Running lightweight usability tests

0:00 Running lightweight usability tests

Here's the fastest way to humble a marketing team. Sit them down and let them watch one real customer try to use their website. Just one. Within five minutes, someone will whisper, but it's right there! And someone else will realize the customer doesn't know what our key product term even means. Usability testing is watching real people try to complete real tasks. It's the most reliable way to learn why people struggle, and it doesn't need a lab or a big budget. In this lecture, you'll learn how many people to test, how to plan and write tasks, how to facilitate, recruit ethically, and turn observations into fixes.

0:47 Why it matters

Why does this matter? Because analytics can tell you that sixty percent of mobile visitors leave at step two. It can't tell you why. Usability testing tells you why. Maybe they can't find the delivery date. Maybe they think the price is per month when it's per year. Here's an analogy. Analytics is a smoke alarm. It tells you there's a problem somewhere in the building. Usability testing is walking into the room and seeing what's actually burning. You need both, but only one tells you what to fix.

1:26 How many participants?

How many people? A widely cited guideline from usability research is that testing with around five users per round reveals many of the most common problems, especially for one fairly similar audience. The key idea is small, repeated rounds, not one big study. Test five, fix what you find, then test again. If you have distinct audiences, like customers in different countries or languages, test a few people from each group. Five people is enough to find problems. It isn't enough to measure percentages precisely, so don't report it as a statistic.

2:06 Plan the test

Planning. Write a one-page plan. The goal: what do we want to learn, like can mobile users book a consultation. The prototype or page, and the device. Participants: who, how many, how you'll recruit them, and the incentive. Three to five realistic tasks. The metrics: task success, time, errors and a confidence or ease rating. The method, moderated or unmoderated, remote or in person. Consent and data handling. And the schedule. Sessions of twenty to forty-five minutes are typical. Then write a screener, a short set of questions to confirm people match your audience without revealing the right answers.

2:49 Writing good tasks

Writing good tasks is where most tests succeed or fail. Tasks should be realistic scenarios that don't give away the answer. Poor: click the pricing link in the menu. Better: you're considering this service for your salon, find out how much it would cost for three staff members. Notice the better version doesn't use the word pricing. If your task uses the exact words on the interface, you're testing whether people can match words, not whether they understand your site. Give people a reason, a context, and a goal, and then let them find their own way.

3:31 Facilitating well

Facilitating well. Ask participants to think aloud, saying what they're thinking as they go. Stay neutral. Don't help, and don't defend the design. If someone asks should I tap here, answer, what would you do if I weren't here? Probe gently: what did you expect to happen? What made you choose that? Focus on behavior over opinions. What people do matters more than I like it. And take structured notes, ideally with a second person as note-taker, so you can focus on the participant. The hardest part for most marketers is silence. Let people struggle for a moment. That's where the learning is.

4:16 Recruit ethically

Recruiting, ethically. Good sources: existing customers, email subscribers and social followers, screened to match your audience. Paid recruitment services for specific profiles. Friends and colleagues only for very early checks, because they aren't your audience, and they'll be polite. Offer a fair incentive, like a voucher or payment, and respect people's time. Get informed consent for recording, and explain how you'll store and use the data, in line with privacy laws. And include people with disabilities and a range of devices when you can. Moderated or unmoderated? Beginners should start with moderated remote tests over video calls.

4:58 Worked example 1: salon software

Worked example one, simple. A salon booking software company tests its pricing page with five salon owners over video calls. Task one: find out what it would cost for three staff members. Four of five can't work it out. One says, is this per salon or per user? Two people scroll past the book a demo button because a sticky banner covers it on their phones. Two ask what happens if it doesn't work for them. The fixes: a worked price example for three staff, the button moved higher with better contrast, and a refund FAQ right beside pricing.

5:41 Worked example 2: tutoring service (illustrative)

Worked example two, a business scenario with illustrative details. An online tutoring service in Karachi tests its sign-up flow with five parents on their own phones, over video calls. Three findings. Parents don't understand the difference between two plans. Most miss the free trial class option entirely. And for one parent, the WhatsApp contact button opens a broken link inside a social app's browser. The team fixes all three within a week: plan names based on outcomes with a one-line comparison, the free trial as the primary button, and a working WhatsApp link. A second round with four parents confirms the improvements.

6:25 Watch me do it: remote moderated test

Watch me do it. I send a four-question screener to newsletter subscribers and book five thirty-minute calls. In each session, I read the intro script: we're testing the website, not you, please think aloud, may I record? I wait for a clear yes. I give task one as a scenario. After each task, I ask how easy was that, from one to five. My colleague fills in the note sheet: participant, task, time, observation, verbatim quote, success and a severity guess. Within an hour of the last session, we list the top three issues. Then I paste the anonymized notes into an approved AI assistant for a first summary, and verify every quote against the recordings.

7:15 Common mistakes and measures

Common mistakes. Testing with colleagues instead of target users. Leading tasks that give away the answer. Helping participants the moment they hesitate. Reporting five-person results as percentages. And the worst: collecting findings and never acting on them. How do you measure success? Compare rounds. The same tasks after fixes should show higher success, fewer high-severity issues and better ease ratings. Then look for the matching live signals: higher completion on the relevant step, and fewer support questions on the same topic.

7:50 Recap and try this now

Let's recap. Watch real people try realistic tasks. Run small rounds of about five participants, fix what you find, and test again. Write scenario tasks that don't give away the answer. Stay neutral, use think aloud, and focus on behavior. Recruit representative people ethically, with consent and fair incentives. And turn every observation into a prioritized fix. Try this now. Write a test plan with three scenario tasks for one of your pages. Recruit three to five people from your audience, run the sessions remotely, and complete a findings table with severity and fixes.