That row of big logos on your wall may be working against you — it proves you can serve large companies, and the person across from you runs a team of five.
Social proof has never worked because "lots of people bought it." It works because a few people in the same position as this buyer have already walked the path. The two sound similar and lead to opposite behaviour: the first sends you hunting for volume and famous names, the second sends you hunting for the one or two customers who look most like them, described in enough detail to be recognised. Volume is what you use to steady your own nerves; similarity is what lowers their risk. There is only one thing they want to confirm: if this goes wrong, has someone already got it wrong once and survived?
MOVE 01
Two switches: are they unsure, and do the people in your proof look like them?
Two switches: uncertainty, and similarity
descriptive normssimilarityboomerang effect
The principle in one line
Social proof is a shortcut switch, not a bonus point. It only carries current when two conditions hold at once: they can't decide on their own, and the people in your evidence read to them as their own kind. Miss one and every extra customer is just background noise; miss both and you push them away.
In their words
"We view a behavior as more correct in a given situation to the degree that we see others performing it."— Robert Cialdini, Influence (1984)
Two switches decide whether proof helps or is noise
How it plays out
The setting: the owner of a sixty-person hot pot chain asks "so who's using this?" You have two large group clients and three small chains.
✗ Leading with the loudest name
"We work with XX Group and YY Catering — over a thousand sites nationwide run our system." —— They say "impressive" and think three things: probably expensive, probably needs an IT person, probably won't care about a customer my size. You meant to show strength; what landed was "you're not for me."
✓ Leading with the one who looks like them
"The closest to you is the place on the west side — four sites, about sixty staff, the owner's wife does the roster herself. She was worried her older staff wouldn't clock in on a phone app either. The first fortnight a few of them did need hand-holding; by week three nobody was asking. Want her number? Ask her how annoying those two weeks were." —— Same size, same role, and the pothole is handed over in advance. That last line is worth a hundred logos: being willing to let them check is itself the strongest proof.
Why it works · mechanism and evidence
Why it works
Mechanism: when you're unsure, "what others do" is a cheap shortcut — but it only works on samples that are like you, because only those carry information that applies to you. And this isn't sentiment: the brain literally encodes "out of step with the group" as a worse-than-expected error signal.
Hard evidence · similarity converts straight into behaviour: Goldstein, Cialdini & Griskevicius (2008, Journal of Consumer Research) ran a randomised field experiment in a real hotel: a standard environmental appeal produced about 35% towel reuse, a descriptive norm ("75% of guests reuse their towels") about 44%, and narrowing the reference group to guests who had stayed in that same room about 49%. Same word count — only "who the people in the evidence are" changed.
Hard evidence · conformity has a neural signature: Klucharev et al. (2009, Neuron) had participants rate faces, then showed them the group's ratings. Disagreement with the group raised activity in the rostral cingulate zone and lowered it in the nucleus accumbens / ventral striatum — the shape of a reinforcement-learning prediction error — and the size of the accumbens dip predicted how far the person later shifted towards the group. Being out of step is processed as a loss. Campbell-Meiklejohn et al. (2010, Current Biology) supply the other half: ventral striatal response rises when others endorse your choice. (cross-ref neuroscience, psychology)
Hard evidence · where it fails and backfires: Schultz et al. (2007, Psychological Science) fed households their neighbourhood's average electricity use. Heavy users came down — and light users went up. Adding an approving symbol (a smiley face) removed the boomerang. Quote "the industry average" to a customer who is already doing well and you have handed them a permit to relax.
Three boundaries. One, social proof fills the "risk" cell, never the "need" cell — if they don't think this is a problem, a hundred peers won't help. Two, conformity strength is not a constant: Bond & Smith's (1996, Psychological Bulletin) meta-analysis of 133 Asch-type studies across 17 countries found it stronger in collectivist cultures and declining over the decades in US samples. Three, the first lookalike case is worth an absurd amount: in Salganik, Dodds & Watts's (2006, Science) artificial music market, merely letting participants see each other's download counts made rankings winner-take-all.
Lines that carry
Worth stealing
"The closest to you is …" — open on similarity, not on fame.
"Want her number? Ask her how annoying the first two weeks were." — dare them to check.
Similar beats impressive.
MOVE 02
A testimonial is credible because it can be checked, not because it is flattering
A testimonial is only as good as it is falsifiable
falsifiabilitytwo-sided messagesspecificity
The principle in one line
What a testimonial is worth depends on whether it could be exposed. "Really useful, highly recommend" is something anyone could write, so it equals zero. "Two hours every Wednesday before, steady at fifteen minutes from week six, and importing the timesheet was painful for a fortnight" is not something anyone would bother inventing — which is why it counts as evidence.
In their words
"Opinions are worthless."— Rob Fitzpatrick, The Mom Test (2013)
One customer, five ways to write them up
How it plays out
The setting: a long-standing customer is doing well on your product and you want a line you can put on the site.
✗ Asking for something nice
"Hi — could you write us a quick review? Just say whatever you like about us!" —— You just handed them the writing job. Usually it drags two weeks and comes back as "your service is professional and trustworthy": polite, sincere, zero information.
✓ Three specific questions, you do the drafting
"Four minutes, three questions. One: before we started, what did that Wednesday afternoon actually look like? Two: you nearly didn't sign — what worried you most? Three: what's the number now, and which week did it settle?" —— Then you write it up and send it back: "Drafted from what you said — change anything that's off, your edit is final." Asking for praise is a chore you impose; asking specific questions is work you take off them. And the answer to question two is the reason your next customer won't sign.
Why it works · mechanism and evidence
Why it works
Mechanism: persuasion follows the cost of saying the thing. An all-praise testimonial costs nothing to produce, so it carries no information. Attach a specific basis and one real flaw and you have put yourself somewhere verifiable and refutable — what the reader is really reading is "will you let me check?"
Fairly strong evidence · admitting a flaw raises credibility: Crowley & Hoyer (1994, Journal of Consumer Research) laid out the mechanics of two-sided messages; Eisend's (2006, International Journal of Research in Marketing) meta-analysis of two-sided advertising found a stable advantage on credibility, with the net effect on attitude depending on how heavy and how relevant the negative is. Practical version: one true small flaw helps, one fatal flaw obviously doesn't.
Moderate evidence · the blemishing effect has narrow boundaries: Ein-Gar, Shiv & Tanner (2012, Journal of Marketing Research) found negative information can lift overall evaluation, but only when it is minor, arrives after the positive information, and the audience is processing with low effort. Outside those conditions it reverses.
Boundary: the effect sizes here are generally modest and depend heavily on which flaw you admit. The test is plain — admit the pothole they will hit anyway (the rough first fortnight, the data migration, the month that needs someone watching). Saying it first moves the moment of discovery from "furious after go-live" to "braced before signing." Never admit the flaw that would be a legitimate reason to refuse; that isn't candour, that's something you shouldn't be selling. And buying or fabricating reviews stopped being an effectiveness question long ago — it's fraud, and one exposure burns the channel permanently.
Checklist · getting a testimonial you can use
A four-minute call, three questions, you draft
· Q1: what that specific moment looked like before (which day, what time, where it jammed)
· Q2: what nearly stopped them signing (often worth more than the praise)
· Q3: the number now, and which week it settled (a week, not "quickly")
· Draft 60–80 words and send it back: "change anything that's off, your edit is final"
· Attribute fully — name · role · company · size; size is the cell you can least afford to drop
· One more ask: "if someone like you wants to talk, could they ask you directly?"
Lines that carry
Worth stealing
"What almost stopped you from signing?" — surfaces your next customer's objection.
"I drafted it from your words — change anything that's off." — takes the work off them.
Praise is free; specifics cost something.
MOVE 03
Numbers: give the basis first, the figure second
Denominator first, number second
the basisprecision cuespre-emptive discounting
The principle in one line
A number without a basis isn't evidence, it's an adjective. "300% efficiency gain" and "really good" sit in the same bin in their head — neither can be checked, so both get discounted automatically. Saying the basis before the figure matters far more than making the figure bigger.
In their words
"The secret language of statistics, so appealing in a fact-minded culture, is employed to sensationalize, inflate, confuse, and oversimplify."— Darrell Huff, How to Lie with Statistics (1954)
The four parts of a number
Every number you say out loud carries these four
· What n is: three sites, not "many customers"
· The window: weeks 6–12 after go-live, not "over the long run"
· Compared to what: to themselves before go-live, not to an industry average
· Who measured: exported from the customer's own dashboard, or counted by you (an order of magnitude apart)
· If one part is missing, say so first: "that's an average of three, a small sample."
How it plays out
The setting: they're staring at "average efficiency gain 300%" on your slide and look up: "where does the 300% come from?"
✗ Answering with a bigger number
"That's the average across over a thousand customers — some of them saw 500%." —— You read the doubt as "the number's too small" and raised it. They asked about method; you answered with volume. From that moment they don't just disbelieve this figure, they start doubting every other figure on the page — one unsourced number contaminates all the sourced ones next to it.
✓ Expose the basis, then shrink the claim
"That number's soft — let me unpack it and you judge. Three sites, weeks six to twelve after go-live, compared with themselves before go-live: rostering went from roughly two hours to around fifteen minutes, exported from their own dashboard. Three is too few for me to call it an industry figure. You've got more sites and night shifts, so I'd expect you to land lower than that in the first two months." —— The number got smaller and the credibility went up. Claiming the small sample yourself, and predicting a worse result for them, buys more trust than 300% ever will.
Why it works · mechanism and evidence
Why it works
Mechanism: the first thing a listener does with a number isn't believe it — it's estimate how inflated it is and discount accordingly. Stating the basis does that discounting for them, so they don't have to keep a private margin, and what's left can be taken at face value. Conversely, one uncheckable number drags down every checkable number beside it.
Hard evidence · precise numbers read as more confident: Jerez-Fernandez, Angulo & Oppenheimer (2014, Psychological Science) found that people who give precise values rather than round ones are judged more confident in their answer and more credible. So "down to about 15 minutes" beats "substantially reduced." But what it buys is confidence, not correctness — which is why the thing to add is the basis, not another decimal place.
Hard evidence · precise anchors are harder to move: Janiszewski & Uy (2008, Psychological Science) found precise anchors (4,998) produce smaller adjustments than round ones (5,000). Applied to pricing and promised results: your first figure fences the range they negotiate in — and a precise but false number does proportionally more damage.
Hard evidence · ignoring the denominator is the default: numerical judgement research reliably shows denominator neglect — "9 in 100" feels riskier than "90 in 1000" even though the latter ratio is higher (the fuzzy-trace tradition of Reyna & Brainerd). If you don't supply the denominator they won't add it; they form an impression from the numerator and only discover the distortion later.
Boundary: "shrink it yourself" is a constraint, not a technique. It works precisely because most people don't do it; the moment "honestly, that figure's soft" becomes a line you deploy, it turns into another performance — and customer circles are smaller than you think. There is one test: would your stated basis survive them phoning those three sites to ask? If not, don't use the number. Not reword it — don't use it.
Lines that carry
Worth stealing
"That's three sites over six weeks, measured off their own dashboard." — all four parts in one line.
"I'd expect you to land lower than that in month one." — lower the expectation yourself.
A number without a denominator is an adjective.
MOVE 04
Write the case study as a particular day, not as a report
Write the case as a day, not as a report
A case study isn't a report card, it's a route map someone can follow. What the reader is hunting for is "which places will I pass through if I do this," so the parts to write clearly are not how good the outcome was, but how they hesitated and what went wrong on the way — those are the only two sections in which the next person recognises themselves.
In their words
"A case passes the Sinatra Test when one example alone is enough to establish credibility in a given domain."— Chip Heath & Dan Heath, Made to Stick (2007)
Six cells make one page
How it plays out
The setting: you're putting a customer story on the site and you have the full data.
✗ Written as a report card
"A leading restaurant chain adopted our intelligent rostering solution, achieving an 87% efficiency gain and significant optimisation of labour cost." —— No person, no particular day, no hesitation, no pothole. The reader can extract nothing about how to walk it, only an impression: this is an ad. And "a leading chain" is itself an announcement that nothing here can be checked.
✓ Written as his one afternoon
"Zhang Wei runs four hot pot restaurants on the west side of the city, about sixty staff, and his wife has always done the roster. Wednesdays she started at two and finished around four, with seven or eight phone calls in between to confirm swaps. He didn't want to use it at first — his words were 'my older staff will never work an app.' The first fortnight they did need hand-holding; by week three nobody was asking. From week six the roster has held at around fifteen minutes (exported from their own dashboard). He says: now I get home for dinner on Wednesdays." —— The next four-site owner reading "my older staff will never work an app" feels a jolt: that's the worry they hadn't said out loud.
Why it works · mechanism and evidence
Why it works
Mechanism: a person with a name and a visible situation recruits empathy and mental simulation; a pile of statistics recruits arithmetic. The first produces "this could be me," the second returns them to spectator. And the pothole matters because it moves the text out of the advertising category — ads don't write up their own bad parts.
Hard evidence · identifiable individual beats statistics: Small, Loewenstein & Slovic (2007, OBHDP) had participants donate to famine relief: a named, photographed girl (Rokia) drew significantly more than aggregate statistics — and, more tellingly, priming people to "think analytically" beforehand reduced donations to her. Once a reader switches into accounting mode, the story's power is switched off by the reader themselves, which is why the case and the ROI table belong on separate pages.
Hard evidence · narrative transportation shifts attitudes: Green & Brock (2000, JPSP) proposed and tested "transportation": the more absorbed a reader is in a narrative, the more they accept the beliefs implied by it, and the fewer counterarguments they generate. That's why a case is harder to rebut on the spot than an argument — rebutting means stepping out of the story, and a good story makes you not want to.
Honest labelling · one thing here has no evidence: the six-cell structure itself has no controlled trial behind it; it's an engineering assembly of the mechanisms above, and it's experience talking. Reorder or merge cells freely, but don't delete ③ and ④ — those run on the two-sided-message evidence from the previous card (Crowley & Hoyer 1994; Eisend 2006), the only part of the six with hard support underneath it.
Boundary: the identifiable individual cuts both ways — it also lets an extreme case mislead. If the customer you feature is your best result, you're using a true story to manufacture a distorted expectation, and the account gets settled three months after go-live. The safe move is to say it: "this is our best site; the median is closer to …". Cases also need explicit permission, in writing where money, headcount or internal process is involved.
Lines that carry
Worth stealing
"Here's the one that looks most like you." — the first line when picking a case.
"This is our best result — the median is closer to …" — head off the extreme-case distortion.
Write the day, not the outcome.
Your Day 19 Action
Build one case today, but build it so it can be checked.
One (10 minutes): list your existing customers by size × sector and circle the one that most resembles your next target — not the biggest one. Use the lookalike even if their result is ordinary: recognisable beats spectacular-but-irrelevant.
Two (a four-minute call): three questions — what that specific moment looked like before, what nearly stopped them signing, what the number is now and which week it settled. The answer to the second is today's real haul: it's the reason your next customer won't sign.
Three (30 minutes): write the six cells onto one page, then check two things: is cell 3 empty? Is there a real pothole in cell 4? Both have to be there.
Four (5 minutes): send it back for approval and settle permission while you're there — if you can get "sector + size + role," don't publish "a leading company," and ideally add "if someone like you wants to talk, could they ask you directly?"
Boundary: every step rests on the numbers being real, the pothole being real, the person being real. Buying reviews, inventing cases, presenting your best site as "typical" — all of it works in the short run, which is exactly what makes it dangerous.
Think It Through
1. I don't have a single customer yet. Where is social proof supposed to come from?
What you're missing isn't customer count, it's one piece of evidence that lowers their sense of risk — a case study is merely the most expensive form of that. Four substitutes. One, write up your first user as a case even if they never paid: a beta user, a friend, your own team all qualify, as long as the basis is explicit — "this is eight weeks of our own usage, we have no paying customers yet." That admission doesn't cost you anything. Two, make the process itself the proof: "I interviewed 27 restaurant-chain managers and 19 of them raised Wednesday rostering" isn't a customer count, but it's checkable, and it shows you understand this better than most vendors they've met. Three, borrow published research to explain the mechanism — say why this approach should work, then add "I don't have my own data yet." Four, substitute risk reversal: do one site, charge on results, no result no fee.
One more thing: the first lookalike case is worth going loss-making for. Salganik et al. (2006) show how a small early visible difference gets amplified into an enormous gap. Your first deal isn't about revenue, it's about buying the ticket everyone afterwards will look at — so pick not the highest bidder, but the one who looks most like the crowd you want next.
2. The customer won't let me use their name or their numbers. Can I still write the case?
You can, but know what you've lost. The name supplies verifiability, the numbers supply recomputability; with neither, you still hold the one thing hardest to fake: the resolution of the process. "A four-site hot pot chain, the owner's wife does the roster herself, Wednesdays from two until about four with seven or eight phone calls in between" contains no name and no percentage, yet anyone in the trade knows you've actually been inside a place like that. A fabricated case can invent the outcome; it can't invent that texture.
Three steps. One, walk down the anonymisation ladder rather than jumping to the bottom: push for "sector + size + role" first — most people agree to that, and it's enough for the next reader to recognise themselves. Don't settle for "a leading company"; those words announce that nothing is checkable. Two, if you can't give absolutes, give ratios: "they won't let me quote the money — what I can say is the time dropped to about an eighth." Three, move the verifiability to your side: write the method out fully enough that people can judge it themselves, or run a small pilot so they generate their own numbers. Numbers they produced beat any case you can offer.
And one step that gets skipped: ask why they're saying no. Afraid of competitors, process not finished, blanket legal policy — or actually less happy than you assumed? The last one matters most: if the real reason is "it isn't that good," what needs fixing isn't the case study, it's the product.
3. My competitors buy reviews and invent numbers. Doesn't playing it straight just lose?
Start with the unpleasant part: in the short run, yes. In markets that are one-shot, opaque and never revisited, the fakers win faster, and no moralising changes that. So the test isn't which approach is nobler, it's whether you're in a one-shot game or a repeated one. Selling souvenirs at a tourist gate is one-shot, so the market rewards faking — which is also why nobody in such a market is trusted, including the honest ones. But if your customers know each other (same city, same trade group, same investors), it's a repeated game: the payoff from faking goes permanently to zero the first time you're caught, because being caught gets retold.
Three practical moves. One, put "checkable" on the table as the selling point: others won't put a customer on the phone, you will — you don't need to accuse anyone, just say "happy to give you her number." Two, when they've been fed a diet of beautiful numbers, lower the expectation yourself: "the going claim is triple the efficiency; my estimate is you'd save about half in the first two months." Amid a wall of 300%, that doesn't sound modest, it sounds professional. Three, accept you'll lose some deals, and pick the right arena: the honest game wins where deal sizes are large, cycles long, renewals real and referrals common; it's clearly disadvantaged in one-off transactions.
One hard note: fabricated reviews and false advertising are regulated in most jurisdictions under advertising and consumer protection law — that's compliance risk, not tactical choice. The simplest test of the line between persuasion and manipulation: if they later learned the whole truth, would they feel helped, or conned?