Most people who can recite the HEART acronym could not build a HEART table if you gave them a product and 30 minutes. The letters are the easy part. The useful part is a three column grid that forces you to say what you want users to achieve, what behaviour would show it, and which number you will actually track. Skip the grid and HEART turns into a list of five buzzwords that sounds organised in a product interview and helps nobody on the job.
This explainer starts from zero, builds one complete example with numbers and ends with how to talk about the framework when an interviewer asks you to measure a feature.
Five categories and three columns, defined
HEART comes from a 2010 paper by Kerry Rodden, Hilary Hutchinson and Xin Fu, researchers at Google, presented at the CHI conference under the title “Measuring the User Experience on a Large Scale: User-Centered Metrics for Web Applications”.
The authors wanted metrics that described the experience of using a product, because the numbers teams usually tracked described the system instead. They named those older metrics PULSE: Page views, Uptime, Latency, Seven day active users and Earnings. PULSE tells you whether the service is running and growing. It says little about whether people find it good to use.
The five HEART categories, in plain terms:
Happiness: how users feel about the product, measured through attitudes such as satisfaction, perceived ease of use or likelihood to recommend, usually by a survey.
Engagement: how much and how often people use the product over a period, for example the number of visits per user per week or the share of users active on 5 of 7 days.
Adoption: how many new users start using the product or a feature in a given period, such as accounts created this month or first uses of a new export button.
Retention: how many users from a given period are still active in a later one, for example the percentage of March users who were active again in April.
Task success: whether people can complete key tasks efficiently and correctly, captured by completion rate, time on task and error rate.
Goals, Signals, Metrics: the process that turns any category into something measurable. A goal is what you want users to achieve, written without numbers. A signal is the behaviour or attitude that would show the goal is met. A metric is the signal turned into a number you can track over time, often as an average, a percentage or a rate.
The paper also makes a point beginners tend to miss: you do not need all five categories for every product. You choose the ones that match what you are trying to improve.
Goals, signals, metrics: why the grid matters more than the letters
Teams that skip goals go straight from a category to a metric they already have on a dashboard. “Engagement? We track sessions.” That shortcut quietly lets the existing data decide what success means. Writing a goal first forces an argument about the user, which is where the disagreements should happen.
A goal must be about the user and readable by anyone. “Users can find the right file without asking a colleague” is a goal. “Increase search usage by 10 percent” is not, because it already picked the metric and the target. Many teams get stuck here for 20 minutes, and that time is well spent: if four people write four different goals for one feature, they would have argued about the dashboard later anyway.
Signals bridge the gap between a feeling and a log file. For the file search goal, a signal could be “users open a result from the first page of results”, or “users stop searching again for the same term within 2 minutes”. Each signal has to be something your instrumentation can capture or something a survey can ask. If neither is possible, the signal is a wish.
Metrics are the signals made comparable. “Share of searches where the user opens a result in the top 5, per week” is a metric. So is “median seconds from query to first result opened”. A good metric moves when the experience changes and stays put when it does not. A metric that jumps every time marketing sends an email is measuring marketing.
Two warnings from practice. First, a single signal can feed several categories, so expect overlaps and resolve them by asking which goal the number really answers. Second, ratios beat raw counts nearly every time. Total searches grow with user numbers; searches per active user tell you something about the experience.
Worked example: a meeting room booking feature in a workplace app
Imagine a workplace app used by a 2,000 person company. The team has shipped a new meeting room booking flow and wants to know if it works. These numbers are illustrative.
| Category | Goal | Signal | Metric |
|---|---|---|---|
| Happiness | Booking feels quick and unstressful | Rating in a 1 question survey after booking | Average rating on a 1 to 5 scale, monthly |
| Engagement | People rely on the app for regular meetings | Repeat bookings by the same person | Bookings per active booker per week |
| Adoption | Staff who book rooms move to the new flow | First booking made in the new flow | Share of all bookers who used the new flow this month |
| Retention | People who try it keep using it | Booking again in the following month | Percentage of October bookers who booked in November |
| Task success | Booking a room is fast and error free | Bookings completed, cancelled within 5 minutes, time to book | Completion rate, quick cancellation rate, median time to book |
Now the numbers. In the first month, 900 of the 1,500 staff who book rooms try the new flow, so adoption is 60 percent. They make 4,500 bookings, which is 5 per active booker per month, or about 1.2 a week. Of the 900 October bookers, 702 book again in November: retention of 78 percent.
Task success is where it gets interesting. Completion rate is 94 percent, but 11 percent of completed bookings are cancelled within 5 minutes and rebooked, almost always for a different room. The median time to book is 41 seconds against 70 in the old flow. The survey average sits at 3.9.
Read together, the table says the flow is faster and people come back, but roughly one booking in nine goes to the wrong room. A team tracking only adoption and retention would celebrate. The quick cancellation rate points to the next fix: the room list probably does not show capacity or equipment clearly enough.
Building a HEART table in one working session
You can fill a first version of the table in about an hour with the product manager, a designer, an engineer and someone who knows the analytics setup. The procedure:
- Write one sentence on what the product or feature is for, from the user’s side. Everyone must agree on it before moving on.
- List the five categories and cross out the ones that do not apply. A one off tool used once a year gains little from Engagement; an internal admin page may not need Adoption.
- For each remaining category, write one goal without numbers. Allow 5 minutes per goal and expect to rewrite at least one.
- For each goal, list 2 or 3 candidate signals, then keep the one you can measure today or with one small tracking change.
- Turn each signal into a metric with a time window and a denominator, for example “per active user per week”, never a raw total.
- Check every metric against a simple test: would it change if the experience got worse while traffic stayed flat? If not, replace it.
- Record the current baseline for each metric before any change ships. Without a baseline, the table is a plan, not a measurement.
- Agree on one metric that leads the discussion and treat the others as context and guardrails.
Step 8 matters because a table of 7 equal metrics produces 7 debates. The original paper treats HEART as a way to choose metrics, and choosing means some rows matter more than others.
Which categories to keep: a decision tree
Beginners often fill all five rows because the acronym has five letters. A short decision tree prevents that:
- Is the product or feature new, or being rolled out to people who used something else before? If yes, keep Adoption. If no, drop it or treat it as background.
- Is the product meant to be used repeatedly, at least monthly? If yes, keep Retention and Engagement. If no, keep neither and focus on Task success.
- Does the user come to complete a defined task, such as booking, paying or uploading? If yes, keep Task success, and measure errors as well as completion.
- Is there a reason to think people complete tasks but dislike the experience, for example frequent support complaints? If yes, keep Happiness and run a short survey. If no, Happiness can wait until you have the budget to survey properly.
- After questions 1 to 4, do you have more than 3 categories for a single feature? If yes, rank them and keep the top 3 for this cycle.
Applied to the booking flow, all five survived because the feature is new, used weekly, task based and replaced an old flow people grumbled about. A settings page for changing a password would end with one category: Task success.
The tree also works in reverse when a dashboard already exists. Take every metric the team currently reviews, place it in a category, and see which rows are empty or crowded. A common result is eight engagement numbers, one adoption number and nothing on task success. That imbalance usually reflects what was easy to log in 2021, not what the product needs to learn in its next quarter.
Using HEART in a product interview, and where it falls short
Interviewers ask “how would you measure the success of feature X?” to see whether you can connect a user problem to a number. Naming HEART earns you little. Using it well earns a lot.
A strong answer picks the 2 or 3 relevant categories out loud and explains why the others are dropped, writes one goal per category in user language, names a signal and a metric with a denominator for each, picks a lead metric and one guardrail, and says what result would make you roll back. That takes about 3 minutes to say and shows the reasoning the interviewer is grading.
HEART has real critics, and an interviewer may push on them. The framework does not tell you which category matters most for the business. It has nothing on revenue, so a commercial product needs a separate link to money, which is where a North Star metric or simple unit economics comes in.
Happiness surveys are noisy: response rates are often low and the people who answer are not a random sample. Engagement can be the wrong goal entirely; for a tax filing tool, more time in the app is a sign of failure. The authors themselves treated the categories as a menu, so using all five by default misreads the source.
My verdict on the trade-off: HEART is worth using whenever a team has to judge a user experience at scale and the default dashboard only counts traffic, because the goals, signals, metrics grid makes people agree on what good looks like before arguing about numbers.
It is the wrong tool when the question is commercial, such as pricing or conversion to paid, and a weak tool when you lack event tracking for the signals you choose. Use it to pick metrics, then let the business goal decide which one leads.