When The Playbook Breaks

Abstraction is not the problem

Large operating systems are governed through abstraction.

No individual can directly observe the full operating environment, so physical conditions are translated into categories, workflows, metrics, dashboards and procedures that allow the organization to understand and act upon them. Somebody has to decide how much maintenance capacity to fund next year, which markets are underperforming, whether a vendor is expensive or merely assigned difficult work. Those decisions cannot be made by reading thousands of individual job tickets. The work has to be counted, and counting requires categories.

The quality of that governance depends partly on what survives the translation.

Every abstraction discards information. A category is useful precisely because it allows things that differ in some respects to be treated as equivalent for a particular purpose. The details left behind may be irrelevant to one decision and decisive to another.

The problem appears when a representation built for one class of decision is carried into another that requires information the representation no longer contains.

That makes the governing question more precise: which details can a particular decision afford to lose?

Three jobs, one line item

Consider three jobs on a maintenance schedule. Each requires the replacement of roughly three hundred feet of damaged fiber. Each restores the same service to the same standard. Each can reasonably be recorded against the same work type.

Their descriptions are nearly identical. Their execution is not.

Two kinds of difference run through the example. Physical conditions travel reasonably well across jurisdictions: buried infrastructure has to be exposed, rock changes excavation and boring, and disturbed surfaces require restoration. Procedural requirements are more local. Utility locating, permitting, traffic control, inspection and restoration standards depend on jurisdiction and ownership of the right of way.

The scenario is therefore illustrative. Its purpose is to show what happens when materially different physical conditions are represented by the same operational category.

The first is aerial. The damaged span runs along an existing pole line with existing attachments. The work requires a bucket truck, technicians, appropriate traffic control and splicing at each end. The physical condition is visible from the ground, allowing much of the job to be understood before the crew begins. Barring a surprise at the poles, the repair is relatively straightforward.

The second is buried in ordinary soil. The same fiber replacement now sits behind a different sequence of work. Locates may be required. Permits may be required depending on the route and jurisdiction. The cable has to be exposed, spoil managed, conflicts navigated and the disturbed surface restored. Inspection requirements may add another step. The splicing may be substantially the same as the aerial job. Much of the surrounding work is not.

The third is buried beneath asphalt with rock below it. Excavation and boring conditions change again. Different equipment and skills may be required. Entry and exit pits create their own excavation and restoration work. Rock introduces uncertainty into production rates. Pavement restoration may require another contractor and another schedule. Roadway involvement can increase traffic-control and permitting requirements. The consequences of encountering an unknown utility during a bore are different from encountering one in an open excavation.

By the time the three jobs are complete, their costs, durations, resource requirements and risk profiles can be radically different.

How much of that variation comes from physical conditions and how much comes from procedural requirements will depend on where the work occurs. The important fact for the operating model is simpler: conditions hidden by the work category can materially change what the work requires.

All three can still be correctly recorded as replace 300 ft damaged fiber.

Correct at one level, insufficient at another

The work type is not necessarily wrong. It may simply be answering a different question.

As a description of what happened to the network, replace 300 ft damaged fiber can be entirely adequate. If the organization wants to know how much fiber was replaced, where damage is increasing, or how often a particular repair occurs, many differences among the three jobs are irrelevant.

Now change the question.

Suppose the organization wants to understand why maintenance spend in one market is materially higher than another.

The same category may no longer contain enough information. A market with more difficult access, pavement restoration, rock or other costly physical conditions can consume substantially more resources per job than one where the same work type occurs under easier conditions. Yet the reporting may show both markets performing apparently equivalent work at different costs.

A conclusion about relative efficiency can then be perfectly reasonable from the representation presented and still poorly supported by the physical conditions underneath it.

The representation remained accurate. Its use changed.

Information sufficient for one class of decision was carried unchanged into another that required greater resolution. Nothing had to be falsely reported or incorrectly measured for the organization to reach a conclusion from information that could no longer support it.

The average illustrates the problem compactly. Combine materially different jobs into one unit cost and the resulting figure may correspond to none of the conditions actually encountered in the field. It can be arithmetically correct and operationally fictional.

That number may then become the basis for a budget, benchmark or unit rate.

What the chain discards

The loss accumulates as information moves through the organization.

A physical condition exists in the world: a damaged span, at a particular location, with particular access, surface and ground conditions.

A technician converts that condition into a work description.

The work description is mapped to a category.

The category becomes part of an aggregate metric.

The metric enters a dashboard.

The dashboard informs a decision.

Each transition reduces the amount of physical reality carried forward. Some reduction is necessary; otherwise the dashboard would simply be the network.

The governance problem is deciding what must survive.

For maintenance economics, surface or access conditions may materially affect the decision while a particular technical detail does not. A different decision may reverse that hierarchy. Capital planning, reliability analysis, vendor pricing and workforce planning can each require different resolution from the same underlying work.

Information has decision-dependent resolution.

The attributes worth preserving are determined by the decisions the representation will be asked to support. An organization does not need every detail at every level. It does need to know when a decision depends on information that was discarded earlier in the chain.

That requires understanding the intended range of each representation rather than treating every legitimate use as an obvious extension of the last one.

Precision is not accuracy

A common response to insufficient information is to collect more of it.

More fields are added. Categories multiply. Completion becomes mandatory. Estimates become more specific.

The resulting system is more precise. Whether it is more accurate depends on what those additional fields actually establish.

Consider the rock bore. Its eventual cost may genuinely be uncertain until work begins. A system that requires a single estimate does not remove that uncertainty. It converts the uncertainty into a number.

The number can then acquire authority simply because the system required one.

When actual cost differs, the difference appears as variance against the estimate even though some portion of that variance may have been inherent in the physical condition from the beginning.

The same problem appears in data collection. Twelve fields completed carefully can contain more reliable information than forty fields completed because the workflow will not advance without them. Increasing the resolution of a form does not guarantee an equivalent increase in the resolution of knowledge.

A highly detailed representation can therefore create a peculiar risk: greater precision can increase confidence faster than it increases understanding.

Not all variance is the same thing

Standardization depends on knowing what kind of variation it is trying to remove.

Some variance is noise. Two technicians perform the same task slightly differently with no meaningful difference in outcome. Standardizing that variation can make the system easier to operate without sacrificing useful information.

Some variance reflects inefficiency. Sequencing, staging, tooling or coordination may make equivalent work unnecessarily expensive or slow. Reducing that variation improves performance.

Some variance is noncompliance. A required locate is skipped. A crossing is installed outside the required standard. A ticket is closed without adequate verification. Discretion in these cases can introduce inconsistency and risk without producing a corresponding operating benefit. Strong systems deliberately remove some choices from the field.

And some variance is signal.

The job took longer because there was rock. The crew returned because the available record did not match the physical condition. The work cost more because access was genuinely difficult.

That variation contains information about the system itself.

Treat all four forms of variance as the same management problem and standardization begins removing information along with inconsistency. The resulting operation may become easier to compare while becoming harder to understand.

Where the variance goes

Removing material variance from a reporting model does not remove it from the work.

This becomes particularly visible in commercial relationships.

A vendor prices work against some representation of what that work is expected to require. If materially different conditions sit behind the same commercial category, actual delivery economics will vary even when the contracted unit does not.

Different contracts and vendors can absorb that variation in different ways. Portfolio mix may offset difficult work with easier work. Pricing may include contingency. Exceptions or change orders may account for conditions outside the assumed basis. Commercial structures may deliberately distribute risk between parties.

Where the model does not adequately account for the variation, however, the difference still has to be absorbed somewhere.

It may appear in margin. It may surface through change requests or selective bidding. It may influence staffing, scheduling, response behavior or future pricing. Under sustained pressure, it can also create incentives to reduce effort in places where quality is difficult to observe.

Which mechanism dominates is an empirical question, and the answer will differ by commercial arrangement.

When the model does not represent material variation in the work, that variation does not disappear. It is absorbed somewhere in the economic relationship.

Nothing about the physical condition changes merely because the commercial category cannot see it.

When the model becomes the system

The more serious condition develops when people throughout the organization understand the limits of a representation and decisions continue to treat it as complete.

Field teams may maintain an informal understanding of what particular jobs actually require alongside the category entered into the system. Vendors may price against the documented structure while planning delivery around conditions the structure does not capture. Budget owners may know that estimates are unreliable in particular circumstances while still needing a single figure for the planning process.

At that point, two operating models begin to coexist.

One is formal: the categories, estimates, dashboards and procedures through which the organization governs.

The other is experiential: the knowledge people use to make the work succeed despite what the formal model cannot represent.

The existence of that gap is not unusual. Every operating model is incomplete. The risk increases when the organization repeatedly depends on the informal model while continuing to make higher-level decisions as though the formal one were sufficient.

Then the representation stops being merely incomplete. It begins shaping decisions in ways the people closest to the work already know require qualification.

The playbook has become easier to govern than the system it describes.

What has to survive

The answer is better governance of abstraction itself.

Every important representation has a useful range. The organization should know which decisions it can support, what information it intentionally discards and which assumptions sit underneath it.

Attributes that materially change cost, risk or execution need to survive far enough through the information chain to reach the decisions that depend upon them. Which attributes those are will vary with the infrastructure, operating environment and decision being made.

Uncertainty also needs a legitimate representation. When the underlying quantity is genuinely a range, the operating model should be capable of carrying a range rather than manufacturing certainty for the convenience of the next system.

Variance requires the same discipline. Noise can be removed. Inefficiency can be improved. Noncompliance can be constrained. Signal has to remain visible long enough for the organization to learn from it.

None of this eliminates abstraction. It makes abstraction governable.

A category can be accurate without being sufficient. A dashboard can faithfully represent the data it received while omitting something essential to the decision being made from it. A standardized process can improve consistency while concealing meaningful differences in the physical system.

The discipline is knowing what the representation knows, what it has forgotten, and whether the next decision can afford the difference.