visible upon breakdown
You pay a company every month to keep your systems running. The invoice arrives, the number matches last month’s, and the honest answer to what did they do for that? is that you do not know. The months they earn it look exactly like the months they don’t. The one morning the arrangement becomes legible is the morning something is broken — which is to say, the only time you see your provider clearly is the moment they have failed you.
That is not a billing problem, and it is not a communication problem. It is the defining property of the thing you bought.
The claim
Thesis. You cannot make a service material. You can only choose which indicators of it the buyer gets to see — and the only indicators worth choosing are the ones the buyer can check without asking you, because everything else can be counterfeited more cheaply than it can be earned.
Four steps:
- Work that functions is invisible. Not under-marketed. Invisible as a structural property, written down in 1999 with “the server is down” as the example.
- So visibility is never a fact about the work. It is a set of indicators somebody selected. You are selecting it whether or not you admit you are.
- Software makes indicator sets very cheap to produce — and the cheapest indicator to manufacture is the appearance of effort, which is the one thing a counterfeit does as well as the real article.
- Cheap indicator sets fail in a specific, named way: they substitute precision for validity. A number to two decimal places, answering a question nobody asked.
Scope. This is an argument about one design move, not a theory of trust. I am not claiming visibility produces accountability — Ananny & Crawford opened their 2016 paper by putting exactly that equation in the dock: “Being able to see a system is sometimes equated with being able to know how it works and govern it.” They argue it should not be. I have read their abstract and not their paper, so I will state that the fight exists and decline to referee it.
1. Invisible because it is working
Susan Leigh Star, The Ethnography of Infrastructure, 1999, listing the properties of infrastructure:
Becomes visible upon breakdown. The normally invisible quality of working infrastructure becomes visible when it breaks: the server is down, the bridge washes out, there is a power blackout. Even when there are back-up mechanisms or procedures, their existence further highlights the now-visible infrastructure.
She was not thinking about managed services. She wrote the managed-services problem anyway, twenty-seven years ago, with the exact example — the server is down — and in one sentence she does something harsher than the usual complaint that services are hard to see. Working infrastructure is not merely unnoticed. Visibility is the signature of failure. The provider who never surfaces is either doing the job perfectly or not doing it at all, and from the invoice those two look the same.
The trade knows this about itself, in its own flatter words. A vendor handbook for MSPs: “If there’s no reason to contact their MSP, many SMBs won’t. They’re satisfied as long as the business continues to run.”
What the trade does not know about itself is the thing it repeats most confidently. The line you will hear is that clients leave because they couldn’t see the value — a consultancy blog states it flatly, adds that “half your clients can’t tell you what you did for them last quarter,” and offers no survey, no sample and no method. I went looking for the study behind it across four rounds of searching. There is no study. Nobody has published a survey of managed-service clients asking why they left. The industry’s most sophisticated benchmarking operation counts departures and prices them by profit quartile, then advises handling each loss “on a case-by-case basis” — which is a reasonable thing to tell an owner and also a good explanation of why there is nothing to aggregate.
The numbers that do exist are one vendor’s surveys of its own audience — no sampling frame, no response rate, no margin of error, and the four reports people cite as corroboration are four reports from the same corporate parent. Read with that discount, they still say something inconvenient. Asked what they expect to struggle with, those providers put retaining current customers at 7%, and acquiring more customers at 36%. The one finding in the whole pile that resembles this post’s thesis — inability to quickly demonstrate value to clients, nearly doubling from 10% to 19% — sits under a chart headed “Biggest challenge when acquiring new customers.” It is a sales finding. Moving it quietly one stage down the relationship, to describe a client who already pays you, would be exactly the kind of thing this post is against.
So I am not going to argue that invisibility is costing the industry its retainers. I don’t know that, and neither does anyone quoting that line at you. I am going to argue something narrower and better supported: that the invisibility is real, that it is structural, and that the standard remedy for it is built wrong.
2. 1977 already answered the easy half
The instinct — put it on a screen, make the invisible thing visible — is correct and it is old. G. Lynn Shostack, Journal of Marketing, April 1977:
Intangibility is not a modifier; it is a state. Intangibles may come with tangible trappings, but no amount of money can buy physical ownership of such intangibles as “experience” (movies), “time” (consultants), or “process” (dry cleaning). A service is rendered. A service is experienced. A service cannot be stored on a shelf, touched, tasted or tried on for size.
Then, decades before there was a screen to put anything on, she writes the instinct and denies its mechanism in the same breath:
But a service is already abstract. To compound the abstraction dilutes the “reality” that the marketer is trying to enhance. Effective service representations appear to be turned 180° away from abstraction.
Her prescription is not de-abstraction. It is the management of evidence — peripheral clues the buyer reasons backwards from, because “service ‘reality’ is arrived at by the consumer mostly through a process of deduction, based on the total impression that the evidence creates.”
Star and Anselm Strauss put the same correction in sharper form in 1999: “No work is inherently either visible or invisible. We always ‘see’ work through a selection of indicators: straining muscles, finished artifacts, a changed state of affairs.” The indicators shift with context, and that shift, they write, “becomes a negotiation about the relationship between visible and invisible work.”
That costs the idea its novelty and improves it. There is no material version of your service waiting to be uncovered. There is a set of indicators, and somebody picked it. Right now most providers have picked: an invoice, a ticket confirmation, and an outage.
And Star and Strauss add the line that turns this from a marketing observation into a software problem. As industrial practice shifts, they write, these negotiations “require longer chains of inference and representation, and may become solely abstract.”
3. Software makes indicators cheap
Here is the step that is mine rather than theirs, and I’ll mark it as mine: software drove the cost of producing an indicator to roughly zero, and it did not drive down the cost of doing the work. A straining muscle is expensive to fake because it is the work. A green tile is a SELECT and a CSS class. When the two inputs to Star and Strauss’s negotiation move that far apart in price, the indicator set stops being a representation of the work and becomes a product in its own right — built by a different team, on a different schedule, to a different spec.
The research literature has a name for the effect this exploits, and — this is the part that should make anyone reach for their wallet-hand — the same researchers who established the effect also established that the fake version works.
Ryan Buell and Michael Norton, Management Science, 2011, named it operational transparency and named its shadow the labor illusion. Across five experiments on simulated travel search and online dating:
when websites engage in operational transparency by signaling that they are exerting effort, people can actually prefer websites with longer waits to those that return instantaneous results — even when those results are identical.
Identical results, longer wait, preferred. You perceive effort; effort triggers reciprocity; reciprocity raises what you think the thing is worth. And then, same paper:
even when the actual operations might take much less time, providing consumers with the illusion of labor can still serve to increase value perceptions, provided participants believe that they are seeing the website hard at work.
To their credit, immediately after: “the fact that firms can induce the labor illusion does not mean that they therefore should induce it.” But the economics are now stated plainly. Faking it is cheaper than doing it and scores nearly as well on everything anyone measured. There is a whole design practice built on that finding, and it is not a fringe one — the spinner that spins a little longer than it needs to is in software you used this morning.
(A terminology warning, because both literatures are in this post. For Star, “transparent” means unnoticed — infrastructure “is transparent to use, in the sense that it does not have to be reinvented each time.” For Buell and Norton, transparency means deliberately shown. Same word, inverse conditions. I have tried to say which one I mean every time it matters.)
One honest note about scope, and then I’ll stop litigating it. The measured effects in this literature come from short interactions: web searches, a dining-hall grill station, a city service-request app. A managed-service relationship runs in months. Whether the perception effect survives that trip has not been shown, and I am not going to assume it does — which is why nothing in the rest of this post rests on it. The argument from here runs on what an indicator is, not on how much it moves a survey score.
4. The failure mode has a name
Star gave it one, in the same 1999 paper:
A common example is the substitution of precision for validity in the creation of a system of indicators or categories.
That is the whole disease in eight words. An MSP dashboard reporting SLA compliance to two decimal places is precise. Whether it is valid — whether that number is about the thing you actually bought — is the one question the dashboard cannot answer, and it is the only question you have.
Star is blunter still about why indicator sets drift: “When large epistemological stakes are at issue in the development of a system, one political tactic is to focus away from the larger question, and instead to seize control of the indicators.” Nobody has to be dishonest for this to happen. The indicator set gets built by whoever is closest to the telemetry, gets tuned against whatever is easiest to turn green, and the larger question — is this arrangement worth what I pay for it — quietly stops being represented at all.
The field results are consistent with this being hard rather than easy, and I’ll offer them as colour rather than proof. Buell, Porter and Norton ran operational transparency on Boston’s service-request system, the app where residents report potholes and broken street lamps. Residents who got photographs of the city fixing what they reported filed 60% more requests over the following thirteen months, and a site visualising requests and responses left residents 14% more trusting and 12% more supportive of government (DOI 10.1287/msom.2020.0877). But the same study has an arm nobody quotes: residents shown the growing backlog of requests the city was failing to fulfil were “no more nor less trusting and supportive of government than residents who received no transparency.”
Transparency about work that isn’t happening bought exactly nothing. Not a collapse in trust, which is the cynic’s prediction — nothing at all, which kills the evangelist’s. Add the 2011 paper’s own boundary conditions — “when outcomes are unfavorable, increasing operational transparency has negative effects on customer value perceptions”, and “there are many processes that are inherently unappealing or visibly inefficient due to poor design” — and the pattern is not “transparency is good.” It is closer to: transparency converts an existing reality into a perception. Real work, it pays. Poor work, it costs. Absent work, it is inert. That reading is mine; none of the papers writes it in that form.
Which is a fine description of a measurement instrument and a terrible description of a sales instrument. It only pays out if the thing underneath is already there.
5. The test: can the client check it?
If the counterfeit scores as well as the real thing on perceived value, then perceived value is the wrong target. You need a property the counterfeit cannot have.
There is one. The artifact has to be checkable by the person receiving it, without going through you.
That is what the pothole photograph has, and it is why I keep coming back to it — not for the 60%, but for the mechanism. A resident can walk down the street and falsify it in two minutes. An artifact the customer can independently verify is expensive to fake in precisely the way the work itself is expensive, which is the only reason the evidence carries information at all.
A patch report they can diff against their own asset inventory. A restore they watch come up from last night’s backup, on their hardware, in front of them. A log with timestamps that cross-reference something they already hold. Those are receipts.
A green dashboard is not. Neither is an activity feed, a status page nobody can audit, or a ticket count. They are claims about work, and they render identically whether or not the work happened. They are precision standing in for validity, on a fifteen-second refresh.
This is where I part company with the trade’s standard remedy. The quarterly business review is a narration of work: the provider assembles the story, and the client’s only means of checking it is to ask the narrator. Against the test above, a QBR deck sits closer to the spinner than to the photograph. That is my inference, not anybody’s finding, and I hold it at medium-high confidence — but I note that I went looking for a single empirical study connecting status pages, audit logs or SLA reporting to trust or perceived value, and found none. One search pass, not a systematic review. Still: the practice is enormous and the evidence under it is close to empty.
And one more caution, because this is where the argument could get too pleased with itself. A vendor report I read while researching this glosses its own survey with the sentence “buyers want proof they can verify.” That is my thesis in their voice, and I am not going to quote it as support: no item in that survey measured verifiability. It is an interpretation, and it happens to agree with me. Agreement is not evidence, and an argument that only notices the data agreeing with it is a dashboard.
What survives
Star was right that working infrastructure is invisible and that breakdown is what makes it appear. Shostack was right that you cannot hand someone a service, only the evidence around it. Star and Strauss were right that the evidence is a selection, which means somebody is always choosing. Buell and Norton were right that showing the work changes what it is worth — and the same paper proves a convincing imitation collects most of the same credit, which is why their result is a finding and not a recommendation.
What is left is smaller than the original instinct and much harder to fake: stop trying to make the service visible, and start producing artifacts the client can check without you in the room. Most of what gets built in the name of transparency fails that test. The city of Boston passed it with a photograph.
A dashboard is a claim. A receipt is a claim someone else can go and check. Only one of them survives being checked.