easier to see. Human-agent coordination is the third layer of FinOps nobody named. Run the math in your head: in an operation that already has AI inside the flow, inference spend is the visible tip, a small fraction of the real bill. Everything else, the senior hour that calibrates, ratifies, and fixes what the machine handed over, sits in payroll, with no category of its own, with nobody adding it up. Buying a tool today without knowing which edge leaks the most in your own house is measuring the small slice precisely and the big one not at all.
The paper inventory settles the precondition. In 30 days it delivers three assets that hold up as real defense in front of the board and in front of a vendor. The map of which decisions actually carry weight in the company. The first reading of where coordination is leaking most. And a shared language for the COO, CFO, and CTO to talk about the category without each one in their own dialect. With those three in hand, instrumenting stops being a bet.
Operational Principle: 3 to 5 Real Decisions Are Enough
The natural next question is how many decisions to map. The answer comes from the nature of the problem itself: three to five recent decisions that mattered. This is not a statistical sample, it is a pattern read. A weighty decision that crosses departments repeats the same graph: the same senior people, the same areas, the same points where the thing stalls and circles back. By the third decision the pattern is already in front of you. By the fifth you have the confidence to defend it at the C-level. From the sixth on, each new case teaches less and charges more hours.
| Criterion | Why it matters | Concrete example |
|---|---|---|
| Crossed at least 3 departments | A decision stuck in a single area does not show the company's coordination | A major renewal pulling in Sales, Finance, Product, and CS |
| Had AI in the path | With no edge to an agent, there is nothing to read on the hybrid side | A churn analysis with an agent pulling data before the head decides |
| Actually mattered | A small decision makes a thin graph, with no signal of expensive coordination | A quarterly budget approval above $100,000 |
| Happened in the last 60 days | Fresh memory reconstructs the path with decent accuracy | A pricing adjustment on an enterprise plan closed last month |
| Different from the others in type | Variety reveals distinct edge signatures by category | A mix of commercial, product, financial, and people decisions |
The "different in type" criterion deserves a word. Inventorying five renewals teaches you less than inventorying one renewal, one pricing adjustment, one senior hire, one budget approval, and one churn analysis. Each type of decision has its own coordination signature, and it is the variety that shows where your company leaks most, and on which front.
Step 1: Pick 3 to 5 Recent Cross-Functional Decisions
Put the COO, someone from operations, and a dedicated senior analyst in a room for four to six weeks. In a 90-minute session, list together the weighty decisions that crossed departments in the last 60 days and had AI in the path. No politics, no censoring. Write down every candidate, and only then run the filter of the five criteria from the table above.
The raw list usually comes out with a dozen candidates, and the filter cuts it to three to five without effort. The types that show up most at this stage: the major renewal, the pricing adjustment on a plan, the quarterly budget approval, the senior hire, the leadership response to an incident. The final selection lives on a sheet with the short name of the decision, the date, who owned it, and the type. That sheet is the index of the inventory.
Step 2: Reconstruct the Graph of Each Decision
For each chosen decision, ask 30 minutes of whoever owned it. The opening question is simple: from the moment it hit the agenda, who got called, in what order, and what came out of each stage. Write it in the order it happened. Mark each node as H (human) or A (agent). The edge between two consecutive nodes is the unit of measure that goes into the radar.
| # | Origin (node) | Destination (node) | Edge type | Edge content |
|---|---|---|---|---|
| 1 | Head of Pricing (H) | Analysis agent (A) | H2A | Opening question on price elasticity |
| 2 | Analysis agent (A) | Head of Pricing (H) | A2H | First output with hypotheses |
| 3 | Head of Pricing (H) | Analysis agent (A) | H2A | Context calibration and refinement |
| 4 | Analysis agent (A) | Forecast agent (A) | A2A | Handing scenarios over for modeling |
| 5 | Forecast agent (A) | Head of Pricing (H) | A2H | Modeled scenarios for ratification |
| 6 | Head of Pricing (H) | Commercial Director (H) | H2H | First walkthrough of the scenarios |
| 7 | Commercial Director (H) | CFO (H) | H2H | Escalation for financial approval |
| 8 | CFO (H) | Forecast agent (A) | H2A | Request for extra margin analysis |
| 9 | Forecast agent (A) | CFO (H) | A2H | Margin analysis for ratification |
| 10 | CFO (H) | Commercial Director (H) | H2H | Decision approved with 2 caveats |
| 11 | Commercial Director (H) | Sales Team (H) | H2H | Rollout of the new price to the field |
The example above has 11 edges. In real life the count usually lands between 8 and 20: simple decisions sit below, the ones that climb the political ladder sit above. Do not force a template, reconstruct how it actually went. What counts is the faithful sequence, not a round number of edges.
Step 3: Classify Each Edge as H2H, A2A, H2A, or A2H
The table above already carried the classification in the row itself. In practice it happens on a second pass, with the raw graph already built. Use the H2H, A2A, H2A, and A2H typology in cash terms as your reference. Each type charges in its own way and swells from its own trigger.
| Question | Answer | Edge type |
|---|---|---|
| Person talking to person (meeting, Slack, email)? | Yes | H2H |
| Human calibrating, asking, or prompting the agent? | Yes | H2A |
| Agent delivering something a human has to ratify? | Yes | A2H |
| Agent passing context to another agent with no human review? | Yes | A2A |
| Agent taking action in the world with nobody ratifying? | Yes | A2A (subtype) |
The classic mix-up is swapping H2A for A2H. The direct rule is to look at who starts. If the human pulls the agent, it is H2A. If the agent delivers to the human, asked for or not, it is A2H. In a genuinely hybrid decision, the two trade places several times in the same case, and each charges a different kind of cost.
Step 4: Note the Rough Time Each Edge Consumes
Forget the stopwatch. A rough estimate gives you the order of magnitude. For the human edges, count in senior person-hours: a 90-minute meeting with three senior people is 4.5 person-hours, even if the calendar only shows 90 minutes. For the agentic ones, count the inference calls plus the time a human spends waiting until the output is good enough to use.
The difference changes the number. A one-hour meeting with five senior people burns five person-hours, not one. And the A2H edge that looked instant (the agent came back in two minutes) may have charged 25 minutes of calibration before and 15 of review after. Note the coordination that actually happened, not the wall clock. The target here is the order of magnitude you defend at the board, not the fifth decimal place.
| Edge type | How to count | Typical range |
|---|---|---|
| H2H senior meeting | Senior person-hours × no. of participants | 60-90 minutes × 3 to 6 senior people |
| H2H asynchronous (Slack, email) | Senior person-hours spent in the back-and-forth | 20 to 90 minutes summed per decision |
| H2A calibration | Person-hours + inference calls | 15 to 45 senior minutes + 3 to 8 calls |
| A2H ratification | Senior person-hours checking the output | 10 to 30 minutes per output to ratify |
| A2A handoff | Inference calls + human fix if there was one | 2 to 5 calls + 0 to 20 human minutes |
Step 5: Consolidate Into an Edge Radar Across 3 to 5 Decisions
The final deliverable is a single sheet, consolidating the three to five decisions into an edge radar. Each row is a type (H2H, A2A, H2A, A2H), each column is a decision, and the close is the aggregate column with the total per type. One glance and the coordination pattern that rules your company jumps off the page.
| Edge type | Major renewal | Pricing adjustment | Quarterly budget | Churn analysis | Aggregate total |
|---|---|---|---|---|---|
| H2H meetings + async | 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 read is direct. H2H eats the bulk of the aggregate, 68 of the 107 hours, close to two thirds. H2A comes second, 21 hours, a fifth. A2H takes 14, A2A takes 4. In your company the shape will be different: where AI adoption is already high, H2A and A2A weigh more; where it is still finding its feet, H2H dominates by an even wider margin. The picture shifts from house to house, but it always reveals which edge pulls the bill.
What does not change in any case is the visibility. Before the radar, the cost of coordination sat dissolved in senior payroll, with no label. After it, you know which edge leaks most and you have a number in hand to defend at the board. You leave guesswork and enter evidence.
What the Inventory Delivers in the First Month
Three concrete deliverables fit inside 30 days with the COO, the operations team, and a dedicated senior analyst. The index sheet with the three to five decisions. The reconstructed graph of each one, with the edge already classified. And the aggregate radar, with senior person-hours per edge type across the decisions.
The cost of doing it is low. Estimate something between 80 and 140 senior person-hours spread over four to six weeks, and redo the math with your loaded hour. Compare it with any platform proof of concept, which asks for three months minimum, six figures of integration, and hundreds of engineering hours, and the paper inventory is the cheap, fast option that comes before the serious conversation with a vendor.
The gain in credibility is large. You move from "we still do not measure what it costs to coordinate people and machines" to "we measured it across three to five real decisions, and this is the house pattern." In front of the board, that is the difference between handing over the microphone and setting the agenda. The AI Multiplier paradox gets explained in cash, not in jargon.
What Comes After the Inventory (Steps 6 to 8, Out of Scope Here)
With the radar ready, three next moves come within reach, each in a 30 to 60-day window. Step 6 pins a cost to each edge, using the company's loaded senior hour. Step 7 puts the aggregate cost of coordination side by side with cloud spend and payroll, on the same radar, to take to the next results conversation. Step 8 picks one or two edges with the highest leakage and proposes a targeted intervention: split the AI agenda from the management agenda, shorten the calibration loop, kill the meeting that only exists to align AI usage.
These steps fall outside the scope of the initial inventory because they depend on the decision to instrument. You can run them by hand, on paper, for another three to five decisions a quarter, or use the finished radar to vet vendors with a demanding eye. On either path, the initial inventory is the precondition. The formal dashboard that feeds off this radar is covered in five metrics that measure economic governance in cash.
Frequently Asked Questions
Why start with an inventory before instrumenting?
Because a tool without a map delivers data with no read. The reflex of someone spooked by a new cost category is to buy the platform first, and three months later discover it measures the wrong thing. The paper inventory answers, in 30 days, three questions no software answers on its own: which decisions actually weigh on the company, who gets called into each one and in what order, and how much senior time each edge consumes. With that map in hand, the decision to instrument becomes defensible and the right vendor stands out on its own.
How many decisions do I need to map to get signal?
Three to five cross-functional decisions are enough for the first inventory. It is not a statistical sample, it is a pattern read. An important decision that crosses areas has a recurring graph: the same areas, the same senior people, the same points of iteration. By three decisions the pattern is already in front of you. By five you gain the confidence to take it to the C-level. From the sixth on, each extra decision teaches less and charges more time. The choice about instrumenting comes after the map, never before.
Who in the company should lead this inventory?
The COO or operations director, with a dedicated senior analyst for four to six weeks. The COO carries the authority to cross areas and ask for the reconstruction of a recent decision with no political noise. The analyst has the muscle to draw process. Where there is no formal operations function, the chief of staff or the CFO's right hand works too. HR is not the natural owner of this: the read is economic, not behavioral. The CFO comes in on the next pass, with the map ready, to pin cash to each edge (step 6, outside this text).
What if the company still uses little AI?
The inventory runs the same. With adoption still early, the human-agent edges show up less often, but heavy in each decision they touch. The human-to-human edges remain the bulk of the map, and that pre-AI snapshot is exactly what you need: it is the ruler you will measure how much changed against when adoption grows. Without that starting snapshot, the return on AI twelve months out is an educated guess. Drawing the map now protects the read later.
Does the 30-day inventory replace a measurement platform?
No. It replaces the bad starting point of buying the platform first. The inventory delivers three things that precede any serious instrumentation: the map of which decisions matter, the read of which edges leak most, and a shared language between COO, CFO, and CTO. With those three assets in hand, the conversation with a vendor changes character. You stop being a generic buyer and become a demanding one, with your own criteria and an internal reference. The platform then accelerates what you already understood, instead of patching the hole of what you had yet to understand.
The Bottom Line
The category that carries the largest slice of the AI bill today is still the invisible vector of AI governance. The economic read is on the table, the theory behind it is almost ninety years old, and the leak repeats from company to company. What is missing is the first practical move: inventory before you instrument.
In 30 days, the COO, the operations team, and a senior analyst close the initial inventory without spending more. What comes out is a sheet that fits on any board agenda and opens the conversation in cash. The next board that asks for a fine-grained explanation of post-AI margin will find a defensible answer from whoever did step zero, and an educated guess from whoever left it for later.