The Pregnancy Prediction Story
The most repeated anecdote in consumer marketing survives in its original document, which is what makes it traceable. The magazine article sources the angry father to one unnamed employee who was present; the book, published the same year, drops that clause and adds a figure the article never printed. The real psychology sits underneath the story, in a field experiment that found a window about three months wide.
Answer to the question
Can habit and purchase data predict a pregnancy?
No published study reports accuracy or precision for predicting pregnancy from retail purchase records, searched four ways. The nearest real test, on about 132,000 bank customers and roughly 60 million transactions, beats a random baseline on four life events including the birth of a child and reports no precision or recall at all.
Key takeaways
- The 2012 magazine article does source the angry father, to one unnamed employee who took part in the conversation, and the book published that year drops the clause with no endnote.
- That same article contradicts itself four paragraphs apart, because the camouflage programme it describes existed to avoid sending the visibly baby-themed mailer the anecdote turns on.
- The eighty-seven per cent figure is introduced as a hypothetical about a fictional shopper with the word say in front of it, and the book strips all three hedges and adds a second case.
- No published study reports accuracy for predicting pregnancy from purchase records, and the one field experiment isolating the habit disruption window found a small effect confined to the first three months.
In 1984 a marketing academic at UCLA, Alan Andreasen, published a paper arguing that people change what they buy when their lives change. In 2002 a statistician named Andrew Pole joined Target. In October 2010 he described his work in a conference keynote that drew little attention. On 16 February 2012 the New York Times Magazine published Charles Duhigg's "How Companies Learn Your Secrets", and the same day a Forbes headline turned one anecdote inside it into the most repeated claim about habit and purchase data in consumer marketing. Later that year a book retold the same scene without that attribution. The story has decayed since, and every step is still on the page.
Where did the retailer story begin?
Andreasen's paper is titled "Life Status Changes and Changes in Consumer Preferences and Satisfaction". Duhigg calls its conclusion a pillar of modern marketing theory and quotes Andreasen's phrase that at such moments customers are "vulnerable to intervention by marketers". We confirmed the citation, the sole authorship and the digital object identifier at Crossref, and have read neither the full text nor the abstract. That is the premise the whole lane rests on, and its published title is about preferences and satisfaction rather than marketing opportunity.
The book adds detail the article lacks: telephone surveys of almost three hundred people around Los Angeles, with roughly a tenth switching toothpaste brand inside six months. Those figures are unverified by us.
Pole joined in 2002, and the request reached him from the marketing department, in the question that opens the book's seventh chapter: can your computers figure out which customers are pregnant, even if they don't want us to know? He built a model on about twenty-five products that, analysed together, let him guess a trimester, estimate a due date and give a shopper what the chapter calls a pregnancy prediction score.
In October 2010 Pole gave a keynote on getting more from guest data. We have not viewed that video, which is the earliest public account of habit and purchase data at this retailer by the model's builder.
What did the 2012 article actually source?
Six people are named and quoted in the magazine piece, Pole at length among them, and his two hedges are the most careful sentences in the affair. A catalogue congratulating someone on a first child they have never mentioned "is going to make some people uncomfortable", he says, and "even if you're following the law, you can do things where people get queasy". Three claims in the article are attributed to nobody nameable, and those three are the habit and purchase data claims everybody repeats.
The eighty-seven per cent score arrives inside an explicit frame. One employee the reporter spoke to provided what the article calls "a hypothetical example" about "a fictional Target shopper", and the figure is introduced with the word "say" in front of it. The camouflage practice comes from "one Target executive". The angry father at the service desk is attributed, once, to "an employee who participated in the conversation". That sentence reaches us from a complete saved copy of the published page, since the nytimes.com original is metered.
The anecdote is therefore sourced in the original, to exactly one unnamed person present. Read at its strongest it reaches us as one employee's recollection of a counter conversation, relayed to a reporter, in which the pregnancy rests on words the father is reported to have said on a telephone call days later. Nobody in that chain is identifiable and nothing in it is documentary. We assert nothing about either private individual, including that any of it happened.
Duhigg printed the retailer's written reply, that almost all his statements contained inaccurate information and that the company would not address them point by point. He also recorded being turned away in Minneapolis.
Why does the article contradict itself?
Four paragraphs separate two passages that cannot both describe the same mailing. The anecdote's envelope "contained advertisements for maternity clothing, nursery furniture and pictures of smiling infants". The unnamed executive then explains that the company mixed in offers it knew pregnant women would never buy, a lawn mower beside diapers, wineglasses beside infant clothes, so baby offers looked random, because "as long as we don't spook her, it works". A mailer announcing itself as a baby mailer is the one output that camouflage programme existed to prevent.
Eric Siegel, the conference chairman quoted by name in that article, makes the point from inside the field: material built that way would "by design not raise any particular attention of the teen's father". He also discloses his own part, a fact about his conduct he volunteered. He invited Pole to keynote in October 2010, Duhigg interviewed him months afterwards, he raised pregnancy prediction, pointed the reporter at the little-watched video and introduced the two men.
Siegel's published verdict is that the piece alleges "an anonymous story" carrying an "unsubstantiated but tacit implication" about the retailer's analytics project. His reading of the article's tone is a judgement we report rather than adopt. The contradiction he identifies is different in kind, because it sits in a public document anyone can check.
We read his passage as excerpted in a 2014 KDnuggets post rather than in his book. The same post is where we take the Forbes headline, "How Target Figured Out A Teen Girl Was Pregnant Before Her Father Did", written by Kashmir Hill the day Duhigg's piece appeared. Duhigg's own headline was "How Companies Learn Your Secrets". The sentence everybody quotes about habit and purchase data is a headline by a second journalist about a first journalist's work, and neither the Forbes piece nor the Financial Times article quoted later was opened.
What did the book change in 2012?
The removals between the magazine and the book of the same year are listable, and they run one way.
The first removal in the chain is not between an author and his retellers. It is between one author's two texts.
| Magazine, February 2012 | The Power of Habit, 2012 | |
|---|---|---|
| The father at the desk | "according to an employee who participated in the conversation" | plain narrative, no attribution, no endnote, page 196 |
| The eighty-seven per cent | a hypothetical example, a fictional shopper, the word "say" | hedges gone, plus a second case at ninety-six per cent |
| Size of the mailing list | "tens of thousands of women" | "hundreds of thousands of women" |
| Revenue | 44 billion dollars to 67 billion, 2002 to 2010 | 44 billion dollars to 65 billion, 2002 to 2009 |
| Pole's department | Guest Marketing Analytics, dozens of members | Guest Data and Analytical Services, fifty members |
That second score has no counterpart in the article at all. The chapter's general footnote records interviews with more than a dozen current and former employees, many not for attribution, and nothing in the apparatus keys to the counter scene. This states what two texts do, not what anybody intended. Set against it, Duhigg disclosed his fact-checking method on a dedicated page, printed the retailer's denial where it bites, quoted Pole's discomfort twice, and recorded that no source saw the complete text.
This forces a correction on our own record. Our Borrowed Certainty note treated the 67 billion dollar figure in later retellings as drift from the book's 65 billion. It is not drift. The larger figure is Duhigg's own, in his own article, for a different date range, and the causal shape the retellers added has a precedent in the article's unsourced sentence that mother and baby sales exploded soon after the campaign began. The first removal in this chain of habit and purchase data claims is in the author's own second text.
Can habit and purchase data predict pregnancy?
No published study reports accuracy or precision for predicting pregnancy from retail purchase records. We searched four differently worded ways and found journalism, a Groningen thesis on European fertility citing this story as motivation, and a teaching exercise on a synthetic dataset balanced five hundred against five hundred. That is an absence of published evidence, weaker than evidence of absence, and it is this section's finding.
The nearest real test is De Caigny, with Coussement and De Bock, on a large European financial services provider: about 132,000 customers and roughly 60 million debit transactions across one year, predicting four life events, among them the birth of a child and a move. Fine-grained records beat a random baseline, improve further alongside aggregated customer data, and recency, frequency and monetary features outperform simple binary logs. We read the abstract and repository record only, and that abstract reports no precision, no recall and no base rate.
Life events are predictable above chance from habit and purchase data at industrial scale. The womb-reading machine of the anecdote has no published counterpart.
The base rate decides how an eighty-seven per cent figure can be read. The National Center for Health Statistics puts the general fertility rate at 53.8 births per thousand women aged fifteen to forty-four, about five and a half in a hundred giving birth in a year. Since a pregnancy runs about nine months, roughly four in a hundred women of childbearing age are pregnant at any moment, and for a girl of school age the rate of 12.6 per thousand puts it under one in a hundred. This arithmetic is ours.
Moving a four per cent prior to an eighty-seven per cent posterior needs a likelihood ratio around 160, which at a generous one-in-two detection rate means about three false positives in a thousand, a specificity near ninety-nine point seven per cent. None of that shows the model was weak. A basket holding unscented lotion, zinc supplements and a sudden volume of cotton balls is a strong signal, and a retailer's denominator is shoppers with a purchase history rather than the population.
What it shows is that the quoted figure cannot be a routine per-shopper probability over a whole file, which the word hypothetical already said. The statistician Kaiser Fung reads the camouflage the same way, quoted in the Financial Times by way of that 2014 post: offers were mixed because the company knows many coupon books go to women who are not pregnant after all. Inference from a psychological profile is a separate question, which [what psychographic targeting can do](https://thewhyweproject.com/articles/what-psychographic-targeting-can-do) covers. A profile guesses who you are; habit and purchase data guesses what just happened to you, and only the second carries a deadline.
A profile guesses who you are; habit and purchase data guesses what just happened to you, and only the second carries a deadline.
How wide is the disruption window?
Here the evidence is real and the finding is modest. Verplanken, with Roy, ran the only field experiment contrasting disrupted people with undisrupted ones instead of assuming the contrast. In Peterborough his team cold-contacted 1,612 people at the doorstep and 800 agreed, half of them known to have moved house within six months and matched to non-movers on house size, home ownership, recycling facilities and transport access, with randomisation clustered geographically so neighbours stayed together.
The intervention was a doorstep interview, a bag holding eco-washing liquid, seeds, a bus timetable and a shower timer, a booklet and two newsletters. Of the 800, 521 completed the eight-week follow-up, reporting how often they performed twenty-five environment-relevant behaviours. The predicted interaction appeared, small and in the right direction: recent movers moved clearly, the average participant slightly, the long-settled not at all. Split by weeks since moving, the intervention worked only in the group under three months past their move.
The authors state the limits of their own result plainly, and it deserves quoting rather than paraphrase:
> "The effect size of the extent to which relocation boosted intervention effects was small. More than anything else, the results should be considered as 'proof of concept'."
They also flag self-report, call the six-month cut-off defining recent arguably arbitrary, and argue against blanket roll-outs. How long a replacement behaviour takes to set is a separate measurement, which [how long a new habit really takes](https://thewhyweproject.com/articles/how-long-a-new-habit-really-takes) handles.
Their lineage reaches us through their own paper rather than the originals. Bamberg gave relocated residents a free transport ticket and information, against a control group who had also moved, which shows an intervention after a move works rather than that the move made it work. Fujii and colleagues watched a freeway closure: most habitual drivers took a longer road, some tried the train, found they had overestimated the journey and stayed.
Walker and colleagues followed an office relocation and recorded old habits decaying as new ones set, which is the window shutting. Larcom and colleagues, whose abstract and summary material is all we read, found a two-day partial Underground strike in February 2014 left a minority of commuters permanently rerouted.
A break in context has no preference about which habits it loosens. The founding moving-house study, which [how much of what you do is habit](https://thewhyweproject.com/articles/how-much-of-what-you-do-is-habit) sets out properly, found habits surviving a move wherever the supporting context survived with them, and that is the same fact from the other side. The capability described in 2012 was real and the anecdote attached to it is unverifiable, and the psychology underneath was never the surveillance. When a context breaks, people briefly start noticing information again, for about three months, and the experiment that measured that window used a shopping bag and a conversation about saving water. Who reaches the window first, and carrying what?
Common questions
Can habit and purchase data predict a pregnancy?
No published figure answers that. We searched four ways and found no study reporting accuracy or precision for predicting pregnancy from retail purchase records. The nearest published work, De Caigny and colleagues on about 132,000 bank customers, predicts four life events including the birth of a child above a random baseline, and reports no precision or recall.
Did Target really tell a father his daughter was pregnant?
The scene is attributed in the 2012 magazine article to one unnamed employee present at the conversation, and to nobody at all in the book that retold it. Everything else rests on words the father is reported to have said on a later telephone call. We treat it as unverifiable reporting and assert nothing about either individual.
Where does the eighty-seven per cent figure come from?
From a hypothetical. The magazine article introduces it as a hypothetical example about a fictional shopper, with the word "say" in front of the number, supplied by an unnamed employee. The book prints the figure without those three qualifications and adds a second case the article does not contain.
How long does the habit disruption window stay open?
In the one field experiment designed to isolate it, about three months. Verplanken and Roy found their intervention worked only among people less than three months past a house move, with nothing in any later group, and they call the result a proof of concept with a small effect size. Beyond that study the window is poorly mapped.
Sources
Duhigg, C. (2012). 'How Companies Learn Your Secrets.' The New York Times Magazine, 16 February 2012
Duhigg, C. (2012). The Power of Habit: Why We Do What We Do and How to Change. Random House Books
Andreasen, A. R. (1984). 'Life Status Changes and Changes in Consumer Preferences and Satisfaction.' Journal of Consumer Research, 11(3), 784-794. doi:10.1086/209014
De Caigny, A., Coussement, K., & De Bock, K. W. (2020). 'Leveraging fine-grained transaction data for customer life event predictions.' Decision Support Systems, 130, 113232. doi:10.1016/j.dss.2019.113232
Verplanken, B., & Roy, D. (2016). 'Empowering interventions to promote sustainable lifestyles: Testing the habit discontinuity hypothesis in a field experiment.' Journal of Environmental Psychology, 45, 127-134. doi:10.1016/j.jenvp.2015.11.008
Larcom, S., Rauch, F., & Willems, T. (2017). 'The Benefits of Forced Experimentation: Striking Evidence from the London Underground Network.' The Quarterly Journal of Economics, 132(4), 2019-2055. doi:10.1093/qje/qjx020
Walker, I., Thomas, G. O., & Verplanken, B. (2015). 'Old Habits Die Hard: Travel Habit Formation and Decay During an Office Relocation.' Environment and Behavior, 47(10), 1089-1106. doi:10.1177/0013916514549619
Bamberg, S. (2006). 'Is a residential relocation a good opportunity to change people's travel behaviour?' Environment and Behaviour, 38, 820-840
Fujii, S., Garling, T., & Kitamura, R. (2001). 'Changes in drivers' perceptions and use of public transport during a freeway closure.' Environment and Behaviour, 33, 796-808
Wood, W., Tam, L., & Guerrero Witt, M. (2005). 'Changing circumstances, disrupting habits.' Journal of Personality and Social Psychology, 88, 918-933
Martin, J. A., Hamilton, B. E., & Osterman, M. J. K. (2025). 'Births in the United States, 2024.' NCHS Data Brief No. 535. doi:10.15620/cdc/174613
Piatetsky, G. (2014). 'Did Target Really Predict a Teen's Pregnancy? The Inside Story.' KDnuggets, 7 May 2014