The Hook Model and habit loops: repeat engagement without dark patterns

The point: a habit is a cue that fires without a decision

Most businesses want customers to come back without having to decide to. That is a habit, and it is worth being precise about before anyone draws a loop on a whiteboard.

Psychologists define a habit as a response that a context triggers automatically, because the two have been paired many times before. Wendy Wood and David Neal put it plainly: habits come from the slow learning of links between an action and the features of the situation it usually happens in, and once the link is formed, seeing the situation triggers the action without a goal in between.¹ Benjamin Gardner, Phillippa Lally and Jane Wardle add the part that matters for a business: because the action no longer runs through a conscious decision, habits tend to persist even after the motivation that started them has faded.²

The Hook Model is Nir Eyal's attempt to turn that into a design method: a four step loop that products repeat until using them becomes automatic.³ It is a useful map. It also leans on some research harder than the research can carry, and it can build things people later wish they had never started. So I go through each step with a worked example, check what the studies behind it actually measured, and draw the line between a habit that serves the customer and a hook that only serves the dashboard.

Every worked example uses one invented business: Shandon Roasters, a small Cork coffee roaster selling beans at shandonroasters.example, as one off orders and by subscription. The shop and its numbers are made up.

The Hook Model in four steps

Eyal names the four steps trigger, action, variable reward and investment.⁴ In his words, what separates a hook from a plain feedback loop is that it creates wanting, not just a response.⁴

A trigger is the cue that starts the loop: external at first (an email, a button), internal later if the loop works (a situation that brings the product to mind on its own). The action is the smallest thing the person does in expectation of a reward. The variable reward is what they get, and the variable part is the point: the outcome is not quite the same each time. The investment is something the person puts in, such as time, data, effort, social capital or money, that makes the next pass through the loop better.⁴

The diagram reads top to bottom, and the last box feeds back into the first. Each step only works if the one before it did. A trigger with no easy action is an annoyance. An action with no reward teaches the person to ignore the trigger.

Worked example: mapping Shandon's reorder journey

First I map what Shandon does today against the four steps.

1. Trigger. A customer who bought a 250g bag gets one email, seven days after delivery, saying "Time for more coffee?". Nothing else.
2. Action. To reorder, they click through to the home page, find the coffee again, choose the grind, add it to the basket, sign in, confirm the address and pay. I count seven separate steps.
3. Reward. The coffee arrives. It is grand, and it is exactly the same as last time.
4. Investment. None. The shop does not remember the grind, the brewing method or what the customer thought of the last bag.

Laid out like that, the weak points are plain: a calendar trigger, a long action, a flat reward and no investment. The rest of the article takes them one at a time.

Triggers: external first, internal is the goal

An external trigger carries its own instruction: an email that says "reorder", a button that says "buy again". An internal trigger is a situation or feeling that the person has learned to pair with the product. Eyal's point is that the solution that comes to mind first wins: once the habit sets, the product comes to mind when the situation arises, unprompted.⁴

That matches the research: a habit is triggered by the context, not by a reminder.¹ So an external trigger should arrive at the moment the real context occurs, consistently enough that the pairing forms. A reminder at a random moment has nothing to pair with.

For a coffee customer the context is obvious: the bag is nearly empty.

Worked example: timing the reminder to the bag, not the calendar

Shandon asks one optional question at checkout: "Roughly how many cups a day?". A customer answers two. The shop's own recipe card uses 18g of coffee per cup.

1. Daily use: 2 cups times 18g is 36g a day.
2. Days a 250g bag lasts: 250 divided by 36 is 6.9 days.
3. The reminder should arrive with about two days of coffee left, so the reorder lands before the bag runs out. Two days at 36g is 72g. The bag has 72g left after 178g has been used, which is on day 5 (178 divided by 36 is 4.9).
4. So the reminder goes out on day 5 after delivery, not day 7, and it says what is true: "You will be down to the last few cups in a couple of days."

A one cup a day customer gets it on day 12 instead. Over a good few cycles the email arrives just as the customer notices the bag is light, and the noticing starts to do the work on its own. That is the external trigger handing over to the internal one.

Action: make the next step smaller than the motivation it needs

Eyal describes this step as pulling on two levers, how easy the action is and how motivated the person is to do it.⁵ BJ Fogg's behaviour model makes the same point. Fogg states it as B = MAP: a behaviour happens when motivation, ability and a prompt come together at the same moment, and when a behaviour does not happen, at least one of the three is missing.⁶ Fogg does not write it as a multiplication, so neither do I. The point needs no arithmetic: the lower the effort, the less motivation the person needs at the moment the prompt arrives.

Raising motivation is hard and usually means louder messages. Lowering the effort is engineering, entirely in your hands. In my experience that is where most of the gain is, and it never annoys anyone.

An example: Shandon's reminder email used to link to the home page. The customer had to rebuild the order from memory. The rebuilt version links to a page that already holds the last order.

1. Before: open email, click through, find the coffee, choose the grind, add to basket, sign in, confirm address and pay. Seven steps.
2. After: open email, tap "Same again", confirm on a page that shows the coffee, the grind, the address and the saved card. Two steps after opening the email.
3. Nothing is hidden. The confirm page shows the price and a clear "Change something" link, and nothing is charged until the customer presses confirm.

That is not a trick, just the shop remembering what the customer already told it. For the wider method of removing friction, the Clarity framework article goes through it screen by screen.

Variable reward: what the research actually measured

Eyal describes three kinds of variable reward. Rewards of the tribe are social: feeling accepted, important and included. Rewards of the hunt are the search for resources, which today means deals and information. Rewards of the self are about mastery and completion.⁷ He supports the idea with brain research, writing that dopamine levels rise when the brain is expecting a reward, and compares the effect to slot machines and lotteries.⁵

The research is real, but narrower than the slot machine story. In 1997 Wolfram Schultz, Peter Dayan and Read Montague brought together recordings from dopamine neurons in primates, which appear to signal the difference between the reward expected and the reward received, a prediction error.⁸ In 2003 Christopher Fiorillo, Philippe Tobler and Schultz recorded dopamine neurons in macaque monkeys while showing them signals that predicted a reward with different probabilities. The quick burst of activity tracked the probability. A second, slower response grew until the moment the reward might come, and it was strongest when the outcome was most uncertain, at a probability of 0.5.⁹

So what was measured is that neurons in monkeys respond most to uncertainty in a laboratory task. That supports the idea that uncertainty catches attention. It is not evidence that a variable discount in an online shop builds a lasting habit in people. The slot machine comparison is an extrapolation, and I treat it as one.

Both ideas can be written down. First, the expected value of a reward:

$$E[X] = \sum_i p_i \, x_i$$

Here $E[X]$ is the expected value of the reward, $p_i$ is the probability of outcome $i$ and $x_i$ is the value of that outcome, summed over every outcome $i$. Foundations: expected value

Second, the uncertainty of a reward that either happens or does not:

$$\text{uncertainty} = p\,(1 - p)$$

Here $p$ is the probability that the reward arrives and $1 - p$ the probability that it does not. Their product is the variance of a single yes or no event, the building block of the binomial distribution. It is zero when the reward is certain ($p = 1$) or impossible ($p = 0$) and largest, 0.25, at $p = 0.5$. That is the shape Fiorillo and colleagues found in the slower dopamine response.⁹ Foundations: distributions

Worked example: same cost, very different uncertainty

Shandon's marketing tool offers a "mystery discount": on a €24 bag, one order in four gets 20% off at random. The alternative is a fixed €1.20 off every reorder.

1. The discount when it lands: 20% of €24 is €4.80.
2. The expected discount per order:

$$E[X] = 0.25 \times 4.80 + 0.75 \times 0 = 1.20$$

3. So both options cost the shop the same, €1.20 per order on average.
4. The fixed discount has no uncertainty at all: $p = 1$, so $p(1 - p) = 0$.
5. The mystery discount has $p = 0.25$, so $p(1 - p) = 0.25 \times 0.75 = 0.1875$, three quarters of the maximum of 0.25.

Same cost, very different feeling. The mystery version keeps people wondering, which is why it works on attention and why I do not use it. It rewards the act of ordering rather than the coffee, and trains people to wait for luck.

Instead, I keep the variety and drop the gamble. Each month one subscriber bag includes a small guest coffee from another roaster, chosen for the customer's taste. It is always good. What varies is the discovery, not whether they win. Nobody regrets a nice coffee.

Investment: stored value the customer can take with them

The investment step is where the person puts something in that improves the next pass. Eyal's list includes inviting friends, stating preferences, building up virtual assets and learning to use new features. He is clear that the point is not getting money out of the customer, it is an action that makes the service better for the next round.⁵

It is also where habit design most often slides into lock in. Stored value makes leaving more costly, which is fine when the value is real and not fine when the cost of leaving is the point. My rule is simple. The investment should make the next order better, and the customer should be able to see it, change it and take it away.

Worked example: a taste profile that earns its keep

After the second bag, Shandon asks two optional questions in the confirmation email: "How do you brew?" and "How did the last bag taste to you?", with five plain choices from "too bright" to "too dark".

1. A customer answers "cafetière" and "a bit bright".
2. The next reminder suggests the same coffee at a coarser grind, or a darker one from the range, with a sentence on why.
3. The "Same again" page now remembers the grind, so the action from the last section stays at two steps.
4. The customer's account page shows the profile in plain words, lets them edit or delete it, and has a "Pause or cancel" link that works in one step.

Each answer improves the trigger, the action and the reward. Because the profile is visible and removable, the customer stays because the coffee got better, not because leaving got harder.

How long a habit takes, and what Lally actually measured

The most quoted number in habit design is 66 days. The study behind it is more interesting than the number.

Phillippa Lally and colleagues asked 96 volunteers to pick one eating, drinking or activity behaviour and do it every day in the same context, such as after breakfast, for 12 weeks. Each day the volunteers filled in the self report habit index, a questionnaire, and recorded whether they had done the behaviour. Automaticity, how automatic the behaviour felt, was scored from a subset of the questionnaire's items. 82 people gave enough data to analyse. The researchers fitted a curve that rises and then levels off to each person's scores over the 84 days. It fitted for 62 people, and fitted well for 39.¹⁰

The time to reach 95% of each person's plateau ranged from 18 to 254 days. Missing one opportunity did not materially affect the habit forming, and more consistent repetition went with a better fitting curve.¹⁰

The 66 days is the typical time to the plateau. Gardner, Lally and Wardle report that automaticity "slowed to a plateau after an average of 66 days", with wide variation between people and behaviours.² A systematic review in 2024 by Singh and colleagues gathered 20 studies with 2,601 participants. Only four reported how long habits took. Those four gave medians of 59 to 66 days, means of 106 to 154 days, and individual times from 4 to 335 days. The reviewers rated 11 of the 20 studies at high risk of bias, and they found that habits chosen by the person themselves, and practices done in the morning, tended to be stronger.¹¹

Two caveats for a shop. First, these are daily health behaviours people chose for themselves, such as drinking water, flossing or walking. None of it measured a purchase, an app or a notification. Second, a daily behaviour gets 84 repetitions in 12 weeks, a weekly coffee order gets 12. My inference, not a finding: a purchase habit takes far longer in calendar time than 66 days.

The shape of the curve is the useful part. A curve that rises fast and then levels off can be written like this:

$$A(t) = a \left(1 - e^{-k t}\right)$$

Here $A(t)$ is the automaticity score after $t$ days, $a$ is the level the person eventually plateaus at, $e$ is the base of the natural logarithm (about 2.718) and $k$ is how fast that person gets there. This is my notation for the general asymptotic shape Lally's team describe, not their exact published equation. Foundations: logarithms and exponentials

Setting $A(t)$ to 95% of $a$ and solving for $t$ gives the time to the plateau:

$$t_{95} = \frac{\ln 20}{k}$$

Here $t_{95}$ is the number of days to reach 95% of the plateau, $\ln 20$ is the natural logarithm of 20 (about 3.0, because $1 - 0.95 = 0.05 = 1/20$) and $k$ is the rate from the formula above. Foundations: logarithms and exponentials

Worked example: two customers, one curve

Take two invented customers building the same daily habit, a morning cup brewed at home. One picks it up fast, the other slowly.

1. Fast: $k = 0.045$ per day.

$$t_{95} = \frac{2.996}{0.045} = 66.6$$

2. So about 67 days to the plateau.
3. Slow: $k = 0.012$ per day.

$$t_{95} = \frac{2.996}{0.012} = 249.6$$

4. So about 250 days, close to the slowest people in Lally's study.
5. At day 28 the fast customer is at $1 - e^{-0.045 \times 28} = 1 - e^{-1.26} = 0.72$, so 72% of their plateau. The slow one is at $1 - e^{-0.012 \times 28} = 1 - e^{-0.336} = 0.29$, so 29%.

The horizontal axis is the day since the person started, over the 84 days of Lally's study. The vertical axis is how far each person has got towards their own plateau. The takeaway: after the same 12 weeks, one customer has the habit and the other is not two thirds of the way there. A plan that assumes everyone settles in by week three will write off the slow half, or nag them until they leave.

Measuring a habit in your own numbers

Nobody can see automaticity in a shop's database. You can see whether people come back, and whether that changed when the loop changed.

The first is the repeat rate of a cohort, the customers whose first order fell in the same month:

$$\text{repeat rate} = \frac{\text{customers who reorder within 60 days}}{\text{customers in the cohort}}$$

Here the numerator counts customers from the cohort who place a second order within 60 days of their first, and the denominator counts everyone whose first order fell in that month. It is a rate like any other, events over opportunities, so $\text{repeat rate}$ sits between 0 and 1. Foundations: ratios and rates

The second is whether the difference between two cohorts is more than noise, the two proportion z test:

$$z = \frac{p_2 - p_1}{\sqrt{p(1-p)\left(\frac{1}{n_1} + \frac{1}{n_2}\right)}}$$

Here $p_1$ and $p_2$ are the repeat rates of the earlier and later cohort, $n_1$ and $n_2$ the number of customers in each, and $p$ the pooled rate across both. A $z$ above about 1.96 means a difference that large would come up less than 5% of the time by chance alone. Foundations: hypothesis tests

Worked example: did the new loop change anything?

Shandon changes the trigger, the action and the investment step in April and compares cohorts.

1. March cohort, old journey: 400 first time customers, 132 reorder within 60 days. Repeat rate 132 divided by 400 is 33%.
2. May cohort, new journey: 380 first time customers, 152 reorder within 60 days. Repeat rate 152 divided by 380 is 40%.
3. The difference is 7 percentage points, a 21% relative rise (0.40 divided by 0.33 is 1.21).
4. Pooled rate: 284 reorders out of 780 customers is 0.364.
5. Standard error: $\sqrt{0.364 \times 0.636 \times (1/400 + 1/380)} = 0.0345$.
6. Then $z = 0.07 / 0.0345 = 2.03$, a two sided p value of about 0.04, just under the usual 5% line.

So the change is probably real, but only just, and March and May are different months. I would keep the new journey and watch the June and July cohorts before calling it settled. If you want to know how many customers a comparison like this needs before it means anything, How much data do you actually need works through it.

Why does a few points of repeat rate matter so much? Because it feeds straight into what a customer is worth over time:

$$CLV = M \times \frac{r}{1 + d - r}$$

Here $CLV$ is the customer lifetime value, $M$ is the margin a customer brings in per year (margin, not revenue), $r$ is the share of customers still buying a year later and $d$ is the discount rate that turns future money into today's money. Foundations: lifetime value

An example: a Shandon subscriber brings €90 a year in margin, and the shop uses a 10% discount rate. If 55% of subscribers are still there after a year, $90 \times 0.55 / (1.10 - 0.55) = 90$, so €90. At 65% it is $90 \times 0.65 / (1.10 - 0.65) = 130$, so €130. Ten points of retention is worth €40 per subscriber, about 44% more.

The horizontal axis is the share of subscribers still buying after a year. The vertical axis is what one subscriber is worth today. The takeaway is the steepness: each extra ten points of retention is worth more than the last, which is why a habit that genuinely holds pays so well. For the wider retention picture, Survival analysis, churn and customer lifetime value and Loyalty: the three Rs go deeper.

Where a habit becomes a hook in the bad sense

The same four steps that keep good coffee in the house can keep someone scrolling at 2am. The model does not care which, so the person using it has to.

Eyal himself saw this. In 2012 he proposed what he calls the Manipulation Matrix: two questions a maker should ask. Would I use this product myself? And will it help users materially improve their lives?¹² A maker who answers yes to both he calls a facilitator. Yes to helping but no to using it themselves is a peddler. Using it but not improving lives is an entertainer. No to both he calls a dealer, and he calls that exploitation.¹²

The questions are easy to answer yes to in a meeting, so I add one of my own: would the customer still want this after experiencing it? If they would regret it the next morning, it was not for them.

Irish and EU law have a view where children are involved, and it is worth being exact about whom it binds.

The Data Protection Commission published its Fundamentals for a Child Oriented Approach to Data Processing in December 2021: 14 principles for services directed at, intended for or likely to be accessed by children, meaning anyone under 18.¹³ Two of its suggested measures go straight at habit design. One says to avoid nudge techniques that encourage or incentivise children to give unnecessary information or to take privacy disrupting actions. The other, on data driven auto features, says to avoid personalised features such as autoplay and reward loops where a child's personal data is used to support them.¹³ The Fundamentals are the DPC's guidance on how it reads the GDPR, not a separate law, but the DPC is the regulator that enforces the GDPR in Ireland.

Underneath sits the GDPR itself. Recital 38 says children merit specific protection because they may be less aware of the risks, in particular when their data is used for marketing or for building personality or user profiles. Article 8 sets the age at which a child can consent to an online service on their own at 16, which a member state may lower to no less than 13.¹⁴ Ireland kept 16.¹³ And the fairness principle in Article 5(1)(a) applies to every customer, of any age.¹⁴

The Digital Services Act adds a duty for online platforms. Article 28(1) requires providers of online platforms accessible to minors to put appropriate and proportionate measures in place to ensure a high level of privacy, safety and security of minors.¹⁵ The Commission's guidelines on that article, announced in July 2025 and published in the Official Journal that October, are explicit about engagement. They list streaks among the features that should be off by default for minors, and they ask that minors not be exposed to persuasive design features aimed predominantly at engagement, naming endless scrolling, autoplay, timed notifications, false urgency and virtual rewards for repeated actions.¹⁶ Coimisiún na Meán's Online Safety Code, for its part, covers video sharing platforms and deals with harmful content, age assurance and parental controls, not reward loops.¹⁷

So what applies to a small shop? Honestly, less than the headlines suggest. A coffee roaster selling to adults is not a service likely to be accessed by children. A shop that does not store and publish content from its users is not an online platform under the Act, and even a platform falls outside Article 28 if it is a micro or small enterprise, unless the Commission has designated it a very large online platform.¹⁵ The Fundamentals and the guidelines bind the platforms and the children's services. The fairness principle binds everyone, including a shop in Cork. And the reasoning travels: when the regulators for the most protected audience name streaks and reward loops as the problem, that is a fair hint about which features to leave out for adults too.

The manipulative techniques themselves, false urgency, shaming decline buttons, cancellation mazes and pre ticked boxes, are covered in detail, with the Irish and EU law each one breaks, in Dark patterns and Irish law. The short version: a habit loop that needs any of them is not a habit loop, it is a trap with good branding.

Worked example: three ideas, three verdicts

Shandon's marketing tool suggests three more features. I put each through the two Eyal questions and my morning after test.

1. A "brewing streak" badge that resets if the customer skips an order. The owner would not use it, it punishes a week away, and it pressures purchases nobody needed. Verdict: no.
2. A "Your next bag is ready to roast" notification sent at 9pm when a customer has not opened the app in a week. It is timed to inactivity, not to the bag running out. Verdict: no.
3. A one tap "Skip this delivery" button in every reminder. It probably lowers orders that month. It also tells customers they are in control, and a customer who knows they can skip without a fight is far less likely to cancel outright. That last part is my experience, not a study. Verdict: yes.

The good feature moves with the customer's real need. The bad ones move with the shop's metric and pretend otherwise.

A habit loop audit, step by step

This is the order I go through when a shop or a software product asks why people do not come back.

An example: run against Shandon's original journey, the audit moves the trigger from day 7 to day 5, cuts the action from seven steps to two, adds a guest coffee instead of a mystery discount, and adds a visible taste profile with a one step pause. The streak badge and the inactivity notification never ship. The cohort comparison says "probably better, keep watching", which is the honest answer most of the time.

If your question is about the first session rather than the hundredth, the moment someone first gets value from a product, that is a separate topic, and Onboarding and habit formation covers it. This article is about what happens after that, when the novelty has worn off and only the habit is left.

Footnotes

1. Wendy Wood and David T. Neal, A new look at habits and the habit goal interface, Psychological Review 114(4), 843 to 863, American Psychological Association, 2007, https://doi.org/10.1037/0033-295X.114.4.843

2. Benjamin Gardner, Phillippa Lally and Jane Wardle, Making health habitual: the psychology of habit formation and general practice, British Journal of General Practice 62(605), 664 to 666, 2012, https://pmc.ncbi.nlm.nih.gov/articles/PMC3505409/

3. Nir Eyal with Ryan Hoover, Hooked: How to Build Habit Forming Products, Portfolio Penguin, updated edition 2019 (first published 2014), https://www.penguin.co.uk/books/278201/hooked/9780241184837

4. Nir Eyal, The Hooked Model: How to Manufacture Desire in 4 Steps, nirandfar.com, https://www.nirandfar.com/how-to-manufacture-desire/

5. Nir Eyal, Hooked, introduction (publisher's sample), Penguin Business edition 2019, https://cdn.penguin.co.uk/dam-assets/books/9780241184837/9780241184837-sample.pdf

6. BJ Fogg, B=MAP, the home page of the model's site, behaviormodel.org, https://www.behaviormodel.org/

7. Nir Eyal, Variable Rewards: Want To Hook Users? Drive Them Crazy, nirandfar.com, https://www.nirandfar.com/want-to-hook-your-users-drive-them-crazy/

8. Wolfram Schultz, Peter Dayan and P. Read Montague, A neural substrate of prediction and reward, Science 275(5306), 1593 to 1599, 1997, https://doi.org/10.1126/science.275.5306.1593

9. Christopher D. Fiorillo, Philippe N. Tobler and Wolfram Schultz, Discrete coding of reward probability and uncertainty by dopamine neurons, Science 299(5614), 1898 to 1902, 2003, https://doi.org/10.1126/science.1077349

10. Phillippa Lally, Cornelia H. M. van Jaarsveld, Henry W. W. Potts and Jane Wardle, How are habits formed: Modelling habit formation in the real world, European Journal of Social Psychology 40(6), 998 to 1009, 2010, https://doi.org/10.1002/ejsp.674

11. Ben Singh, Andrew Murphy, Carol Maher and Ashleigh E. Smith, Time to Form a Habit: A Systematic Review and Meta Analysis of Health Behaviour Habit Formation and Its Determinants, Healthcare 12(23), 2488, MDPI, 2024, https://pmc.ncbi.nlm.nih.gov/articles/PMC11641623/

12. Nir Eyal, The Art of Manipulation, TechCrunch, 1 July 2012, https://techcrunch.com/2012/07/01/the-art-of-manipulation/

13. Data Protection Commission, Children Front and Centre: Fundamentals for a Child Oriented Approach to Data Processing, December 2021, sections 5.1 and 7.3, dataprotection.ie, https://www.dataprotection.ie/en/dpc-guidance/fundamentals-child-oriented-approach-data-processing

14. Regulation (EU) 2016/679, the General Data Protection Regulation, recital 38 and Articles 5(1)(a) and 8(1), Official Journal of the European Union, 2016, https://eur-lex.europa.eu/eli/reg/2016/679/oj

15. Regulation (EU) 2022/2065, the Digital Services Act, Articles 3(i), 19(1) and 28(1), Official Journal of the European Union, 2022, https://eur-lex.europa.eu/eli/reg/2022/2065/oj

16. European Commission, Guidelines on measures to ensure a high level of privacy, safety and security for minors online, pursuant to Article 28(4) of Regulation (EU) 2022/2065, C/2025/5519, sections 2, 6.3.1 and 6.4, announced 14 July 2025, Official Journal 10 October 2025, https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=OJ:C_202505519

17. Coimisiún na Meán, Online Safety Code, cnam.ie, https://www.cnam.ie/online-safety-code/