From Field Progress to Verified Progress: Where Can AI Help?

What does 40% complete mean?

A progress report says a conduit package is 40% complete. The number looks settled. It isn’t, until three questions are answered: 40% of what, measured how, and accepted by whom?

In workflows I have worked with, installed progress is calculated from upstream quantity and earned-hour information. Foremen review the calculated percentage and can accept it or override it with a reason. Project controls and management review the movements, and earned and remaining standard hours follow from the result.

That sequence holds a distinction every downstream forecast depends on. A calculated percentage is a proposal from the data. Verified progress is a percentage someone accountable has accepted against a declared rule.

This article follows one synthetic electrical work package along that path, then asks where 360° imagery and AI can help.

From installed quantities to progress

Synthetic work package (invented for teaching) Level 3 East, branch-circuit EMT conduit rough-in Planned quantity: 1,000 m · Budget: 200 hours (0.20 h/m) Last approved progress (previous period): 25%, 50 earned hours Actual hours to date: 75

This period, upstream quantity reporting shows 400 m installed.

• Calculated progress = 400 ÷ 1,000 = 40%
• Calculated earned hours = 200 × 40% = 80 h

The arithmetic is simple. What gives it meaning is the measurement method behind it. Earned value guidance expects objective performance criteria to be set for each work package before work starts, with progress then statused against those criteria (EIA-748 guidelines, Rev E summary). In construction, those criteria usually take the form of rules of credit, and they are not universal. One AACE conference presentation notes that there is still no consensus on which measurement method suits linear construction, so owners and contractors typically negotiate them (AACE).

In this package the rule is units completed: a metre earns credit when it is permanently supported and complete as rough-in.

The rule matters more as scope grows. Suppose the package also covered wire pull and terminations, under an illustrative weighted rule crediting rough-in at 50%, pull at 30% and terminations at 20%. Then 400 m of conduit with no wire pulled would earn 400 ÷ 1,000 × 50% = 20% of the package, not 40%. The field is the same; the rule is different.

So “40%” is not a property of the field. It is the output of a declared method applied to reported quantities.

V1

Practitioner’s Planning Notebook

The progress chain

Flow from 400 m installed through a units-completed credit rule to 40% and 80 earned hours, with an inset showing a weighted rule giving 20%.

Before field review · Calculated, not verified or approved

1. Installed quantity400 m
2. Credit ruleUnits completed
3. Calculated progress40%
4. Calculated earned hours80 h

Alternative scope: weighted credit rule

Same field. Different rule.

Rough-in 50%Pull 30%Terminate 20%

400 ÷ 1,000 × 50% = 20% of the weighted package

40% of a declared method, not of the field.

When the foreman disagrees

The foreman reviews the 40% and proposes 30%. The reason: about 100 m in the corridor ceiling is hung on temporary supports, with permanent supports and fittings still outstanding. Under the credit rule, that length is not yet creditable.

• Proposed creditable quantity = 400 − 100 = 300 m
• Proposed progress = 30%
• Proposed earned hours = 60 h
• Adjustment = −20 earned hours

Is the override defensible? EVMS practitioners warn that percent-complete estimates without backing data drift toward guesswork. Their standard remedy is quantifiable backup data: predefined steps or milestones that make a progress claim auditable (Humphreys & Associates). Applied here, a defensible override carries five things:

1. a location;
2. a quantity basis;
3. a reason tied to the credit rule;
4. evidence, such as photos or a marked-up drawing;
5. an author and a timestamp.

This one has all five. “Feels more like 30%” would have none.

Overrides in either direction carry risk. An upward override can credit work before it meets the rule, inflating earned hours and flattering productivity. A downward override can be accurate field knowledge, as here, but it can also hold hours in reserve, making productivity look worse and prompting recovery effort that isn’t needed.

A downward correction can also point upstream: if quantity reporting counted temporarily supported conduit as installed, the data feed needs attention too.

The test is the same in both directions: does the evidence support the rule?

What changes in project controls?

The override creates a number in a new state. It helps to keep four states distinct:

State	Progress	Earned hours	Meaning
Calculated	40%	80 h	Output of the data and the rule
Proposed	30%	60 h	Foreman’s correction, with basis
Reviewed	30%	60 h	Project controls has checked the basis
Approved	30%	60 h	Accepted by the accountable manager and posted

Teaching assumption (synthetic, not a standard): Any override needs a location, quantity basis, reason code and evidence. A movement of more than 5 percentage points or more than 10 earned hours against the calculated value requires project-controls review and project manager approval before posting.

This correction is −10 points and −20 hours, so it triggers both thresholds.

V2

Practitioner’s Planning Notebook

Four states of one number

Ledger showing one package moving from calculated 40% to approved 30%, with the reason, the threshold trigger and the pending-state rule.

A. While awaiting approval

Last approved

25% / 50 h

Proposed

30% / 60 h

Status

Pending

Reviewed does not mean approved.

B. After approval

Approved: 30% / 60 h

Accepted by the accountable manager and posted.

−10 percentage points versus calculated 40%.

+5 percentage points versus prior approved 25%.

Illustrative approval thresholds: more than 5 percentage points or more than 10 earned hours against the calculated value requires project-controls review and project manager approval before posting. This correction: −10 points / −20 h. Teaching assumption.

Four-state ledger and supporting evidence

From the calculated result to the accepted value

State	%	Earned h	Who / what
Calculated	40%	80 h	Data + declared rule
Proposed	30%	60 h	Foreman’s correction + basis
Reviewed	30%	60 h	Project controls checks basis
Approved	30%	60 h	Accountable manager accepts; posted

100 m on temporary supports, Corridor L3E, photos + markup, Foreman, date.

What does the report show while the correction is pending? In this example, the last approved value holds at 25% and 50 earned hours, with the proposed 30% visibly flagged beside it. If no approved baseline existed, the package would report as pending/unverified rather than borrowing the calculated figure. The principle is that an unapproved number never quietly becomes the reported one.

Once the correction is approved:

• Remaining standard hours = 200 − 60 = 140 h

That is not a forecast. In earned value terms, budget minus earned value is “work remaining,” which implicitly assumes the rest of the work will be done at the budgeted rate (SAS EVM documentation). A forecast needs an explicit productivity assumption:

• Observed productivity = 60 earned ÷ 75 actual = 0.80
• Forecast hours to complete at 0.80 = 140 ÷ 0.80 = 175.0 h
• Forecast hours to complete at an assumed 0.85 = 140 ÷ 0.85 = 164.7 h

Article 001 argued that measured productivity should inform the forward assumption but not automatically become it. Here, 140 hours is a statement about scope. 164.7 or 175 hours is a statement about performance.

Notice what the unverified figure would have done. Had the calculated 80 hours been posted, remaining standard hours would read 120 and observed productivity 1.07, which tells a very different story from the same corridor ceiling.

V3

Practitioner’s Planning Notebook

Two kinds of “remaining”

Bar comparison of 140 remaining standard hours against forecasts of 164.7 and 175 hours.

Remaining standard hours versus forecast hours to complete · Common scale: 0–200 h

Remaining standard140 h

scope (measured)

200 − 60 = 140 h

At 0.85 assumed forward productivity164.7 h

performance (assumed)

140 ÷ 0.85 = 164.7 h

At 0.80, assuming observed productivity continues175.0 h

performance (assumed)

140 ÷ 0.80 = 175.0 h

050100150200 h

The observed 0.80 becomes a forward performance assumption when used for the forecast.

Using unapproved 40% would show 120 remaining standard hours and 1.07 observed productivity.

Can 360° imagery and AI help?

Commercial reality-capture platforms make strong claims. OpenSpace, as one vendor example, now describes its progress-tracking approach as AI-powered 360° capture paired with expert human review, and claims reliable, accurate results (OpenSpace). The cited drawing-set guidance belongs to its legacy OpenSpace Track solution: that support article says percentages and quantities are based on the most recent drawing set and explicitly identifies the solution as no longer offered (OpenSpace legacy support). These descriptions are vendor claims, not independent accuracy evidence. I would test their applicability and accuracy on your own scope.

Peer-reviewed research frames the problem more cautiously:

• A 2022 review by Reja, Varghese and Ha describes a pipeline of data acquisition and 3D reconstruction, as-built modelling and progress assessment, and calls for benchmarking of these pipelines (Automation in Construction, 2022).
• Comparing images or point clouds with a model requires line of sight to the elements, and research is still developing ways to infer the state of occluded components (Advanced Engineering Informatics, 2015).
• A 2026 MEP case study notes that densely overlapping components make MEP scenes particularly prone to occlusion (EC3 2026).

For our package, AI could potentially:

• Observe exposed conduit in the corridor ceiling before close-in, tied to date and location;
• Extract an estimate of visible installed length against the drawing;
• Reconcile that estimate with the 400 m reported;
• Flag the gap between calculated and proposed figures, an override without evidence, or areas with no capture.

Visible features may be detected. Images alone cannot establish full installation compliance or acceptance. In this package, that means AI cannot:

• confirm that the supports meet the standard the credit rule requires;
• see inside the conduit;
• confirm terminations or test results;
• decide what the credit rule should be;
• approve the movement.

The wire-in-conduit point matters for electrical work. Under the weighted rule above, the steps a camera sees least, pull and terminations, carry half the credit.

So the realistic potential is in organizing evidence and surfacing exceptions for people to examine. I have not seen evidence that it makes progress verification faster or more reliable, and it does not approve.

V4

Practitioner’s Planning Notebook

What the camera sees

Matrix contrasting what AI could potentially do with 360° imagery against what still needs human verification, noting that visible features do not establish compliance or acceptance.

AI could potentially

Observe

Exposed conduit before close-in, tied to date and location.

Extract (estimate)

Visible installed length against the drawing.

Reconcile

Compare the estimate with the 400 m reported.

Flag

Gaps, missing override evidence or areas with no capture.

Needs a person

• Support compliance
• Inside the conduit
• Terminations / tests
• Credit rule
• Approval

Human verification may need inspections, records and tests beyond the image.

Visible ≠ compliant ≠ accepted.

Vendor claims are not independent evidence.

Closing reflection

The 40% was not wrong arithmetic. It was a correct calculation on incomplete information, and therefore an unverified claim about the field. The foreman’s 30% was better information, but it was still a proposal until someone accountable accepted it.

That is the chain I would keep in any progress system, with or without AI: calculated, proposed, reviewed, approved. Earned hours come from the approved state, remaining standard hours from earned hours, and forecast hours from remaining hours plus an explicit productivity assumption.

AI may help organize the evidence along that chain. Accountability for the number still sits with people.

All figures in this article are synthetic and invented for teaching. They do not represent any real project.