Consumer behaviour is the basis of marketing strategy, not the afterthought

Three founders, three stories about the customer

Vanlog is an invented Rails SaaS from Galway for trades businesses: carpenters, electricians, tilers, the one van and two apprentices kind of firm. It does scheduling, quotes and invoices, and its three founders agree on everything except the one thing that matters. Ask them why customers sign up and you get three confident answers.

Sorcha, who built the product, says people come for the branded quotes, because they want to look as professional as the big firms. Dónal, who does sales, says people come because double bookings are costing them jobs and their diary is a whiteboard in the van. Aisha, who runs marketing, says people come because their accountant has told them to stop invoicing from a spreadsheet.

All three stories are plausible, and all three have been used to write landing pages, price plans and ad copy. None has ever been checked against what customers do inside the product. A hamper shop has the same problem in a different coat: the owner sees one crowd of "people who like nice food", while the orders say a gift buyer in December and a self buyer in March are two different customers.

Strategy is supposed to be built on how people decide. Analytics is supposed to be how you find out. If the strategy rests on founder folklore and the analytics reports last week's clicks, neither is doing its job.

Where this sits in the series

This is the second of 21 chapters in the Marketing Analytics series, which follows the structure of Mike Grigsby's book Marketing Analytics¹. The explanations, examples and numbers here are my own. The previous chapter, the statistics every marketer actually needs, gave you the vocabulary. This one supplies the questions those tools answer. The next, what is an insight, and why most reports do not contain one, asks whether an answer changes a decision, the only test of an insight I accept.

Consumer behaviour is how people decide, and every part of it leaves a trace

Consumer behaviour is the study of how people recognise a need, search, weigh options, buy, use and judge what they bought. Each building block hides a variable you can measure.

Needs and motivation. A need is the gap between where someone is and where they want to be, and motivation is the pressure that gap creates. A tiler with three jobs booked for the same Tuesday has an urgent need. A tiler who vaguely wants to "look more professional" has a slow one. An example: urgency shows up as time to first action. The first tiler creates a job in the calendar within ten minutes of signing up. The second browses quote templates for a week.

Perception. People respond to what they think your product is. If Vanlog's homepage leads with invoicing, a visitor with a scheduling problem sees an accounting tool and leaves. You can measure it through what people click first and where they leave.

Learning. Customers learn from experience, from other people and from your onboarding, and it shows up as habit: the second hamper order that arrives almost exactly a year after the first.

Attitudes. An attitude is a stored evaluation, "subscriptions are for big firms". Attitudes are invisible in transaction data, which is why surveys exist, and why surveys mislead, as you will see shortly.

Reference groups. Nobody decides alone. A carpenter asks the other carpenters in the WhatsApp group what they use. It shows up as referral sources, adoption clustered by trade or region, and the phrases people paste into support chats ("my mate said you do...").

The decision process has loops, and the loops are where you lose people

The textbook version of the purchase decision has five stages: need recognition, information search, evaluation of alternatives, the purchase decision and post purchase evaluation². It is a useful skeleton. The trouble is that the textbook draws it as a straight line, and real decisions double back.

Each box is a stage, each decision a fork, and the two notes are the loops the tidy version leaves out. Every arrow is a place where data can be collected and the customer can be lost. The postpone loop is where Vanlog's "after the summer" trials live.

Not buying is a choice, and it has been studied

The postpone step is an option the buyer actively picks. Ravi Dhar showed across seven studies that people defer a choice more often when no option has a decisive advantage³. Waiting is the sensible response to not knowing which option is best.

This is where a famous and contested idea comes in, choice overload. Sheena Iyengar and Mark Lepper set up a tasting stand in a grocery store with either six or 24 jams. The bigger display drew more people in (60% of passersby stopped, against 40%), but 30% of those who stopped at the small display bought a jar, against 3% at the large one⁴.

The evidence since then is mixed, and I say so plainly to anyone quoting the jam study. A meta analysis of 50 experiments with 5,036 participants found an average choice overload effect of virtually zero, with large differences between studies⁵. A later meta analysis of 99 observations found the effect does appear under specific conditions: complex choice sets, difficult decisions, uncertain preferences and a goal of minimising effort⁶. So "more options hurt sales" is not a law. "More options can push unsure people into the postpone loop" is defensible, and testable.

An example: a Kilkenny tea shop shows 40 loose leaf teas on one page. Visitors from "tea gift ideas" scroll, leave and often come back days later. They have uncertain preferences, so the question to test is whether they postpone more on the long page than on a short curated one.

Jobs To Be Done turns a story into a test

The most useful reframing I know for all of the above is Jobs To Be Done (JTBD). People do not buy products, they "hire" them to make progress towards a goal in a particular circumstance^{7 8}. The carpenter does not want scheduling software. He wants to stop losing Tuesdays. I have written about the method at length in Jobs To Be Done in growth strategy, so here only the analytical point.

A job is a behaviour hypothesis with a built in test. An example: "the job is winning the work and getting it into the diary in one motion" predicts that users who create a quote and schedule the job in the same session convert to paid at a higher rate than users who do only one of the two. That prediction can be wrong, which is exactly what makes it useful. Founder folklore cannot be wrong. "People want to look professional" survives any data, because "professional" can mean anything.

Stated and revealed preference disagree, and the gap is the finding

This distinction does more work than any other idea in this chapter. Stated preference is what people say they want, in surveys, interviews and the "why did you sign up" dropdown. Revealed preference is what they do with their time and money.

Revealed preference comes from economics. Paul Samuelson described it in 1938: if someone chooses A when B was available and affordable, A is revealed preferred to B⁹. The theory rests on what people choose, not on what they report feeling.

The two disagree constantly, and not because people lie. People are poor witnesses to their own motives. In 1977 Richard Nisbett and Timothy Wilson reviewed experiments in which people's explanations of their own choices missed what drove them¹⁰. In one, shoppers judging four identical pairs of nylon stockings laid out in a row preferred the rightmost pair by almost four to one, and virtually all denied that position had anything to do with it. That was a small study of 52 people, and the wider claim was disputed: Ericsson and Simon argued in 1980 that verbal reports are good data when people describe what they are attending to at the moment, rather than explain afterwards why they acted¹¹. Both sides support the practical lesson. Ask people what they did and saw, not why.

Two more findings make the gap concrete. A meta analysis of 47 experiments found that a medium to large change in what people intend to do (an effect size of d = 0.66) produced only a small to medium change in what they then did (d = 0.36)¹². And survey answers drift towards what sounds socially acceptable, a well documented bias called social desirability¹³. "I want to look professional" is a respectable reason to give a dropdown. "I double booked a customer and she shouted at me" is not.

An example: a Limerick kitchen fitter picks "look more professional" in the signup dropdown, then spends his first session entering next week's jobs into the calendar. Stated preference says quotes. Revealed preference says diary.

A purchase is a choice between buying and waiting

Economists put a little arithmetic under revealed preference, and it changes how you read a funnel. In a random utility model, the utility a person $i$ gets from option $j$ is

$$U_{ij} = V_{ij} + \varepsilon_{ij} = \sum_{k=1}^{K} \beta_k x_{ijk} + \varepsilon_{ij}$$

Here $U_{ij}$ is the total utility, $V_{ij}$ is the part that can be explained, $x_{ijk}$ are the $K$ observable attributes of option $j$ for person $i$ (price, delivery time, whether a quote template exists), $\beta_k$ is the weight people put on attribute $k$, and $\varepsilon_{ij}$ is everything not measured, treated as random. The person picks the option with the highest $U_{ij}$. Nobody sees it directly. You see the choice, and infer the weights $\beta_k$ from thousands of choices. Estimating weights like these is regression, explained from scratch in Foundations: regression.

If the random part follows a particular distribution (the extreme value distribution, the standard assumption), the probability that a person buys rather than waits takes the logit form, shown below¹⁴.

$$P(\text{buy}) = \frac{e^{V_{\text{buy}}}}{e^{V_{\text{buy}}} + e^{V_{\text{wait}}}}$$

Here $P(\text{buy})$ is the probability of buying now, $V_{\text{buy}}$ is the explainable utility of buying now, $V_{\text{wait}}$ is the explainable utility of postponing, and $e$ is the base of the natural logarithm, about 2.718. Probability as a long run share is explained in Foundations: probability, the logistic form in Foundations: regression, and $e$ in Foundations: logarithms.

Notice that postponing has a utility too. Most marketing treats "did not buy" as an absence. It is a choice, usually a sensible one, and the model gives it a seat at the table. Chapter 7 builds this into a full logistic regression.

Worked example: a free trial against waiting

The numbers are invented. For a typical tiler on Vanlog's pricing page, say $V_{\text{buy}} = 0.4$ and $V_{\text{wait}} = 0$ (one option is always fixed at zero, because only the difference matters). Then $e^{0.4} = 1.492$ and $e^{0} = 1$, so $P(\text{buy}) = 1.492 / 2.492 = 0.599$, about 60%.

Now Vanlog adds a free trial with no card required, which the example assumes raises $V_{\text{buy}}$ to 0.9. Then $e^{0.9} = 2.460$, and $P(\text{buy}) = 2.460 / 3.460 = 0.711$, about 71%. The same shift would move a buyer from 20% to 29%, or from 90% to 94%, because the logit curve is steepest in the middle. A lever works hardest on the people closest to undecided.

Expected utility maps every marketing lever to one term

For a single purchase under uncertainty, the older expected utility view is also worth keeping close. It is the probability weighted average of the value of each possible outcome¹⁵. Written for one purchase, with the cost taken out separately:

$$EU(\text{buy}) = p \cdot u(\text{works}) + (1 - p) \cdot u(\text{disappoints}) - c$$

Here $EU(\text{buy})$ is the expected utility of buying, $p$ is the buyer's belief that the product will do the job, $u(\text{works})$ is the value to them if it does, $u(\text{disappoints})$ is the value if it does not (often negative, the wasted setup time), and $c$ is the cost in money and effort. It is an expected value with utilities in place of euro, covered in Foundations: expected value.

Every marketing lever maps onto a term. Testimonials and a free trial raise $p$. A clearer promise raises $u(\text{works})$. A guarantee lifts $u(\text{disappoints})$. A simpler signup lowers $c$. When a founder says "we need more awareness", I ask which term they think is the problem. Usually it is not awareness.

Worked example: which term to work on

The numbers are invented, and to keep it simple the example measures utility in euro. A Cork electrician believes Vanlog will work with probability $p = 0.6$. If it works, it saves her about €1,200 of lost jobs in a year, so $u(\text{works}) = 1{,}200$. If it disappoints, she loses about €100 of evenings spent setting it up, so $u(\text{disappoints}) = -100$. The plan costs €30 a month, so $c = 360$ for the year.

Step by step: $0.6 \times 1{,}200 = 720$, then $0.4 \times (-100) = -40$, then $720 - 40 - 360 = 320$. Positive, so buying makes sense to her, but not by much.

Now compare two levers. Cutting the price by €5 a month lowers $c$ by €60 and lifts the result to 380. Three reviews from other electricians that raise her belief to $p = 0.75$ give $900 - 25 - 360 = 515$. In this example the proof is worth more than three times the discount, and it costs Vanlog no margin.

Marketing strategy chooses whom to serve and how to win them

Strategy, stripped of the slideware, answers four questions: whom to serve, what to offer them that others do not, how to make that visible, and how to make money doing it. The vocabulary is segmentation, targeting, positioning, the marketing mix and competitive advantage.

Segmentation and targeting

Segmentation divides the market into groups that behave differently and would respond to a different offer. The key word is behave. Company size or age is useful only if it predicts behaviour, and it often does not. A good segmentation is one where you would happily run a different campaign for each group (chapters 13 and 14 go deep on this). Targeting is choosing which groups to serve and, harder, which to ignore. Vanlog cannot be the best tool for a solo plumber and a 40 person building firm at once.

Positioning

Positioning is deciding how you want the market to see your product relative to the alternatives, and then communicating that. The classic tool is a perceptual map, which plots the offerings on the two attributes buyers use to decide¹⁶. I would take the axes from support chats and lost deal notes, not from the founders.

An example: Vanlog's two axes, scored from 0 to 1 for illustration, are how far a product is built for the van rather than the office, and how far it covers the whole job rather than invoicing only.

Product	Built for the van (0 to 1)	Whole job in one place (0 to 1)	Where it sits
Vanlog	0.78	0.82	Built for the van, whole job
Big suite A	0.25	0.85	Office, whole job
Big suite B	0.35	0.70	Office, whole job
Invoicing app	0.70	0.25	Van, invoicing only
Accountant portal	0.15	0.40	Office, mostly invoicing
Spreadsheet	0.20	0.15	Office, invoicing only

Vanlog is alone in the corner that is built for the van and covers the whole job. An empty corner is valuable only if enough buyers want what it offers. Otherwise it is empty for a reason.

The marketing mix

The marketing mix is the set of levers you pull to occupy that position: product, price, place and promotion, the four Ps¹⁷. For a SaaS business it is the roadmap, the plans, the channels and the message. The mix is downstream of positioning, which is downstream of behaviour. Most companies work upwards from the mix, which is why they end up optimising ad copy for a promise nobody wanted.

Competitive advantage

Michael Porter's two basic types of competitive advantage are lower cost and differentiation, combined with a choice of a broad or a narrow market¹⁸. Differentiation in a narrow market is the natural home of a small firm, because cost leadership needs scale. But difference is worth money only if it lands on an attribute with a large $\beta_k$ in the utility model above. Being different on something nobody weighs is just being odd.

Every analytics technique answers a behaviour question

The claim of this whole series in one sentence: every analytics technique answers a consumer behaviour question, and every strategic choice depends on the answer to at least one of them. The table below is the map I keep on my wall, with the chapter where each question gets its proper treatment.

Behaviour question	Technique	Chapter
How much less will they buy if we raise the price?	Regression, elasticity	4
How long does a campaign keep working after it stops?	Polynomial distributed lags	5
How many times a year will a customer come back?	Poisson regression	6
Who is likely to buy, and what goes with what?	Logistic regression, market basket	7
When will this customer leave, and what are they worth until then?	Survival analysis, customer lifetime value	8
Did the new layout lift sales, or was it the season?	Panel regression	9
How much will we sell next quarter?	Forecasting, seasonality	10
Does marketing drive sales, or do sales drive the marketing budget?	Simultaneous equations	11
Which of these 40 survey items measure the same thing?	Factor analysis, principal components	12
Who are our customers, in groups that behave differently?	Segmentation, k means, latent class	13 and 14
Where should the next euro go?	Media mix modelling	15 and 16
Why do they stay, and what makes them leave?	Loyalty models, structural equations	17 to 19
Did the change work, or did we get lucky?	Testing, factorial designs	20

Not one question mentions a dashboard metric. Nobody's behaviour is "sessions". The metric is a shadow of the behaviour, and the technique reads the shadow.

Worked example: three beliefs become three questions

Back to Vanlog. The job is to turn each founder's story into something the data can contradict. My framework has four steps and fits on a napkin.

1. Write the belief as a sentence about a person in a situation. Not "customers want professionalism" but "a carpenter who has just lost a quote to a bigger firm signs up to send better looking quotes".
2. Name the behaviour that would be different if the belief were true. That carpenter opens the quote builder first, uses a template, and sends a quote within days.
3. Name the data that records that behaviour, and what is missing. Signup time, first feature used, quote sent, source. Missing: whether they lost a quote recently, which only a signup question captures.
4. Write the decision that changes with the answer. If the belief holds, the homepage leads with quotes and onboarding walks through a template. If it fails, it does not.

For the three founders the napkin looks like this.

Founder belief	Behaviour if true	Data question	What would contradict it
People sign up to look more professional (Sorcha)	Quote builder is the first feature, branded template applied early	Share of trials whose first action is a quote, and their conversion against the rest	Quote first users convert no better than others
People sign up because double bookings cost them jobs (Dónal)	Calendar is the first feature, several jobs entered in the first session	Share of trials with three or more jobs in the first 48 hours, and their conversion	Heavy calendar users cancel like light ones
People sign up because the accountant said so (Aisha)	Invoice export or accountant invite early, signups bunched before tax deadlines	Share of trials inviting an accountant in week one, and the seasonality of signups	No seasonal peak, no accountant invites

Now the results. Vanlog has a "why are you here" dropdown at signup, so stated and revealed preference sit side by side. The numbers are illustrative. The pattern is one I have seen in real products more than once.

Stated reason at signup	Share of trials	Most common first feature	Trial to paid rate
Look more professional	41%	Quote templates (52%)	18%
Stop double bookings	33%	Calendar (71%)	34%
Accountant or bookkeeping	26%	Invoice export (64%)	27%
All trials	100%		26%
Created a quote and scheduled it in the first week, any stated reason	22%		51%

Sorcha's story is the most common thing people say and the worst converting group. "Look professional" is the socially comfortable answer to a dropdown, and it describes a slow need. Dónal's is a third of signups with nearly double the conversion, and the tightest link between stated reason and first action. Aisha's converts decently and is worth checking for the seasonal peak she predicted. As a check, the weighted average of the three groups is $0.41 \times 18 + 0.33 \times 34 + 0.26 \times 27 = 25.6$, which rounds to the 26% shown.

The last row pays for the analysis. Cutting across all three stated reasons, the 22% of trials who created a quote and put the job in the calendar in the same first week converted at 51%, almost double the baseline. None of the founders described that job. The customer's job is not "look professional" or "stop double bookings". It is "win the work and lock it into the diary before I forget". That is a positioning, an onboarding flow and a homepage headline, and it came from a napkin plus one SQL query.

The horizontal axis shows the five groups: three defined by what people said at signup, all trials, and one defined by what people did in their first week. The vertical axis is the share of each group that became paying customers. The takeaway is the last bar: the group nobody named, defined by behaviour, converts at about twice the baseline.

Worked example: two kinds of hamper buyer

Northgate Hampers, an invented gift hamper shop in Cork, gives the same lesson in retail clothes. The owner, Priya, sees one customer. The orders describe two. Again the numbers are illustrative.

Attribute	Gift buyers	Self buyers
Share of orders	63%	37%
Share of revenue	71%	29%
Average order value	€68	€47
Ships to a different address	94%	6%
Includes a gift message	81%	2%
Orders placed in November and December	58%	24%
Repeat within 12 months	19%	44%
Most common repeat interval	52 weeks	nine weeks

The revenue shares follow from the orders and the order values: per 100 orders, gift buyers bring $63 \times 68 = 4{,}284$ euro and self buyers $37 \times 47 = 1{,}739$, so the gift share is $4{,}284 / 6{,}023 = 71\%$.

A gift buyer spends more, buys mostly in the last eight weeks of the year, and comes back almost exactly a year later if at all, buying for the same person's Christmas again. A self buyer spends less, buys all year, and comes back every couple of months. These are not two segments of one shop. They are two shops. The gift buyer needs a reminder in early November, a "send it again" button and a delivery promise she can trust. The self buyer needs a subscription or a loyalty scheme, chapter 17's territory. The behaviour told Priya the strategy, not the other way round.

Why does the gift buyer pay more? Economists have argued that a gift works partly as a signal of how much the giver values the relationship, which is why cash or the exact thing someone asked for can feel wrong as a present¹⁹. A signal is priced differently from groceries. The behavioural economics frameworks article goes further into how customers actually decide.

How to run this with three tables and a fortnight

You do not need a data warehouse. The data: a customer or account table with a signup or first order time and the source. An event or order table with times and a type. Whatever stated preference you collect, from the signup dropdown to the first support chat. If you have none, add one question at signup this week. One question, not a form. For the Vanlog question, a first pass in SQL is about a dozen lines.

WITH first_week AS (
  SELECT account_id,
         BOOL_OR(event = 'quote_created')   AS made_quote,
         BOOL_OR(event = 'job_scheduled')   AS scheduled_job
  FROM events e
  JOIN accounts a USING (account_id)
  WHERE e.created_at < a.signed_up_at + INTERVAL '7 days'
  GROUP BY account_id
)
SELECT made_quote, scheduled_job,
       COUNT(*)                          AS trials,
       AVG(CASE WHEN a.paid_at IS NOT NULL THEN 1.0 ELSE 0 END) AS trial_to_paid
FROM first_week f
JOIN accounts a USING (account_id)
WHERE a.signed_up_at < NOW() - INTERVAL '45 days'
GROUP BY 1, 2
ORDER BY trial_to_paid DESC;

The last WHERE clause matters. Count only trials old enough to have had a fair chance to convert, or recent signups drag the rate down and your newest feature looks guilty. Framing the beliefs takes an afternoon. Cleaning the events takes most of a week, because there is always one event that was renamed in March. The analysis is two days, the write up another two.

Checks before you trust it

Group sizes first. A 51% rate on 14 accounts is a rumour, on 400 it is a finding. The interval method is in Foundations: confidence intervals.

An example: with 400 accounts at 51%, the standard error is $\sqrt{0.51 \times 0.49 / 400} = \sqrt{0.000625} = 0.025$, so the 95% interval is $51\% \pm 1.96 \times 2.5$ points, about 46% to 56%. Comfortably above 26%. With 14 accounts the standard error is $\sqrt{0.51 \times 0.49 / 14} = 0.134$, and the interval runs from about 25% to 77%, which includes the baseline. Same rate, no finding.

Direction next. Does creating a quote cause conversion, or do people who were always going to convert happen to create quotes? Only a test can separate the two, which is chapter 20. Then try three days and 14 days as well as seven. And check that the event fires on mobile, where half your joiners live.

Pitfalls I see most often

Asking people why and believing the answer. Surveys are for attitudes and for things that leave no trace in the data, and they are poor at motives. Put the stated answer next to the revealed behaviour and treat the gap as the finding.

Quoting a famous effect as a law. Choice overload, the jam study and many of their cousins in popular books rest on single experiments that later work has qualified or failed to repeat. Use them as hypotheses for your own data, never as proof.

Segmenting on what is easy instead of on what differs. Company size, region and age are in every CRM, so that is where people cut. If two segments would get the same campaign, they are one segment.

Mistaking the founder's job for the customer's job. Sorcha built quote templates because she is proud of them. The customer hired Vanlog for a different job, and the roadmap should follow that one.

Treating the decision process as a straight line. Real journeys pause for months, restart from a trigger, or skip the search because a mate recommended you. A funnel report that assumes a straight line calls a postponed customer lost and a returning one new.

Confusing positioning with copy. Positioning is the place you occupy in the customer's comparison, copy is how you describe it. You can rewrite the copy every week. The position moves only when the product, the price or the proof moves. Frameworks such as StoryBrand and the psychology of why users click or bounce are worth knowing, but they improve the delivery of a promise, not the promise.

How I do this for clients

The deliverable is a behaviour brief, in place before anyone touches a model, a campaign or a redesign. It starts with the free workshop, where we write down every belief about why customers buy, in the "person in a situation" form above, and rank them by how much of the strategy rests on each. Then come two weeks of real work on me. I need read access to your orders or events, your CRM or account table, and whatever stated preference you collect. I write the queries, put stated next to revealed, and check the group sizes so nobody gets excited about 14 accounts.

What you get is a short document: each belief, the behaviour that would confirm it, the number we found, the interval around it, and a verdict of confirmed, contradicted or not yet testable, with the one question to add at signup for the last kind. Attached is the positioning map on the axes your buyers actually use. You own the queries, the document and the map.

If you then want the models built, that sits under Data Science. If you want the findings turned into channels and campaigns, that is Growth Hacking Plus, and for an online shop there is also a commission model where I only win when you do. The pricing page says what each of those costs in plain terms. The workshop and the two weeks cost you nothing but access and honesty.

Questions to put to your own team or agency

• Which three beliefs about why our customers buy is the current strategy built on, and when did we last check any of them against behaviour?
• For each belief, what would customers do differently if it were true, and do we record that behaviour anywhere?
• Where do our stated reasons and our revealed behaviour disagree most, and what have we done about that gap?
• Which two attributes do our buyers actually compare us on, and how do we know it is those two?
• If we split customers by what they do rather than who they are, how many groups do we get, and would each get a different offer?
• Which question in the behaviour to technique table would change our biggest decision this year, and who is answering it?

Footnotes

1. Mike Grigsby, Marketing Analytics: A Practical Guide to Improving Consumer Insights Using Data Techniques, third edition, Kogan Page, 2022, https://www.netgalley.co.uk/catalog/book/267990

2. Maria Gomez Albrecht, Mark Green and Linda Hoffman, Principles of Marketing, section 3.3 The Consumer Purchasing Decision Process, OpenStax, 2023, https://openstax.org/books/principles-marketing/pages/3-3-the-consumer-purchasing-decision-process

3. Ravi Dhar, Consumer Preference for a No Choice Option, Journal of Consumer Research, volume 24, issue 2, pages 215 to 231, 1997, https://doi.org/10.1086/209506 (copy at https://bear.warrington.ufl.edu/brenner/mar7588/Papers/dhar-jcr1997.pdf)

4. Sheena S. Iyengar and Mark R. Lepper, When Choice is Demotivating: Can One Desire Too Much of a Good Thing?, Journal of Personality and Social Psychology, volume 79, issue 6, pages 995 to 1006, 2000, https://faculty.washington.edu/jdb/345/345%20Articles/Iyengar%20&%20Lepper%20%282000%29.pdf

5. Benjamin Scheibehenne, Rainer Greifeneder and Peter M. Todd, Can There Ever Be Too Many Options? A Meta Analytic Review of Choice Overload, Journal of Consumer Research, volume 37, issue 3, pages 409 to 425, 2010, https://econpapers.repec.org/RePEc:oup:jconrs:v:37:y:2010:i:3:p:409-425

6. Alexander Chernev, Ulf Böckenholt and Joseph Goodman, Choice overload: A conceptual review and meta analysis, Journal of Consumer Psychology, volume 25, issue 2, pages 333 to 358, 2015, https://www.kellogg.northwestern.edu/faculty/research/detail/2015/when-product-assortment-leads-to-choice-overload-a-conceptual

7. Clayton Christensen Institute, Jobs to Be Done Theory, Christensen Institute, https://www.christenseninstitute.org/theory/jobs-to-be-done/

8. Clayton M. Christensen, Taddy Hall, Karen Dillon and David S. Duncan, Know Your Customers' "Jobs to Be Done", Harvard Business Review, September 2016, https://hbr.org/2016/09/know-your-customers-jobs-to-be-done

9. Hal R. Varian, Revealed Preference, survey prepared for Samuelsonian Economics and the 21st Century, revised September 2006 (on Samuelson's 1938 and 1948 papers), https://people.ischool.berkeley.edu/~hal/Papers/2005/revpref.pdf

10. Richard E. Nisbett and Timothy DeCamp Wilson, Telling More Than We Can Know: Verbal Reports on Mental Processes, Psychological Review, volume 84, issue 3, pages 231 to 259, 1977, https://web.math.princeton.edu/~sswang/literature_general_unsorted/Nisbett-Wilson-PsychologicalReview1977.pdf

11. K. Anders Ericsson and Herbert A. Simon, Verbal Reports as Data, Psychological Review, volume 87, issue 3, pages 215 to 251, 1980, https://acs.ist.psu.edu/ist597/papers/ericssonS80.pdf

12. Thomas L. Webb and Paschal Sheeran, Does changing behavioural intentions engender behaviour change? A meta analysis of the experimental evidence, Psychological Bulletin, volume 132, issue 2, pages 249 to 268, 2006, https://eprints.whiterose.ac.uk/1580

13. Ivar Krumpal, Determinants of social desirability bias in sensitive surveys: a literature review, Quality and Quantity, volume 47, issue 4, pages 2025 to 2047, 2013, https://ideas.repec.org/a/spr/qualqt/v47y2013i4p2025-2047.html

14. Kenneth Train, Discrete Choice Methods with Simulation, second edition, chapter 3 Logit, Cambridge University Press, 2009, https://eml.berkeley.edu/books/choice2nd/Ch03_p34-75.pdf

15. R. A. Briggs, Normative Theories of Rational Choice: Expected Utility, The Stanford Encyclopedia of Philosophy, first published 2014, revised 2023, https://plato.stanford.edu/entries/rationality-normative-utility/

16. Maria Gomez Albrecht, Mark Green and Linda Hoffman, Principles of Marketing, section 5.6 Product Positioning, OpenStax, 2023, https://openstax.org/books/principles-marketing/pages/5-6-product-positioning

17. Maria Gomez Albrecht, Mark Green and Linda Hoffman, Principles of Marketing, section 1.2 The Marketing Mix and the 4Ps of Marketing, OpenStax, 2023, https://openstax.org/books/principles-marketing/pages/1-2-the-marketing-mix-and-the-4ps-of-marketing

18. Institute for Manufacturing, University of Cambridge, Porter's Generic Competitive Strategies, summarising Michael E. Porter, Competitive Advantage, The Free Press, 1985, chapter 1, https://www.ifm.eng.cam.ac.uk/research/dstools/porters-generic-competitive-strategies/

19. Colin Camerer, Gifts as Economic Signals and Social Symbols, American Journal of Sociology, volume 94, supplement, pages S180 to S214, 1988, https://authors.library.caltech.edu/records/nskm8-qzq73