Process Defects With No Observer: Why They Survive for Years
When we broke down commercial-proposal preparation into recurring defects, we found six. No one inside the company noticed four of the six. Not "rarely noticed," not "noticed late" — not noticed at all, because there was no report, no role, and no moment at which those defects were ever supposed to surface.
A defect with no observer never enters the statistics, never comes up at a planning meeting, and never affects anyone's bonus. That's exactly why it survives for years and costs more than all the other defects combined. What follows is about how to find defects like this and how to assign them an observer — not about how to fix ones you've already found.
A process defect gets fixed only after it acquires an observer — a person or a counter obligated to see it. Defects visible only to the client have no observer by design: the client almost never complains, they just stop responding. So the defect map has to be built not from complaints, but from the points where a piece of work crosses outside.
What counts as a defect's observer
An observer isn't whoever suffers from a defect — it's whoever the defect reaches as a signal that can't be ignored. The distinction matters. A calculation error has an observer: accounting reconciles the numbers and eventually runs into a total that doesn't add up. A document sent without the responsible rep's contact details has no observer: the client, with no one to ask, doesn't write in about it — they simply stop replying.
That gives us a working rule: a defect exists for an organization exactly to the extent that it produces a signal inside it. Everything else isn't a problem — it's background noise.
Observed through the ledger. The defect breaks a reconciliation, and whoever reconciles the numbers finds it. These defects get fixed fastest, because they have both an addressee and a deadline — the end of the month.
Observed through escalation. The defect is painful enough that the client picks up the phone. The signal arrives loud and rare, so these defects get fixed, but usually one at a time rather than as a class.
Observed through a counter. The defect doesn't generate a complaint, but it does generate a measurable event: an email that was never opened, a document that never got assembled, a field left blank. This is the cheapest and most underrated type of observer, because a counter is set up once and runs without any people involved.
No observer. The defect is visible only from the outside and leaves no trace inside. Inconsistent document formatting, random illustrations, the tone of correspondence, a template-y feel — the company learns all of this about itself last, if it learns it at all.
| Defect type | Who observes it | Speed of fix |
|---|---|---|
| Breaks a reconciliation | Ledger | By the end of the reporting period |
| Painful enough to prompt a call | Client escalation | One at a time; the class stays unfixed |
| Produces an event in the system | Counter | Immediately, once the counter exists |
| Visible only to the client | None | Never |
The practical takeaway from the table: improving process quality is, first and foremost, about moving defects out of the last row and into the third one. Not "be more careful," but put a counter where there's currently silence.
Why complaints are the wrong foundation for quality control
Building quality control on client complaints is tempting: complaints are free, specific, and arrive on their own. The problem is that a dissatisfied client's silence isn't the exception — it's typical complaint behaviour, and it's been studied.
A client's propensity to complain is explained not just by how severe the incident was, but by a stable personal trait — the willingness to seek redress. Some dissatisfied customers stay silent because they "do not believe that complaining will help, or lack the time and energy for it."
I'll flag the limitation up front: the sample is deliberately balanced roughly fifty-fifty by design, so it doesn't measure the real share of silent customers in the population — it compares two groups against each other. For our purposes, that's enough: what matters is that silence is explained by a personal trait, not by how severe the incident was — meaning you can't judge how often defects occur from the volume of complaints you receive.
I'll also address a figure you've probably seen: "96% of dissatisfied customers don't complain." It circulates through presentations and blog posts citing a study from the mid-1980s, but I haven't been able to trace a primary source describing the method or the sample. So I don't use it. A claim without a verifiable method doesn't become true by being cited often.
The gap between what's visible inside and what's visible outside
There's recent data on exactly how large this gap is, and it measures precisely the discrepancy we care about: executive perception versus client perception.
Roughly 9 in 10 executives believe their customers' loyalty has grown — and only 4 in 10 consumers agree. Meanwhile 52% of consumers have already dropped a brand over a bad experience, and 83% of surveyed executives admit they lack the tools to measure what actually drives purchases.
This is a corporate study from a consulting firm, not peer-reviewed research: the methodology is described, but the instrument and raw data aren't published, and the sample is limited to the U.S. Even so, the direction of the gap confirms the mechanics: wherever there's no observer, executive confidence fills the vacuum.
Why a defect found late costs more
There's a second multiplier on the cost of an unobserved defect — time. The later a defect is found, the more it costs to fix, and this has been measured in engineering practice.
The cost of fixing a defect grows as it moves through the lifecycle: per the engineering data compiled in the report, the gap between fixing a defect at the requirements stage versus at the in-service stage reaches hundreds of times over. The report estimates the U.S. economy's aggregate losses from inadequate software testing infrastructure at $22.2–59.5 billion a year.
The specific cost multipliers in this report aren't the authors' own measurements — they're figures compiled from older engineering work, and that's worth keeping in mind. The "later means costlier" relationship itself still holds regardless: a defect no one observes is, by definition, discovered at the latest possible stage — the single most expensive point on the scale.
How to map defects without relying on complaints
The method is simple and needs neither a platform nor a budget. It's built not around the client but around handoff points: the places where a piece of work moves out of one person's responsibility and into another's, or outside the company entirely.
The last node in the diagram is the most important and the least obvious. A client's silence has to be an event in the system, not the absence of one. As long as "no reply" looks like an empty cell in a record rather than a line in a report, no one can tell disinterest apart from an email that never arrived in the first place. That's the same territory covered in the piece on why automation pays off at the gaps, not in the quality of the writing.
Four steps that take one week
Pull your last twenty outputs
Twenty sent documents, closed tickets, shipped orders. Not a curated sample of "good" ones — the last twenty in a row.
Flag the defects and name an observer
For each defect, one question: who inside the company was supposed to see it. An empty cell in that column is the finding.
Put a counter on the outward handoff
Was the document assembled or not, delivered or not, opened or not, answered or not. Four boolean fields close most of the blind spot.
Assign an owner to the empty cells
Not "improve quality" — a specific person and a specific number they check once a week. Without that, the counter becomes just one more report no one opens.
If the "who notices this" column says "no one," the defect will not get fixed — no matter how obvious it is.
Frequently asked questions
How do you find process defects that no one reports?
Not through employee surveys and not through complaints, but by reviewing the last twenty outputs and by examining the handoff points where a result changes owners. For every defect found, fill in a column: "who was supposed to see this." Defects with an empty column are the ones you're looking for: they don't get fixed not because they're hard, but because they don't exist for the organization.
Can process quality be measured through customer satisfaction?
Partly. Satisfaction metrics show the outcome, but not what produced it, and they lag by months. Managing a process requires events inside it: was the document fully assembled, was it delivered, was it opened, was there a reply. Those four facts give you more to work with than an annual survey does.
What do you do if only the client sees a defect, and they stay silent?
Turn the defect into a measurable event. Inconsistent document formatting isn't measured directly, but the share of documents assembled outside the template is. Missing contact details don't generate a complaint, but they do generate an empty field. Almost every "soft" defect leaves a hard trace in the system — the task is to make that trace start counting.
How representative is the "four out of six" ratio?
This is a breakdown of one process at one company, not an industry statistic. What's representative isn't the ratio — it's the fact that it was produced at all: most organizations never compile a list of their defects, so the question of an observer never even comes up. The exercise itself takes a few hours and usually turns up more than leadership expects.
The ratio "four defects out of six with no internal observer" comes from a breakdown of the commercial-proposal preparation process in Alego.Digital's sales department; the list of defects and their consequences comes from the company's own product description (a commercial-proposal generator). These are the author's company's internal materials; they have not been independently verified and are presented as an illustration of the mechanics, not as an industry benchmark. The four-type classification of observers is the author's own generalization from practice, not a borrowed methodology.
- Chebat J.-C., Davidow M., Codjovi I. Silent Voices: Why Some Dissatisfied Consumers Fail to Complain. Journal of Service Research, May 2005. journals.sagepub.com
- PwC. 2025 Customer Experience Survey (5,511 consumers and 406 executives, May–June 2025). pwc.com
- RTI International for NIST. The Economic Impacts of Inadequate Infrastructure for Software Testing, May 2002. rti.org