HEART metrics: Google’s UX metrics framework as a table

Build your HEART table

Choose an example product, or start with an empty table. For each category, write the goal, the signal that shows it and the metric you track. Then copy the table as Markdown into your document.

What is Google’s HEART framework?

HEART comes from the paper “Measuring the User Experience on a Large Scale” by Kerry Rodden, Hilary Hutchinson and Xin Fu of Google. The definitive version appeared in the Proceedings of CHI 2010. The conference took place from April 10 to 15, 2010, in Atlanta.

First, the authors gave the common metrics a name: PULSE. It stands for Page views, Uptime, Latency, Seven-day active users and Earnings.

According to the authors, these measures are important, but they measure the user experience either at too low a level or only indirectly. If page views for a feature go up, that can mean two things. Either the feature is genuinely popular, or confused users are clicking around looking for a way out.

The count of seven-day active users also doesn’t distinguish between new and returning users. According to the paper, this number could in theory still rise even in the worst case, where the user base turns over completely every week.

HEART was built to close this gap. The paper presents it as a framework that complements PULSE.

Five categories: Happiness, Engagement, Adoption, Retention, Task success

The letters in HEART aren’t metrics but categories. Each team defines the metric that fits its product within these categories.

Happiness

Attitudinal measures: satisfaction, visual appeal, likelihood to recommend, perceived ease of use. In the paper’s example, the iGoogle team first saw a drop on its 7-point satisfaction scale after a major redesign. The score recovered over time, and the team kept the new design.

Engagement

The frequency, intensity or depth of interaction. The Gmail team chose the share of active users who opened the product on 5 or more days in the last week. The paper recommends reporting these values as an average per user rather than as a total.

Adoption

New users who start using the product in a given period. The example in the paper is the number of accounts created in the last 7 days.

Retention

Users from one period who are still using the product in a later period. Example: the share of a week’s seven-day active users who are still active three months later.

Task success

Efficiency, effectiveness and error rate. The Google Maps team tested a single search box instead of the dual one in an A/B test. From the error rates, they saw that users adapted to the single box.

Read Adoption and Retention together. During the stock market meltdown in September 2008, both page views and seven-day active users surged at Google Finance. With these two measures, the authors separated whether the surge came from new users or from existing ones checking their investments.

You don’t have to use every category. The paper writes that Engagement may not be meaningful in an enterprise product, because users open the product for their work anyway. For a single feature, Engagement can still be useful.

The “Hide empty rows” button in the tool brings this decision into the table.

Goals, signals, metrics

The second half of HEART is a process. According to the paper, a metric is unlikely to be useful in practice unless it explicitly relates to a goal. The process runs in these steps.

1. Goal. Write down what the product or feature should achieve for the user experience. Team members may disagree, and this step is the place to reach agreement. The goal of a feature can differ from the overall goal of the product. Don’t get too distracted at this stage by whether it can be measured.
2. Signal. What changes in users’ behaviour or attitude when the goal is met? Choose the data source here too: logs or a survey? The signal should move only when the experience gets better or worse. Sometimes failure is easier to see: abandoned tasks, undo actions, frustration.
3. Metric. Turn the signal into a number that can be tracked over time on a dashboard. Raw counts grow with the user base. Ratios, percentages or averages per user are often more useful.

The authors write that they applied the framework and process to more than 20 products and projects at Google. To me, the real value is that the team talks about the goal before it talks about metrics.

NPS, SUS and task success within HEART

HEART doesn’t exclude other measures; it gives them a place. The paper counts likelihood to recommend and perceived ease of use as part of Happiness. That makes the NPS calculator and the SUS calculator a good starting point for the metric in this row.

The Task success row bundles the classic behavioural measures. How to measure the share of tasks completed is covered in the guide to the task success rate. If you compare two versions, use the A/B test calculator to check whether the difference is chance.

My note

What I see most often in audits: the dashboard has dozens of charts, but none is tied to a goal. The team sees page views rising, but can’t say whether that is good or bad. A HEART table exposes this gap in the first meeting.

When I build the first table with a team, I start with two or three categories. A team that wants to fill all five rows usually ends up tracking none. Striking a row with no goal from the table is a decision too.

I always remember one warning from the paper: metrics shouldn’t stand alone. The number tells you what changed; a usability test gives you the reason. If your metric is dropping and you can’t find the reason, I can go through every step of up to 10 flows in your application with the UX Audit Pro.