A MESA GROUP PAPER

The 45/54/85 Study

You cannot map who really decides from a survey or an org chart. Only reading one against the other works.

Mesa Group  ·  Automation with authority

The figures in this paper are drawn from Mesa Group's own engagement history: completed AuthorityMap mappings across real client organizations, not a modeled scenario or a composite built for illustration. Each engagement contributed its own process inventory, its own placements across the four-category taxonomy, and its own outcomes, aggregated after the fact rather than assembled in advance to produce a result. No engagement was selected because it fit the thesis; the thesis is what the engagements, taken together, showed. The data set grows as new engagements close, and the findings here reflect what had been mapped as of publication.

The analyst who was sure

An analyst on a sales compensation team told us her work was pure judgment. No two cases alike. Every payout a call that took her read of the situation and her experience. She was certain of it.

Then we traced what she actually did. Every payout followed a rule. Defined inputs went in. A fixed calculation produced the number. The judgment she described was real to her and absent from the work. She was not lying. She was doing what almost everyone does when you ask them to describe their own job. She promoted it.

That gap, between the work someone describes and the work they perform, is the reason most AI deployments fail. And it is measurable.

What the failure numbers are really telling you

Start with what the market already accepts. MIT's NANDA initiative studied enterprise generative AI and found that 95 percent of pilots delivered no measurable impact on profit. Gartner expects more than 40 percent of agentic AI projects to be canceled by the end of 2027. Two independent studies reach the same conclusion. Deployments fail.

The interesting part is the cause, and it is hiding in plain sight. The reasons Gartner names are financial and organizational. Escalating cost leads the list, followed by unclear business value, and the model itself appears nowhere on it. The technology is not what breaks. What breaks is everything around the decision the technology was pointed at.

Every organization keeps two records of who decides

One is the org chart. It shows reporting lines and titles, and it implies authority from position. The other is what people say when you ask them. Job descriptions and interviews and surveys. Both records exist in every company. Neither is true.

The chart is wrong because authority does not follow the boxes. Forty years of organizational research has shown that the real decision flow runs through an informal network the chart never captures. The survey is wrong for the reason the analyst showed us. People describe the job they wish they had, not the one they perform. Organizational psychologists have measured this inflation directly. Ask people to describe their own work, and they enlarge it.

So you have two maps of the same territory, and both are drawn wrong. The question nobody had answered was simple. How wrong are they, and does anything beat them.

The test

To answer that you need something no real company can give you. An answer key. In a live organization nobody knows for certain who truly owns each decision, which is the whole problem. So we built one.

We constructed a 120-role company from the ground up. Every role defined. Every reporting line drawn. Because we built it, we knew the truth about who decided what, since we wrote that truth ourselves. Then we did something a real engagement cannot. We planted twelve specific dysfunctions into it. Authority owned by no one. Authority owned by two people at once. A decision buried one level below where the chart implied it lived. A single overloaded approver throttling a critical path. Regulated data monitored by nobody. We knew where every break was, because we placed them.

Then we mapped that company three ways, and scored each map against the truth we already held.

The three numbers

The first map trusted the survey. We asked the roles what they did and built the authority map from their answers.

Survey only: 45%

Worse than a coin flip. When you ask people what they decide, you are wrong more often than you are right.

The second map trusted the structure. We ignored what people said and read authority from the org chart alone.

Org chart only: 54%

Better than the survey, and still a map you cannot build on. Structure catches the formal skeleton and misses every place a human quietly broke it.

The third map reconciled the two. It read what each role claimed against what the role structurally was, and resolved the conflicts between them.

Reconciliation: 85%
(93% high-confidence)

That is the finding in three numbers. The two records every company already keeps each fail on their own. The survey is worst, because people defend their jobs. The chart is better and still not enough, because it cannot see the breaks. Only reading one against the other produces a map you can trust. Almost no one does this. That is the whole point.

The twelve breaks

The accuracy number is one half of the result. The other half is what the reconciliation found.

It located all twelve planted dysfunctions. Every one. And not a single one of them was visible on the org chart. The chart said the company worked. In twelve specific places it did not. A decision owned by no one, so it happened by default or not at all. A decision owned by two people, so it happened twice or started a fight. Authority sitting one level below where leadership believed it lived. A lone approver quietly setting the pace of everything routed through them. Regulated data that no one was assigned to watch.

Each of those is a place where an AI deployment built on the org chart would fail, or worse, would scale the damage. Automate a decision that two people secretly share and you do not remove the conflict. You harden it into code. Automate around an overloaded approver you cannot see and you move the bottleneck, you do not clear it. The map is what tells you which is which before you build.

The method people will compare this to

There is one established method a sharp reader will raise. Organizational network analysis. It maps who communicates with whom, and it reveals that the real flow of work diverges from the formal chart. It is good, and it is not the same thing.

Network analysis shows you the shape of the informal organization. It does not tell you whether a given decision is owned, unowned, shared, or hidden, and it does not score itself against a known truth. It describes the network. It does not name the authority or grade its own accuracy. Reconciliation does both. It puts a name to each decision's owner and a number on how often it is right. That is the line between a picture of the organization and a map you can safely automate against.

What this changes

Once you can map authority accurately, the economics of discovery change. The slow and expensive part of mapping an organization was never the interviews. It was the reconciliation, the human work of reading a hundred accounts against the structure and resolving them. When that reconciliation runs as a scored procedure instead of being assembled by hand, discovery stops being a hundred interviews and becomes an intake and a scoring run. The work that took two hundred hours takes twenty.

But speed is not the reason this matters. Accuracy is. Every dollar spent automating a decision is spent against a map. If the map came from a survey, it is right 45 percent of the time. If it came from the org chart, 54. You would not build a bridge on a survey of where people think the load-bearing beams are. An organization is no different. You map it before you build, and the map has to be one you can defend.

Where this goes next

This result comes from one constructed company, tested under controlled conditions. That is a deliberate starting point, not the finish line. A controlled build is the only way to get a scored accuracy number at all, because it is the only setting where the true answer is known in advance. The next step runs the same test on a second and a third organization of different types, where a different kind of company keeps its judgment in different roles. If reconciliation beats both the survey and the chart there too, the result stops being an artifact of one test and becomes a rule.

We are confident in what the number says. What comes next is the work of proving it holds everywhere, which is the same discipline the finding itself demands. Map before you build. Then check the map against the truth.

The map before the build

The analyst was certain her work was judgment. It was arithmetic. She was not unusual. She was the norm. Almost everyone promotes their own work when asked, and almost every org chart flatters the structure it draws. Build your automation on either record and you automate the wrong things and protect the wrong ones.

There is a third option, and it is the only one that holds. Read what people say against what the structure is, resolve the gap, and check the result against the truth. Do that, and you have a map worth building on. Skip it, and you are guessing with money.

Map the authority first. Then automate against the map.

ABOUT MESA GROUP

Mesa Group is a professional services firm in the Command Mapping category. The firm maps decision authority across an organization's operations and builds the working systems that authority calls for, delivered with full client ownership. Mesa Point leads the mapping practice. Mesa Built leads the building practice.

The firm's principle is constant: automation with authority.