Image source: generated by Unbounded AI

Latest news, AI English learning product Speak announced the completion of US$78 million in Series C financing, bringing the total financing to US$162 million , the valuation reached US$1 billion. If I remember correctly, this should be the first real toC AI unicorn. Not long ago, in June this year, Speak just completed a B-3 round of financing of US$20 million, with a valuation of US$500 million. In other words, Speak’s valuation doubled in just half a year.

Speak’s financing history

In addition to the high amount of financing and doubled valuation, Speak’s investor lineup this time is also quite luxurious. The lead investor is top VC Accel, who previously led OpenAI Startup Fund, which invested in Speak’s B-1 round of financing, as well as Khosla Ventures, Y Combinator and other institutions also participated.

According to the exact information obtained from the investment internship, Speak’s ARR is currently close to 50 million US dollars, with an annual growth rate of 100%. As a product deeply embedded in AI technology, it is rare to be able to deliver such "beautiful" commercial results.

Market changes: Finding new growth markets in Asia, Europe and the United States are still making no progress

In June this year, we reported a valuation of US$500 million. Seize Japan and South Korea, AI language learning products become the top 3 in the track (Part 1)" I observed Speak in the topic selection, and the ARR data given by some sources at that time was 20 million+ US dollars. The recent surge in Speak’s revenue may come from two aspects, one is the expansion of the market, and the other is the launch of advanced subscription plans that can provide customized courses.

Let’s talk about the market level first. When we observed it in June, Speak’s revenue mainly came from the Korean market, which it has been deeply involved in since its establishment in 2017. The Korean market’s revenue will account for more than 90% until 2023. Verification of PMF in the Korean market was also an important reason why Speak was able to obtain financing before. After the B-1 round of financing at the end of 2022, Speak began to expand into the Japanese market.

Speak revenue data from 2021 to the present

From 2023 to the present, Speak’s total revenue in the Korean market is US$14.375 million, and the total revenue in the Japanese market is US$9.314 million. Approximately 64% of the Korean market. In addition to the app store, some users will also subscribe to Speak through the web. If the investment internship calculatedThere is no problem with the ARR data, so the payment amounts on the web and app are roughly the same, each accounting for 50%.

In addition to the two traditional markets of Japan and South Korea, Speak’s revenue in the Taiwan market began to grow rapidly in the second half of this year, with revenue reaching US$260,000 in the past 30 days.

Looking at the number of users, Speak’s “success” is limited to Asia. Apart from Japan, South Korea, and Taiwan, only the U.S. market currently has more than 10,000 DAU. Although Speak has launched a "Learn Spanish in English" course suitable for the U.S. market and has achieved certain user growth, it has not yet made good revenue. Speak's ARPU in the U.S. market is far lower than that of several major markets in Asia.

When it comes to learning, we also need to look at Asians, and we can also see that Asians want to escape from the current environment...

In the official statement on financing, Speak CEO said that this round Financing will be used to expand the number of supported languages, improve Spanish courses, and expand French courses. However, it still needs time to test whether Speak, which has achieved great success in the Asian market, can be accepted by users in the European and American markets.

Speak’s per capita usage time data｜Image source: Diandian Data

At the data level, another major change in Speak is that the per capita usage time has dropped significantly in the past year. . According to Diandian Data’s statistics on iOS user behavior, the average usage time of Speak in January 2024 can reach 1.14 hours/month. In October, this data dropped to 0.54 hours/month, a drop of more than 50%.

There may be multiple factors behind this data, and we are also trying to gain some understanding from the product level.

Product changes: gamification, emphasis on scenes

Compared with before, the homepage was directly the course page, now Speak has added Two functions, one is ChatBot and the other is quick practice. ChatBot is shown in the middle picture. In addition to regular communication with ChatBot, users can also let it give learning suggestions, assess English proficiency, play subtitle solitaire and question games, etc.

The quick practice section currently includes two exercises: word fill-in-the-blank and proverb fill-in-the-blank. By completing these exercises, you can also accumulate winning streaks. This design is a bit similar to Duolingo, and Speak is also trying to gamify it.

Both of these functions can be used without subscription, and the user paths are relatively short, which is more like a design made to penetrate fragmented time. And when the user uses other functions, ChatBot will also be presented on the right side in the form of a floating bubble (as shown below) for the user to recall at any time.

In the "Dialogue Practice" section, the main changesIn addition to regular topic conversations, Speak has added conversation exercises for several actual scenarios such as talking about work, preparing for IELTS, and Thanksgiving (common conversations on holidays). It is speculated that Speak hopes to reach some more purposeful users.

Product design pays more attention to fragmented time and accepts some more purposeful users who want to practice through AI. This may be one of the reasons for the sharp decline in Speak user usage time. But on the other hand, this can also expand Speak’s user base.

Speak’s user feedback function

Challenges and personal information first-level labels when testing Speak 1 year ago

The third major section, "Challenges", is One of the big changes is mainly the upgrade of the user behavior feedback mechanism. Although these upgrades draw heavily on Duolingo's successful experience, they are relatively "restrained." First, the user's winning streak progress will be displayed in the most eye-catching position on the homepage (the dialogue practice page will also be presented in the form of floating bubbles). In addition, compared with the previous challenge module only (which was not displayed during the test), Speak introduced a ranking function and detailed daily challenge tasks to better urge users to learn. However, neither the rankings nor the challenge system provide much "feedback" to users (such as Duolingo's stickers, trophies, skins and other rewards). There is still a question mark as to whether users will participate in rankings and tasks frequently.

Based on this, Speak CEO said that although the approach of enhancing product retention through gameplay like Duolingo has been proven effective, Speak is also actively learning from this idea. However, if there is a conflict between gamification, user participation and learning effects, we will 100% choose to ensure learning effects.

In addition, in terms of subscription benefits, Speak has added a higher-priced premium subscription. After subscribing to this benefit, AI will customize courses and learning plans according to the user's level, and try to match the user's needs. After the user completes the course or exercise, AI will also correct errors for the user in a targeted manner.

Although the author is unable to experience it personally due to the subscription price, judging from the purchase situation on the iOS side in the Japanese market in the past 30 days, this advanced subscription function, which is more than twice as expensive, accounts for half of the revenue. The proportion is not low, which also reflects that user acceptance is still relatively high.

The situation of iOS applications in the Japanese market in the past 30 days, the information comes from Diandian data, translated using ChatGPT

Generally speaking, there are three main ideas for Speak product iteration. One is to join Some more basic functions and tools, and more consideration is given to users’ actual application scenarios when designing courses. The second is to actively integrate gamification design to improve stickiness and retention. The third is to use personalized customized courses as a selling point based on large model capabilities. furtherExpand ways to commercialize products. Judging from the data feedback, except for the change in stickiness, the other two changes may have certain effects.

Write at the end

TalkAI speaking practice global dual-end revenue data (the Chinese market only has iOS data)｜Picture source: Diandian Data< /p>

Judging from Speak’s results and our previous observations, Asian users are quite willing to spend money on learning English, especially practicing speaking. Speaking practice products have great potential for monetization. Data shows that a domestic spoken language learning product "TalkAI Spoken Language Practice" has a revenue of US$1.32 million, or about 10 million yuan, on the iOS side alone this year, plus domestic Android and tablet products that are not included in the data. The revenue of other platforms can also reach tens of millions of RMB in ARR.