The rise and limits of the Carolyns

How former language teachers took over the world just to find the talking gap at the other end

I started studying the market segment of the small businesses for language learning with the purpose of detecting a market niche during the course of the Founder Institute program for startups at Frankfurt, which was extremely valuable.

Online language learning has exploded during the last years: from a few language apps, like Duolingo, to a wealth of YouTube channels, hundreds of blogs and podcasts, you name it. There are a few huge language businesses offering their polished apps like, well, Duolingo, but also Mosalingua, Rosetta Stone, Mondly, Babbel, Busuu, and more, but what struck me was the sheer quantity of language products offered by companies composed of a small team –or even a single person.

Those small language businesses started when a language teacher was fed up about exchanging time for money (that is, being paid by the hour, either by a boss or by a client), so they decided to start posting videos on YouTube, or short clips on Instagram or trying to monetize the blog that they were writing for the students anyway. It’s nice to receive some money from YouTube ads or affiliate programs on the blog, though generally, it’s not enough for relying on that for a living.

Sometimes former teachers didn’t stop there. Later on, they started selling digital courses, which gave them a more substantial income, so in the end, the language side hustle ended up being their main income.

As incredible as it sounds, there is a very polished recipe for putting in place a language learning business, that you can read in this blog post, as a sequence of steps that you can follow one by one. It’s not an easy way, though, and you need to put a lot of work into it to be successful.

I called the former language teachers (now language business owners) the “Carolyn”s, just because the most successful ones happen to be female –it’s not sort of patronizing in any way, and they have my utmost respect and admiration. One of them, Shayna Oliveira, built the amazing “Espresso English” collection of English digital courses, built all by herself, with no employees until recently. I bought her “Shadowing with Shayna” course, which helped me a lot in improving my English intonation and rhythm. Despite additionally raising her babies, she even managed to have the excellent entrepreneurial podcast “Entrepreneurs in Motion”, useful not only for language businesses.

Well, when I started the Founder Institute program, I thought there were a few hundreds of Carolyns, but as I deepened my research, I found thousands more of them, and that for English only; if you add all the other languages, their number is just mind-boggling. And I’m sure I just saw the tip of the iceberg.

Each Carolyn has on average around 2,500 customers, which is not the same as their following (except for the case of YouTubers, such as the British Lucy), and they make often above 10 grand per month. This is of course not a fortune, but you can make a living out of it. And it’s not just any business, they can make more money than in an average professional job, doing something they love. They also help their students to learn languages in a more modern and flexible way than the traditional classroom-based school, so they make a positive impact in the world. They are liked and trusted by their students, as they give not only information but also guidance and encouragement to them.

At this point, you can guess what was that made possible the transformation of former teachers to lifestyle-business owners: it was… the Internet, of course! The transition from uploading material for just their students to selling (and then promoting) their own courses is rather smooth once you have access to the Internet platforms: web pages, email lists, online payments, YouTube, social media, etc.; this doesn’t mean it’s easy, but with hard work, organization and common sense it’s doable and there is a clear path for achieving relative success (see the above link).

From the mere existence of the Carolyn legions, we can smell the possibility of business opportunities: any service (preferably a SaaS) useful for a big percentage of Carolyns is a clear business opportunity because they face hundreds of friction points while traveling their path to success. Anything, literally anything you can offer them that removes friction points, could be a B2B. For instance, pioneer Carolyns had to build their webpage the hard way, dealing with WordPress technicalities; now anyone can set up a page from templates with Wix or Squarespace, so that’s one friction point removed. Some of them offer their courses directly inside a course-selling platform like Teachable.

One wonderful example of friction reduction in language teaching is Feedback Panda, a small SaaS for English teachers working as freelancers for VIP Panda (English live lessons for Chinese kids). Arvid Kahl’s wife, Danielle, was such a freelancer and had the friction point that after each lesson she had to provide feedback about the lesson, but the company’s software didn’t make it easy, so Arvid developed a basic software product that expedited the feedback task. Customer need validation was automatic, as Danielle felt firsthand the problem, and could tell Arvid with no delay if the product fitted the need or not. Danielle was the main promoter of the product on social media because she was already in the relevant forums! Long story short, Feedback Panda became so successful that it was sold after a couple of years of operation.

Another friction factor the Carolyns face in their businesses is that the content they provide gives very little help in one of the main aspects of language learning: talking practice!

Do you really learn to speak a language by watching instructional videos? I bet you don’t. Watching videos can make you familiar with the language, learn words and phrases, but…. What about talking?


From the 4 language skills (reading, listening comprehension, writing, talking) the latter is the most:

1.- Desired. Those who want to learn a new language visualize themselves speaking it.

2.- Difficult. Talking involves a lot of sub-skills, from phrase construction to pronunciation and prosodic skills (intonation, rhythm).

3.- Underserved. After more than 10 years of work, the Carolyns (and also the big schools) have flooded the internet with content, but the speaking development tools are far from satisfactory.

Why do we say that the market for speaking skills is underserved? There are currently two main solutions offered to the public:

1.- Private lessons with a tutor, with platforms like iTalki. While extremely effective and easy to use, this is a very expensive solution for the general public.

2.- Language exchanges, which are mostly free in platforms like Tandem, but have plenty of friction, from how hard it is to agree on a schedule (and then stick with it) to how awkward it feels for shy people to choose a partner in a session that feels like a date.

A few Carolyns have set up practice sessions using Zoom or an equivalent system, with mostly good results (e.g. Adriana Lucic, “English with Adriana”), but though this solution is well known to Carolyns, it’s also very time consuming and leaves too little revenue, so they think twice before offering such a service to the public. Shayna Oliveira comments on how she offered a subscription service only to shut it down shortly after.

We call “the talking gap” the current situation of scarcity of talking practice resources, compared to the wealth of online content available (YouTube videos, courses, podcasts, books, etc.)

Are there or could be other possible options (convenient ones, we mean) for talking practice? The answer we think is affirmative, but the solutions are not going to be created by the Carolyns; they are not going to create the needed tools for talking practice because this time it’s not a matter of leveraging the Internet “as it is” and just uploading content; you need to develop more complicated tools using advanced technologies. This is something that former language teachers just can’t do.

Growing an audience incrementally, one follower a day is within the Carolyns reach, but to build an AI tool for matchmaking is a different beast. Here we reach the Carolyns limit. They are so stuck with speaking practice that they ask their students to do “self-talk” as a way to practice talking, like “I am doing the dishes”, “My dog is barking”, and so on, preferably aloud.

Well, we can ask what would be some suitable, convenient tools for talking practice? A convenient tool would be:

  • Cheap to the final user (at least compared to private tutors).
  • Available anytime (that is, no need for scheduling).
  • Interactive (when you speak you get a response back or at least an evaluation about how you are doing, as opposed to self-talk).

Some hypothetical products that would fill the bill are:

1.- Voice assistant teachers: This would be an extension of an assistant in the style of Siri, Alexa, or Google assistant, but armed with scripts to give kind of “language lessons”. The assistant will not only converse, but also give corrections to the student, detect pronunciation flaws and give advice for correcting them, give encouragement to the student, adapt the material and methods to the student profile, provide bits of humor, and more. This would be an extremely powerful solution, and I think that in a decade or so this will be the beginners’ default language learning method (because the replication cost is essentially zero), but for sure voice assistant teachers won’t be ready for primetime in the next few years, so I won’t discuss it further.

2.- Human-powered talking online practice in virtual rooms, which could take several forms:

  • Video meetings, like Airmeet, which is a good meeting product, but in order to be effective as language practice needs to incorporate some match-making tools (more on this later). Also, people are kind of fed up with Zoom-like meetings due to the pandemics.
  • VR-based environments with virtual rooms populated with avatars representing people, like Spatial, but for the rest would work very much like the first option (without the hassle of dressing up, setting a tidy place, etc.).
  • Voice-only virtual rooms: after the success of Clubhouse voice-based meetings have become a thing, but for all accounts, meetings are just like the two other options, minus the graphic clutter; the advantage is that you don’t need to stare at a screen in order to participate in the virtual meetings. Voice VR is not exactly like a phone call, as some systems include 3d sound, which promotes immersion and makes it easy to identify who is speaking.

We can see that the bottleneck for human-powered talking online practice is the matchmaking part: which people should be in each small group so that the conversation is, first of all, understandable, and also engaging, and useful for language learning purposes.

Current systems leave the matchmaking burden to humans, either the organizer (like in Zoom “breakout rooms”) or the participants (in Airmeet), but in either case, the solution is not scalable (imagine a participant trying to choose between 10,000 virtual rooms).

But the matchmaking could be made automatic as well (a master Thesis by one of my students proposed a solution as a multiobjective optimization problem, which falls into a branch of Artificial Intelligence), and an AI system could group people together in the most suitable way for maximizing aspects such as:

  • Interest compatibility: conversation tends to be more engaging when the participants have common interests to speak about.
  • Level compatibility: it’s not a good idea to put a complete beginner in a group with very advanced students because the former feels like an idiot and the latter get annoyed. One study found out that it’s best to have most students of a certain level and include one of a slightly lower level.
  • Accent diversity: when all the participants have the same mother tongue background (for instance, all Chinese), their accent tends to be reinforced, so it’s better to include participants from as diverse backgrounds as possible.

Of course, each of the mentioned factors generates a problem to solve, and perhaps not a trivial one. For example, level compatibility implies that we have a way of determining the students’ speaking level.

There are some experimental automated systems for speaking level evaluation, based on Machine Learning, and I’ve developed myself a basic one with my students. The speaking evaluation service would be provided as a cloud-powered service to which the students would connect, so it can scale with almost no limits. Even as a standalone product, this could be useful and could even support a profitable business.

To wrap up, we can say that the lack of talking practice (the “talking gap”), which is the main limitation of the language teaching services the Carolyns can offer, could have some practical solutions that involve increasingly sophisticated technology that involves different kinds of Artificial Intelligence. At the beginning of the 2030 decade, the use of voice assistant teachers will be the default language learning method for basic levels, complemented with AI-enhanced talking practice online services.

Then what is going to be the Carolyns’ place in this new AI-powered world? We don’t know for sure, but if the new tech tools are simple enough to use, they will find ways of incorporating them into their always buzzing businesses. Hey, the internet was the basic requirement for the rising of the Carolyns, but it wasn’t their creation in the first place. In the end, all it takes is to be able to adapt to the new needs and tools. Isn’t this always the case for successful species?

Former AI researcher, CEO of Avalinguo

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store