An entire instrument category sells a measure of a quantity that does not exist. Not a quantity that is hard to measure, not one the instruments have not gotten good enough to catch yet, not one that needs a bigger sample and a better sampling frame. A quantity with no referent. Guest Experience preference is not measurable today and will not be measurable in ten years, because preference requires an object that holds still while a chooser ranks it from outside, and the Guest is a constituent of the Guest Experience. There is no outside vantage point. There is no second version sitting on the shelf to compare the first one against.
That is the whole indictment, and it does not depend on the instruments being sloppy. The instruments are not sloppy. They work exactly as designed. They just work on a different object than the one on the label.
The Instrument Has No Object
Here is the specimen. One of the experience-management vendors publishes a piece arguing the industry should move experience work upstream, out of post-visit recovery and into design. Correct instinct. Then it names the method: concept testing. Put the experience in front of a panel before you build it, read the preference, build the winner.
The thesis and the instrument contradict each other inside the same document. The thesis says move the work earlier. The instrument requires the experience to already exist. You cannot test an experience that has not been lived. The GX is the aggregate as the Guest lived it — the room, the greeting, the timing, the recovery, the read the cast member made at the table, the way the Guest felt walking out. A concept has not been lived. There is nothing to aggregate and nothing to read. The instrument has no object.
Preference is a Road 1 construct and it belongs to the [Customer]. A [Customer] is a chooser comparing spot transactions on attributes that decompose: price, speed, portion, wait, accuracy, cleanliness. Every one of those holds still long enough to be ranked, because every one of them is a property of a thing rather than a property of a relationship. Nothing relational decomposes. That is not a limitation of the survey. That is the definition of relational.
So when the instrument returns a preference reading, it has not measured a GX poorly. It has measured a [Customer] accurately. The instrument converts the Guest into a [Customer] at the moment of measurement, reads the [Customer], and hands the operator a portrait of a chooser ranking attributes — which is exactly what it was built to produce, no matter who sat down at that table. The operation then designs against that portrait. And what comes out the other end is a [Customer Experience] with a GX label on the door.
Five arguments hold that category up. Each one fails at the mechanism level, each one compounds the one before it, and each one has a move that would have worked instead.
I Have Been Saying This Forever And A Day
I have spent 44 years in this industry watching its measurement layer answer a different question than the one it asks, and watching the industry respond by buying better versions of the same instrument. That is not a failure of data literacy. It is a failure of definition, and the definition failure is upstream of every dashboard the industry has purchased since.
Under this specimen sits the study half the category quotes. A conference company that publishes market studies ran 530 adult American consumers through a self-report survey in May 2024, asked them to recall sentiment across “the past year,” disclosed no sampling frame, and put four vendor logos on the cover. Three vendor content sections run inside the table of contents — roughly twelve of forty-four pages. The study concludes that self-service is failing and that live agents need better real-time assistance, which is the product category the sponsors sell.
I am not going to spend this piece on the sponsorship. The sponsorship is the least interesting thing in the document. The interesting thing is that the study is internally honest enough to publish the numbers that dismantle its own headline, and nobody in the category noticed, because the category was reading the headline for a preference and got one.
Argument One: The Experience Can Be Tested Before It Is Lived
The dismantling. The argument says an experience is a designable artifact you can put in front of a panel, like a package or a price point. Test three versions, read the preference, build the winner. It fails because the GX is not an artifact. It is what accumulated in one Guest across one visit, produced live, and it exists only after the visit. Before the visit there is an architecture and an intent. After the visit there is a GX. There is no moment in between where a testable object sits still and waits.
What the concept test actually reads is the panelist’s prediction of their own future ranking of attributes they have been shown on a screen. That is a [Customer] read. Useful for pricing a package. Useless as a read of a GX, because the thing it predicts is not the thing that gets produced.
The cumulative impact. This is where the frame gets set for everything downstream. Once the operator accepts that the GX is an object standing apart from the Guest and available to be ranked, every subsequent instrument inherits that acceptance and no instrument in the stack ever re-opens it. The Perspective was decided at purchase. And it was decided by default — the operator never held a meeting about which road to run, because the instrument came pre-loaded with the answer. That is [The Summers Principle] operating at the instrument level: by design or by default, and the default was sitting in the vendor’s onboarding deck. [Two Roads] is a fork, not a spectrum. A better concept test does not walk you toward Road 2. It walks you further down Road 1 with more confidence.
The move that would work. If you were running my framework, the sequence reverses. Name the object before you buy the instrument. Write down, in one sentence, what you believe the GX is in your operation and who produces it. If the sentence contains the word “score” or the word “rating,” you have described a [Customer Experience] and you should stop there and fix the sentence. Then design the [Guest Production Architecture] — the arrangement of cast, sequence, judgment authority, and recovery that produces the aggregate you named — and read what happened after it happened. Design forward, read backward. Never test sideways.
Argument Two: Preference Decomposes Into Rankable Attributes
The dismantling. The argument says you can get at the GX by breaking it into components and scoring each one. This is the load-bearing assumption of the whole category, and it inverts the Product read. The GX is the Product, and the Product is the aggregate. Decomposing the aggregate destroys the object it claims to read. You do not get a GX by summing attribute scores any more than you get a marriage by summing calendar entries.
The study proves the point on itself twice, in public, on two pages.
First, the scale violation. The published line reads: “Only 7% of consumers feel experiences have gotten much better over the past year, and a troubling 55% feel they have gotten worse.” Their own distribution, on the same page: 6.60% much better, 13.77% somewhat better, 24.91% no change, 31.51% somewhat worse, 23.21% much worse. All improvement is 20.37%. All decline is 54.72%. The honest gap is 2.7x. The published gap is 8.3x. The headline compares a top box against a combined two box inside a single sentence. The vendor did not distort the study. The study distorted itself in its own headline finding, and the vendor repeated it, and the category cited the vendor.
Second, the decomposition itself. The study runs a ten-item attribute list against the prompt “Consider your typical customer service interaction over the past year. Which of the following would you say are true?” The top item is “the experience was quick, convenient, and easy” at 26.42%. “None of the above” lands at 21.70%, tied with consistency across touch points. And dead last, at 13.21%, is the one item on the list that requires a human being exercising judgment in the moment: the company or employee went “off script” to give a special resolution. The study does not remark on it. It frames the section around frictionless delivery and omnichannel consistency instead.
That is the decomposition performing exactly as decomposition performs. The relational item does not score, because relational content does not survive being broken into a list item and rated in recall. Its low score is an artifact of the instrument, not a finding about Guests.
The cumulative impact. The decomposed list does not stay a list. It becomes the design brief. That is [The Dashboard Trap] with the trap door already open: the operator reads 26.42% on “quick, convenient, and easy” and 13.21% on off-script judgment and funds the first one, because the first one is where the score is. So the operation gets faster, smoother, more consistent, and less capable of the single act the Guest would have remembered. Stack that on Argument One and the loop closes: a frame that says the GX is rankable, an instrument that ranks the parts, and a budget that follows the parts.
The move that would work. Stop scoring components. Ask one aggregate question — would you bring someone who matters to you here — and then throw the answer away as a number and use it only as a doorway to the follow-up. Then read behavior against it: return interval, party size on return, who they brought, what they ordered the second time, whether they asked for the same table or the same cast member. Behavior after the fact is readable. Preference in advance is not. That is [Relational VoG] doing what [Transactional VoG] cannot: listening, rather than scoring.
Argument Three: The Producer Is Not In The Instrument
The dismantling. Every GX is produced by a specific cast member, at a specific table, in response to a specific Guest. No instrument in the category has a field for the producer. So the portrait comes back with the producer deleted, and a portrait with the producer deleted cannot show that the producer is where the variance lives. Which is why the remedy always arrives as downstream retraining — the instrument can only see the outcome, so the fix can only be aimed at the outcome.
The study’s own People question is the proof, and it is the most damning page in the document. Prompt: “Consider the customer service employees with whom you’ve recently interacted. Which of the following do you feel is true for most of them?” The top answer, at 22.58%, is “none of the above.” Higher than “they seriously care about solving my problems” at 22.01%. Higher than knowledgeable at 19.17%, higher than focused at 19.17%, higher than thinking outside the box at 16.51%.
And then the line that should have stopped the presses: “eager to listen to my feedback and share it with their company” scores 14.04%. That is the survey category auditing its own channel and failing the audit. The instrument asked whether the human in front of the Guest carries the Guest’s read upward, and fewer than one respondent in seven affirmed it — the lowest item on the list but one. The category’s entire value proposition is that it substitutes for that broken channel. The honest conclusion is that the channel is the asset and the instrument is the workaround. The study drew the opposite conclusion.
I am going to hold myself to the standard I just applied to them. That 14.04% is a selection rate on a multi-select list, not a referendum, and I am not going to convert it into a percentage who denied the statement — doing that is the same top-box-for-two-box move I prosecuted a section ago. What it supports is narrower and still sufficient: of everything a Guest could affirm about the person in front of them, feedback-carrying is next to last, and the top answer on the whole question is that none of it applied.
The cumulative impact. Stack this against the study’s own segment data and the compounding becomes visible. Fast food, coffee, and fast casual reads as the top improving segment at 28.79%. Hotels at 24.43%. Bars and sit-down restaurants at 19.70%. The transactional segments read as improving. The relational ones read at the bottom. The study itself concedes the best-performing segment “has little reason to celebrate,” then leaves the ranking standing. An operator in a relational segment reads that table and concludes the relational operation is losing. What the table actually shows is an instrument scoring transactional attributes and correctly reporting that transactional operations produce them better. The relational operator then imports the transactional remedy — faster, smoother, more consistent, more automated — and prosecutes his own architecture on the strength of a measurement that was never pointed at it.
The move that would work. Put the producer in the read. Every GX read in your operation carries three fields before it carries anything else: which cast member, which table, which shift. Then hold a standing debrief where the cast member reports what they read at that table and what they did about it, and where that report is treated as primary data rather than anecdote. You have not added an instrument. You have restored the channel the instrument was purchased to replace.
Argument Four: The Experience Holds Still Long Enough To Be Measured
The dismantling. The argument is never stated, which is what makes it load-bearing. Measurement presumes an object that holds still. The GX is produced live, once, in front of one Guest, and does not repeat. Nothing on the stage holds still. Two Guests at adjacent tables on the same night, same menu, same cast, same timing, do not receive the same GX, and neither of them receives it twice.
Look at what the instrument did with that problem: it recalled. The prompt asks the respondent to consider a typical interaction over “the past year.” So the object being measured is not a GX. It is a year-long composite memory of a category, flattened by 530 people into an average of an average. Then that number is published as a finding about experiences, and an operator somewhere reads it as a benchmark for a room he ran last Tuesday.
The cumulative impact. This is the Performance read collapsing. Once the baseline is a recalled category average, the operator loses the ability to distinguish a bad night from a bad architecture, because the instrument reports at a cadence that cannot see nights at all. Stack it on the producer deletion from Argument Three and the operator is now blind in both directions: he cannot see who produced the variance and he cannot see when. What remains is a quarterly number and a hunch, which is precisely the condition the dashboard was bought to end.
The move that would work. Read the production while it is running. [The Read] is the aggregate discipline sitting above the three thirds of your job, and it runs on the stage, in the kitchen, in the cast, and in the numbers, in the shift, not in a report. Walk the room with one question live in your head — what is this table’s read of us right now — and log what you saw before you leave the building. Then let the lagging numbers do their actual job, which is setting next month’s plan, not diagnosing tonight.
Argument Five: The Score Decides What Gets Funded
The dismantling. Scores decompose only into transactional attributes, so capital routes to transactional fixes, and the margin architecture of the operation ends up shaped by the limits of the survey instrument. That is the Profit read, and the specimen hands over the case study.
Tipping. The study files it under the heading “Trendy CX: How Consumers Feel About Buzzy Issues,” alongside political stances and data privacy. Inside that section: 52.46% say the tip has become “more of a built-in tax” and “less about rewarding exceptional service.” 48.30% say expectations are too high. 46.21% say too many types of workers expect tips. 44.70% object to the tip screen at checkout. Only 7.58% chose “none of the above” — the one exclusive option on the list — which leaves better than nine in ten selecting at least one active grievance about the payment architecture. The study’s read: “today’s consumers are not universally furious about the state of gratuities.” Its remedy: none.
Tipping is not a trend topic. It is the payment architecture of this industry and the instrument that mediates cast compensation. A majority of the sample just reported that the instrument has flipped from reward to tax — which is a report that the mechanism connecting Guest gratitude to cast income has been severed — and the study classified it as a buzzy issue and moved on to the next chart.
The cumulative impact. Now stack it with the pain-point page. The number one pain point, at 52%, is difficulty getting in touch with a live person. Number two, at 49%, is unhelpful automated platforms. Long waits at 48%. Repeating information at 37%. Every leading complaint is a report that the human was removed and the removal hurt. The study’s conclusion is that live agents need better real-time assistance — which is to say, more instrumentation around the human rather than more authority inside the human. That conclusion is the product category the sponsors sell, and it is where the capital goes. The operator reads the study, funds the assistance layer, leaves the compensation architecture and the judgment authority untouched, and reports an improvement in the attributes the score can see.
The move that would work. Run capital allocation against the [Guest Production Architecture] instead of against the score. Before any experience spend, name which part of the architecture the money strengthens: the cast member’s judgment authority, the recovery latitude, the compensation instrument, the sequence, or the room. If the honest answer is “it improves a number,” the spend is buying a portrait, not a Product. [Architectural Coherence] is the test — every funded element points at the same aggregate, or the operation is paying twice to work against itself.
The Deeper Argument: The Measurement Act Is The Road Switch
Underneath all five arguments sits one mechanism, and it is the reason this category is not fixable from inside.
The operator who runs the instrument has already chosen Road 1 before reading a single result. Not when he acts on the data. At the moment of measurement. Because the act of measuring preference requires the Guest to be positioned outside the object, ranking it, and the only entity that can occupy that position is a [Customer]. Running the instrument converts the Guest into a [Customer] as a precondition of getting a reading at all. That is [Guest As Input Not Reference] in its cleanest form: the Guest enters the system as raw material for a score rather than as the reference against which the operation is designed.
So the road got switched, and no one was in the room when it happened. There was no strategy session, no debate about relational versus transactional positioning, no decision the operator could later point to and defend. There was a procurement decision and a login. [The Summers Principle] says the operation is designed by design or by default, and this is the purest default in the industry: the road chosen by an instrument the operator believed was neutral.
Watch what happens when a category that has made that switch tries to define the destination. The study asserts, on the strength of “its fifteen-plus years of research,” that “the ideal customer experience” is one that is “simultaneously frictionless, personalized, predictive, and consistent throughout the omnichannel journey.” Read that again as a claim about a restaurant. Frictionless is not the ideal in a relational operation — the right friction at the right moment is a production tool, the pause that makes a moment land, the deliberate beat before the recommendation, the hand-sell that takes eleven seconds longer and changes both the check and the memory. Predictive is not the ideal either, because prediction requires the Guest to be a repeatable pattern rather than a person at a table tonight.
And notice what kind of statement that is. A conference company published a category definition as a research finding. That is not measurement. That is a vendor-adjacent business asserting the shape of the destination and then selling the road to it. The definition is upstream of every number in the document, and it was never tested, because it is not the kind of claim their instrument can test.
Three lenses on it, since this is where an operator has to actually decide. The devil’s advocate says the instruments still catch real failures — long waits, cold food, broken automation — and he is right, and that is precisely the tell: those are the transactional attributes, and the instrument is excellent at its actual job. The operator read says the board wants a number next quarter and no board has ever accepted “the GX is not a measurable quantity” as a status report. The third position is the only one that holds: report the number as what it is, a [Customer Experience] index, and report the GX separately as architecture and behavior — what you designed, who produced it, what the Guests did afterward. One page, two objects, honestly labeled. The board does not need a fake measure. It needs to know which object it is looking at.
What You Do Monday Morning
Pull the last report your experience instrument produced. Take the top three items by score and the bottom three, and against each one write a single word: attribute or aggregate. You will find every top item is an attribute and the aggregate items are at the bottom or absent. Then open your next scheduled experience spend and check which of those six items it is aimed at. If it is aimed at an attribute — and it will be — cancel the line item as written and re-aim it at one element of the [Guest Production Architecture]: judgment authority at the table, recovery latitude, the compensation instrument, or the sequence. One spend, re-aimed, before the week starts. That is the switch back, made on purpose this time.
The Closer
The instruments are not broken and the vendors are not lying. The category built an excellent measure of a [Customer] and mislabeled it a measure of a Guest, and the industry bought the label. Every operator who runs it converts the Guest into a chooser to get the reading, designs against the chooser’s portrait, produces a [Customer Experience], and files it as a GX. There is no version of that sequence that ends in a relational operation, no matter how good the instrument gets, because the thing it claims to read was never there. There is no such thing as Guest Experience preference. There is only what you designed, who produced it, and what the Guest did next.
To understand the ideal state, go to Restaurant Physics.
Digging Deeper — The Framework Adoption Engine
Positions on the record.
-
What The Guest Experience Actually Is — https://physics.jeffreysummers.com/what-the-guest-experience-actually-is
-
The Summers Principle — https://jeffreysummers.com/read-the-summers-principle/
-
The Right Friction At The Right Moment — [URL TO CONFIRM]
-
The Dashboard Trap: Why Your Numbers Got Better And Your Room Got Worse — [URL TO CONFIRM]
-
Voice Of The Guest Is Not A Survey — [URL TO CONFIRM]
-
By Design Or By Default: The Instrument Decides For You — [URL TO CONFIRM]
Term definitions from the Knowledge Base.
-
[Guest Experience] — https://kb.jeffreysummers.com/guest-experience
-
[Customer Experience] — https://kb.jeffreysummers.com/customer-experience
-
[Customer] — https://kb.jeffreysummers.com/customer
-
[Transactional VoG] — https://kb.jeffreysummers.com/transactional-vog
-
[Relational VoG] — https://kb.jeffreysummers.com/relational-vog
-
[Guest As Input Not Reference] — https://kb.jeffreysummers.com/guest-as-input-not-reference
-
[The Summers Principle] — https://kb.jeffreysummers.com/the-summers-principle
-
[Two Roads] — https://kb.jeffreysummers.com/two-roads
-
[The Dashboard Trap] — https://kb.jeffreysummers.com/the-dashboard-trap
-
[Architectural Coherence] — https://kb.jeffreysummers.com/architectural-coherence
-
[Guest Production Architecture] — https://kb.jeffreysummers.com/guest-production-architecture
-
[The Read] — https://kb.jeffreysummers.com/the-read
