Murf AI: building a video library out of the questions users were stuck on

Client

Murf AI, AI voice generation platform, San Francisco

Founded

2020, by three IIT Kharagpur alumni

Scale

10 million+ users, 300+ Forbes 2000 companies as customers

How it began

Inbound enquiry, referred by a previous client

Volume

250+ minutes of feature tutorials, 20 to 30 minutes of social

Team

One project manager, two pods, 2 graphic designers and 4 animators

Cadence

Monthly delivery, 2024 into mid 2025

Language

English

Murf’s problem was never getting people in

Murf’s core promise is easy to grasp, because the saving is visible before you have finished reading the pricing page. Professional voice-over is expensive. At standard US rates a voice artist charges between $30 and $55 per finished minute for e-learning narration, and most will not take a job under $350. A single ten minute training module therefore costs somewhere between $300 and $550 in talent fees before anyone has recorded a second take. Murf’s entry plan lists at $29 a month.

When the economics are that stark, acquisition takes care of itself. Murf had built a strong product, the interface was designed around the user, and there was a written FAQ covering the platform. On paper a new customer already had everything they needed.

The difficulty showed up after signup. Murf is a hands on, multi step tool, and a written FAQ has two structural weaknesses when it is asked to teach one. The first is that reading an instruction and performing it at the same time is genuinely difficult. The user has to hold half a dozen steps in their head, switch tabs, and translate a sentence into an action on a screen that never looks quite like the one being described. The second weakness does more damage. A written FAQ only works for someone who already knows what to search for, and a customer who cannot get a feature to behave usually cannot name what is going wrong. They never find the article that would have helped, because they do not know what it is called.

The result was a slow leak. Users would reach a feature they could not make work and, rather than asking, some of them simply stopped coming back. Murf confirmed it afterwards in their own words: the drop-off was real, and the videos noticeably reduced it.

That is expensive in two ways at once. There is the revenue that walks out, which at least shows up in a dashboard, and there is the reputational cost of a customer who left frustrated, which never does. Neither corrects itself with time. A drop-off caused by confusion will keep happening to every new cohort of users until somebody changes the way the product is explained.

The brief contained a contradiction

The videos had to do two things that pull against each other. They had to be instructional enough to genuinely teach a feature, and engaging enough that nobody clicked away halfway through.

That is harder than it sounds. Thorough instruction tends to be slow, because thoroughness means covering the exceptions and the edge cases. But a tutorial that loses its viewer at forty seconds has taught nobody anything. It has simply moved the drop-off out of the product and into the video, which is worse than where it started, because now the customer has failed twice.

Make it a cake walk.The internal brief

We started by getting stuck ourselves

We did not begin from Murf’s feature list. Our project lead opened an account and used the platform the way a new customer would, not testing it for faults but genuinely trying to produce something with it. Every point where he had to stop and work something out, he wrote down as a question, phrased the way he would have phrased it in that moment.

That list was then set against the questions Murf’s own team was already fielding from customers, and the two overlapped. The places where an experienced video producer got stuck turned out to be broadly the same places customers were getting stuck.

Those overlapping questions became the script titles. Not the feature names from the product documentation, but the questions themselves, worded the way a confused user would word them. That distinction does real work. It means a customer looking for help finds a video whose title matches the sentence already in their head, which is precisely what the written FAQ could not do.

Our approach: show the payoff before the process

Most product tutorials open by naming the feature and then listing the steps. We inverted that.

Every video opens with the finished result. Before a viewer is asked to follow a single instruction, they see what the feature actually produces, the output they could have in front of them within a couple of minutes.

The reason is motivational rather than instructional. A viewer who has seen what they are about to be able to make has a reason to stay for the steps. A viewer who has only been told which buttons to press has been handed a task rather than a reward, and tasks are easy to abandon. Putting the payoff first turns the instruction that follows into the route to something the viewer now wants, which is the difference between watching a tutorial and finishing one.

Everything after that opening is deliberately plain. One feature per video, start to finish, with no detours into adjacent features. A user arrives at a given video because of one specific problem, and that video’s only job is to solve it and let them get back to work.

Two libraries, doing two different jobs

The work split into two streams, each handled by its own pod.

• Murf Academy took the larger share, growing past 250 minutes of feature tutorials. It was built as a library rather than a course. Each video answers one question, so it works the way a reference works. You do not watch it, you consult it.
• The social series ran to 20 or 30 minutes in total and had a different job. Timeless Tales, a short storytelling series narrated entirely in Murf voices, took a small piece of history and told it properly. It was aimed at people who were not stuck at all, because they had no idea the capability existed. Rather than explaining what the product could do, it demonstrated it: the argument for the voice quality was the voice doing the work.

That distinction matters more than it might appear. One library catches the customer who has hit a wall and is close to giving up. The other reaches the customer who is quietly under-using the product and will leave later for reasons they will never quite be able to articulate. Those are different people at different moments, and one video series cannot serve both of them well. So we did not ask it to.

What came of it

The most useful evidence is what Murf did next, repeatedly. This was not a single batch of videos. It ran as monthly delivery from 2024 through to the middle of 2025, with Murf commissioning the following month’s work each time the previous month’s landed.

For a client under no obligation to continue, choosing to renew month after month is a clearer signal than most numbers we could quote. Every month was a fresh decision to keep going, made by people who had just watched the previous month’s videos.

“High on Tales created tutorial videos for Murf Academy and creative social media videos that showcased our AI voice quality through engaging storytelling. The tutorials cut new user learning time from days to hours and noticeably reduced drop-offs. Project management was seamless, every deadline met, feedback acted on quickly, and communication always clear. A truly hassle-free partnership.”

Murf put two numbers on it themselves. New users were getting to grips with the product in hours rather than days, and the drop-off that started the whole project had noticeably fallen.

The company itself was moving quickly through the same stretch. When Murf raised its $10 million Series A in September 2022, it had just synthesised its first million voice-over projects. By November 2025 it was reporting six million users across 195 countries, naming more than 300 Forbes 2000 companies as clients, and had been listed by G2 among the fastest growing products of the year. The homepage today says ten million and counting.

We are not going to stand here and claim that four million of those people arrived because of our tutorials. That is the arithmetic this industry does far too often, and it does not survive thirty seconds of scrutiny.

What we will say is smaller, and true. A product growing at that rate is onboarding an enormous number of people who have never seen it before, and a fair few of them, somewhere in a confused first afternoon inside Murf Studio, went looking for help and watched something we made. We are quietly pleased about that.

What the library does now, with nobody attending to it

The videos are still there, still working. That is the quiet economics of customer education video, and the reason it tends to outperform its budget over time.

A written FAQ has to be rewritten every time the interface moves. A support agent answers the same question for the four hundredth time and is no faster at it than the first. A good walkthrough gets made once, and then it answers that question for every new user who arrives, at three in the morning, in whatever time zone they happen to be in, without anyone being paid to be awake. Murf commissioned the library across 2024 and 2025. It has been earning its keep every month since.

If this sounds like your problem

Most SaaS teams already know which features their customers get stuck on, because their support inbox tells them every week. The gap is rarely knowing. It is having something to send.

Tell us the three questions your team is tired of answering, and we will show you what they look like as video. If the answer turns out to be a better help article rather than a walkthrough, we will say so. We do this kind of work for technology and SaaS companies alongside our explainer video and custom eLearning practices.

Ravish Bohare, Founder, High on Tales. Led delivery on this project.

Talk to us about your product videos All case studies