Site Observations

Why observations are critical

The economics of early detection are the whole argument. The UK’s Get It Right Initiative puts the total cost of avoidable error at 10–25% of project cost once indirect and latent effects are counted — roughly seven times the sector’s annual profit. And the quality profession’s 1-10-100 rule describes the escalation precisely: a dollar to fix at source, ten once it is built in, a hundred after delivery.

An observation is the one-dollar intervention. The same deviation, uncaught, resurfaces as an NCR — formal, commercial, root-cause-bearing — then as a failed handover inspection, then as a latent defect in year three, with a delay attached to each step.

The safety literature supplies the second argument, and it is the one people get backwards. Observation volume is a leading indicator. Programmes built on high-frequency observation data show measurably fewer severe outcomes, and rising observation counts with steady closure signal a healthier reporting culture — not a worse site. A falling observation count is as likely to mean people stopped looking as it is to mean there is less to find.

The role of observations in project performance

• Get the taxonomy right — and be honest that it is convention. An observation is an early, often minor or unclear issue: documented, assigned, corrected quickly, with no formal QA machinery attached. An NCR is confirmed non-compliance, carrying containment, root cause and verified closure. A snag is a minor defect at or near completion, managed through closeout. None of this is codified anywhere — it is convention, and it varies by project. Which is exactly why a project must publish its own definitions and its own escalation triggers. Sites that escalate by mood produce both failure modes at once: paper wars, where everything becomes an NCR, and buried problems, where nothing does.

• The escalation ladder, written down. Observation, fixed and photo-verified. Repeated or ignored, and it becomes a site instruction directing correction. Confirmed as actual non-compliance, and it becomes an NCR. Unresolved at handover, and it blocks completion. A published ladder with trigger criteria is what separates a quality system from a quality mood — and the triggers are what make it fair, because they are applied by rule rather than by whoever is annoyed that week.

• Trends are the culture gauge, not the volume. The KPIs that matter are closure time, repeat-category rate, and the observation-to-NCR conversion rate. Steady observation volume with a falling conversion rate is what a maturing quality culture actually looks like: problems caught earlier, fixed faster, escalating less. Volume alone misleads in both directions, and reading it alone is how good programmes get punished.

“Observed” is not “inspected” — and the words carry liability

This is the part most software, and most consultants, get carelessly wrong.

Consultant field reports occupy a deliberately bounded legal position. The consultant visits at intervals and observes the work for general conformance with the design intent. The standard forms reserve the word “inspection” for substantial and final completion precisely in order to bound the duty of care — the language is doing legal work, not stylistic work.

So a field report that casually says “inspected and approved” silently raises the standard the consultant will be held to when a defect emerges years later. Nobody notices at the time. It costs nothing to write and it can cost a great deal to have written. Precision in report language is not pedantry; it is professional-liability hygiene.

What happens without an observation layer

Observations raised without owners or due dates age silently. The condition gets covered by the following trades. Discovery arrives at handover as an NCR — or worse, in the defects liability period as a latent failure, at ten to a hundred times the early-fix cost, with a delay attached.

Safety observations logged but never trended leave repeat hazard categories invisible until an incident forces the review that would have found them.

And consultants who under-document create an evidential asymmetry that surfaces later: in a dispute, the silence in your records is not neutral. The contractor’s records will happily fill it.

How Zepth runs site observations

Observations are captured in seconds with photo and location, assigned with an owner and a due date on creation, and tracked to evidence-based closure. The escalation ladder is enforced by workflow, so repeats and overdue items climb it automatically rather than at somebody’s discretion.

Registers, trends and conversion rates are live, and every observation is linked to the inspections, instructions and NCRs it relates to. The early-warning layer and the formal QA system are one record, not two filing cabinets that disagree.