Annie Su, Analyst (蘇怜媛 / 分析師)

負責投資，表面上是個冷靜理性的金融人，但其實對創業有莫名的熱忱，立志成為台灣網路創業家與資本市場的橋樑。先前於凱基投顧擔任 Research Associate，主要負責產業研究和財務分析。台大財金系畢業，大學時期活躍於創意創業學程、不一樣思考社。

串流影音龍頭 Netflix，正在應用 AI 技術顛覆傳統的娛樂產業，透過個人化推薦，將不同但更適合的內容推送到個別用戶眼前。對 AI 領域的創業者來說，來自 Netflix 最重要的啟示，就是產品與服務未必要是人人追逐的大主流，相反地，讓多元、長尾的內容，為個別用戶帶來最大價值，才是 AI 時代最重要的顛覆力量。

Netflix 的娛樂帝國正在成型。五年來，Netflix 股價大漲十倍，並一度在今年六月市值突破 1,700 億美元，超越 Disney，儘管隨後股價回檔，但成長力道依舊強勁。今年第二季，Netflix 共新增 620 萬用戶，其中 510 萬來自美國以外的海外市場，根據 Cowen 預估，Netflix 海外用戶在 10 年內，將從 2018 年底的 8,360 萬，以每年平均 20% 的速度成長至 2.55 億。

Netflix 影音內容的質量與數量，也在同步提升。今年 Netflix 入圍 112 項艾美獎，首度超越歷年大贏家 HBO 的 108 項，顯示致力於拍攝原創的策略終獲碩果。根據《經濟學人》的報導，Netflix 今年的內容投資將達 120 至 130 億美元，超越任何一家電影公司、電視台 (不含體育頻道)，用戶每年可收到 82 部電影 (每年華納兄弟推出 23 部電影，Disney 則為 10 部)，正在製作或採購的電視節目則有 700 部，其中包括 100 部劇本和喜劇、數十部紀錄片和兒童劇、脫口秀喜劇，以及無劇本的真人秀和脫口秀節目。

Netflix 拍攝原創內容的標準是什麼？以及，Netflix 又是如何讓觀眾在上千種影音內容中，找到自己喜歡的影片？

重點不是最優質的內容，而是「每小時觀看成本」 

Netflix 的商業模式，奠基於向用戶收取固定月費。對公司來說，最重要的任務就是提供足夠 「優質」的內容滿足用戶需求，這樣用戶便沒有理由離開平台。而作為串流影音服務商，Netflix 並不像傳統電視台，受限於節目表以及每日播放 24 小時，而是能夠提供無限的影音內容供用戶隨時選擇，因此 Netflix 所認定的 「優質」內容，其實是針對個人而言的，因為他們能提供各種類型的內容 (無論大眾或小眾) 給長尾市場。

身為一間位於矽谷的科技公司，Netflix 經常被誤解為運用演算法來決定原創內容，但實際上，在內容創作 Netflix 給予創作者相當大的自由，且時常強調他們並不會根據用戶行為，來影響原創內容的方向，產品長 Ted Sarandos 曾提及「千萬不要沈溺於演算法，過去的資訊，很有可能限制對於未來的想像力。」

大數據與演算法對於 Netflix 來說，是運用在評估某個作品上架後，是否符合成本效益。公司在官方網站上明講：評估內容成效的關鍵指標是 「每小時觀看成本」(cost per hour viewed)，亦即 「這個內容是否能在一定的成本內，最大化用戶觀看時數」。由此可見，Netflix 想的並不是 「這個作品是否是最優質的內容」 或 「這個作品是否能吸引最多的觀眾數」，而是 「只要這個作品，能吸引一定數量級的觀眾群，並且符合每小時觀看成本的門檻」。

Netflix 在官網清楚說明挑選影音作品的標準 (圖片來源)

也因此，Netflix 的原創內容其實是相當多元的。從英國女王傳記《王冠》、懸疑驚悚片《怪奇物語》、現代科技反思路線《黑鏡》到青少年霸凌題材《漢娜的遺言》，不管是什麼類型、多少製作成本，只要成本效益指標夠好，都會被 Netflix 認為是適合製作的好內容。也因此，Netflix 製作影集成本的範圍很廣，《王冠》的每集製作成本高達 1,300 萬美元，而《漢娜的遺言》「只有」約 500 萬美元。相對的，一旦不符合 Netflix 的成本效益評估，公司便會決定停止該內容，例如 2017 年中宣告停拍《超感8人組》與《布朗克斯：街頭少年音樂夢》，這兩部影集只各拍了兩季與一季。

相較於 Netflix 的長尾策略，其他電視台則較傾向重壓資源在單一內容上，並預期能擄獲所有觀眾的心。HBO 的大製作《冰與火之歌》，號稱在最後一季的每集成本將創新高至 1,500 萬美元 ，另外一個作品《西方極樂園》的每集成本也據稱達到 1,000 萬美元。上述兩部作品，也於今年艾美獎各拿到入圍 22 項與 21 項的好成績，為入圍數最多的前兩名作品，相較之下，Netflix 的 《王冠》 只入圍了 13 項，但可別忘了 Netflix 是總入圍數最多的贏家，其他獎項，則是由另外 40 個作品拿下，證明了 Netflix 投資於各類型題材的模式。

Netflix 的影片推薦系統，讓你可以無腦挑影片

一旦擁有上千種可滿足各類用戶的優質內容，Netflix 另一件重要的任務，便是用更有效率的方式，把適合的內容，推薦給有興趣的人。精準內容推薦並不是一件新奇的事，Amazon、Facebook、Google 都是藉由用戶歷史行為資料，來推薦商品或產出個人化頁面，以優化使用者體驗。但這件事對 Netflix 尤其重要，因為在使用影音串流平台時，用戶並沒有很明確的目的要購買商品或搜尋資訊，大多數的時候，是漫無目的尋找能打發時間的內容，要是 Netflix 無法在短時間內精準推薦用戶喜歡的影片，用戶很容易就被別的平台或傳統電視吸引走。根據 Netflix 2015 年發表的文章，80% 的用戶觀看時數都是靠推薦而來的，也佐證了這一點。

為了滿足口味各異的用戶們，Netflix 一直致力於優化推薦演算法。在過去，Netflix 試圖去預測每個用戶對於每部影片的評價 (分數 1-5)，藉此推薦用戶可能有興趣的內容。不過隨著 Netflix 掌握更多用戶行為資料 (包括用戶觀看的內容、使用設備、觀看時間、觀看頻率、觀看地點)，現在更以機器學習 (Machine Learning) 來建立推薦演算法，以捕捉更多 rule-based 演算法可能漏掉，但對預測喜好相當有幫助的重要資訊，例如：觀看影片的順序、不同因素之間的交互作用。

讓 Netflix 演算法來推薦最合你胃口的影片內容

有使用 Netflix 的人都知道，Netflix 的首頁是由不同主題的影片列組成的，這些主題選擇、影片挑選、排列順序背後便是由不同的演算法驅動：

• Personalized Video Ranker, PVR (某類型影片)：推薦你喜歡的影片類型
• Top-N Video Ranker (最佳推薦)：推薦你喜歡的影片，和 PVR 的差別在於這邊沒有類型的限制
• Trending Now (現正熱播)：依據當下熱門話題，如聖誕節慶，推薦你喜歡的影片類型
• Continue Watching (請繼續觀賞)：推薦你可能會想繼續觀看的影片
• Video-Video Similarity (因為您觀賞過 …)：推薦用戶可能會想看的類似影片
• Page Generation：最後，將上述演算法排序出最適合你的的個人化首頁

上面的演算法看似很多種，不過大致上可以歸結為兩類：Content-based filtering 與 Collaborative filtering method。簡單來說，前者是根據影片本身特性，找出類似影片並推薦給用戶，後者則是先找出喜好類似的用戶，藉此判斷 A 可能會喜歡 B 看過的影片。關於詳細的演算法判斷流程，有興趣的人可以參考 CS50 的影片。

AI 演算法除了應用在推薦影片，不知道大家有沒有發現，其實 Netflix 還會依照個人興趣，來客製化電影圖像！就拿《黑色追緝令》這部片來說好了，如果用戶 A 曾看過較多鄔瑪舒曼的電影，則演算法會判斷 A 是鄔瑪舒曼的粉絲，因此會呈現在電影海報上；同理可證，如果 B 用戶是約翰屈伏塔鐵粉且看了很多他過去的作品，那麼演算法當然會用約翰屈伏塔來吸引 B 用戶。

Netflix 連電影圖像都為你客製化 (圖片來源)

擔任 Research and engineering director 的 Justin Basilico 在受 NVIDIA 訪問時提及，公司透過 A/B test 來不斷優化推薦演算法，而由於推薦系統的成效短期內難以衡量，因此 Netflix 鎖定的是長期指標，如：用戶每天觀看時數、某段期間觀看天數。

推薦系統在全球化時遇到的挑戰

由於 Netflix 的服務地區涵蓋 190 個國家，用戶多達 1.3 億，用戶觀看影片的喜好受到各種因素影響，因此在設計推薦系統時，也面臨了許多挑戰。

首先，由於某些內容供應商在與 Netflix 談授權時，會限制該影片只能在某地區，或某段期間上架，因此部分國家的用戶無法觀看這些影片，而演算法便會因此判定這些用戶對這些影片沒興趣。又或者是，A 影片的授權期間只有一個月，B 影片的授權期間長達一年，而造成 B 影片的觀看次數較 A 高出很多，此時演算法也會判斷 B 影片的熱門程度較 A 高，以上狀況都會造成演算法誤判用戶的喜好。

另外，Netflix 用戶橫跨這麼多國家，每個地區的文化習俗、語言又不盡相同，因此 Netflix 在設計推薦系統時，也須特別考量到用戶的所在地與熟悉的語言。譬如說，印度用戶可能會較偏好寶萊塢電影。語言方面，由於多數的用戶，可能會偏好觀看自己熟悉語言的電影，因此系統也應該要納入這些因素來設計演算法，但問題是系統並無法取得用戶熟悉哪些語言的資訊，因此只能靠用戶過去觀看行為來判斷。

為此，Netflix 也持續增加推薦演算法考量的因素，包括影片能上架的地區 / 時段、用戶所在地、用戶看過影片的語言，藉此不斷優化推薦演算法的精準度。

總結來說，由於用戶對於觀影這件事通常有獨特偏好，因此即便是後進者，只要能掌握觀眾喜好，都有機會趁勢而起，Netflix 便是靠著許多優質小眾內容，抓住長尾市場而崛起。不過，當各家串流影音業者，例如 Hulu、Apple、Disney 也都狹著雄厚資金與原創內容加入戰局，未來誰將在這場戰役中勝出，目前下定論都過早。

唯一能確定的是，如果影音平台能藉由推薦系統，精準地將內容呈現給最適合的觀眾，必定能大幅提升用戶觀影體驗，如此一來，用戶也就有更大的誘因繼續留在平台上，而不會輕易被新進業者的低價促銷或大製作內容吸引走，這不僅是訂閱制影音平台的護城河，也是在消費性領域投入 AI 創新最重要的原則之一。

【歡迎所有 AI / IoT、Blockchain / DeFi、面向東南亞市場的創業者，加入專為你們服務的 AppWorks Accelerator】

Photo: Pexels

Natalie Lin, Analyst (林楓 / 分析師)

Natalie is an Analyst covering AppWorks Accelerator and Greater Southeast Asia. Before joining the team, she worked in the search engine marketing and email marketing teams at Zappos, America’s leading shoes and fashion online retailer, where she primarily focused on KPI management, campaign optimization, and project management. Born in Canada and raised in the Middle East, Natalie returned to Taiwan for high school before moving to the US for college and work. She received her Bachelors of Marketing at Case Western Reserve University in Cleveland, Ohio. Outside of work she likes to read, travel, and play video games.

About a month into AppWorks Accelerator AW#16, one of the startup founders approached me with a problem he was facing.

“My EDM (Electronic Direct Mail) person is leaving my company. Should I continue sending emails out to my customers? I don’t have a lot of time to dedicate to this type of marketing…what are my options?”

To answer his first question, yes. Absolutely. His second question required a deeper dive into the fundamentals of email marketing. In this blog post, I will explain the virtues of implementing an email marketing strategy, how to analyze data from your emails to simplify your marketing strategy, and what actionable takeaways you can use to create value for your customers through email marketing.

Why Email?

Email marketing is known as a way to improve your retention rate and increase the LTV of your customers. I experienced this first-hand in my email marketing role previously at Zappos, a subsidiary of Amazon and a leading online retailer of shoes and clothing. This marketing channel was strategically focused on the retention of existing customers for the company. When I asked the founder who posed the aforementioned question if his competitors were sending emails to their customers, what do you think his answer was? By implementing a deliberate email marketing strategy, you’re creating value for your customers in the long-run by offering relevant content instead of being seen as a platform for one-time use.

One of the most important KPIs for any business to measure is ROI (return on investment). How much bang can you get for every buck you spend? Compared to some other marketing channels that require you to invest a lot of your time and money, email marketing has for years generated the highest ROI, aptly dubbed the “king of the digital mountain.” Not only is sending emails incredibly cheap but it is also estimated that for every dollar spent on email marketing, there is a US$ 44 return to be had.

As more and more of the world’s population make the move online, email’s influence will continue to grow. Almost 46% of the world’s population are active internet users and one of the first things any digital user needs is an email account. Progressively, websites are requiring users to sign up and login with email accounts for access to content and services. With emails being completely free to create, email accounts are expected to grow at a faster rate than internet users worldwide as many users have multiple email accounts for different needs, so the market is only expected to grow over time.

Be that as it may, not every startup needs an email marketing strategy. Perhaps you’re still in the very early stages of your startup. You’re swamped with building an MVP, your SEO needs work, or you don’t have any customers yet. Acquisition of new customers may be more important than retention efforts. It wouldn’t make sense to spend time sending emails to your (lack of) customers. One day though, you’ll start focusing on building a, hopefully, life-long relationship with your customers. It will be worth your time to start thinking about what to do when that moment comes.

A “How To” Guide to Email Marketing

You either have never tried sending emails out to your customers or you already have some data you can use to improve your email marketing strategy. If the former, there are a multitude of blogs and articles that can help you get started. As mentioned above, email platforms won’t put a strain on your budget as there are many that offer free or affordable email services to use, such as MailChimp, Amazon SES (Simple Email Service), and even one of our own alumni NewsLeopard (AW#4). For this particular case, the founder was using MailChimp, which had a pretty good UI but had some reporting features missing that I would have liked to use. I’d like to add that one specific advantage to using a startup email service such as NewsLeopard is that they are more flexible with pricing if that’s a big concern of yours and can customize services and features to your needs. So for beginners, we highly recommend NewsLeopard.

Before you begin any sort of marketing campaign, you need to determine how you’ll measure the success of your campaigns. What KPIs should you look out for? Which KPIs do you need to focus your efforts on? While these questions sound similar, they are actually quite different from one another. Take a look at some of the important ones related to email marketing campaigns.

For optimization, you mainly want to focus your efforts on improving your open rate and your CTR (click-through rate).

Open rate = the number of times a user opens your email in their inbox.

Your open rate will tell you two things. The first is how well you’re managing your email lists. Not all your customers are the same. Some might come to your site and but not purchase anything. Some may only purchase once and are unlikely to return. Your favorites will probably come back again and again to buy your products or use your services. Therefore, you’ll want to organize your email lists accordingly. If you have the time to dedicate to this, you may want to create different email campaigns depending on what type of customer you’re targeting. If you don’t, my suggestion is to focus on the ones that are engaging with your emails – they’re more likely to benefit from your future emails.

The second thing that open rates tell you is how engaging your subject line is. Obscure or unrelated subject lines will only confuse your customer – you have one chance to get them to click into and read your email. When the founder approached me with his campaigns, I took a look at his subject lines and the accompanying open rates and came up with the following suggestions:

If you have the time, you should test out what types of subject lines work best for you. After some time collecting data, pull a report of all your campaigns (include subject lines as a column) and sort by open rate. It will give you a high-level overview on which subject lines work and which don’t.

CTR (click-through rate) = the number of times a user clicks on a link in your email

Your CTR can tell you a number of different things: how compelling are your images and do you even need them? How coherent is the format/design/structure? How engaging is the message? How relevant is the content? Does the length of the email matter? Depending on your business, you’ll get various answers to these questions, and the only way to find the answers is to test out different strategies of building an email. After looking through the founder’s email campaigns, I realized there were some tests that can be performed to improve CTR:

• Images vs. text: a well-known media company, theSkimm, is a subscription-only newsletter that sends out text emails of news stories that are intended to be simple and easy to read. Maybe your customers are looking for something minimalistic and straightforward. If your business doesn’t benefit from the power of images, this might be a good test to run.
• Formats: let’s say images are helpful for your customers. You can test out the types of hero images to use (the main image at the top of a section in the email). You can test out whether tiles or banners work. What colors do your customers respond to? If you work on SEO, it’s similar to how you would think about structuring your website. A quick tip: clear text is vital. Extravagant backgrounds that drown out your text will only confuse your customers.
• Messaging and CTAs: what kind of content will engage your customers? Depending on your overall marketing strategy, are they looking for something short and sweet, or do they want you to write more personal thoughts like a blog? If you have multiple CTAs, you can also dig deeper and see what content your customers are clicking on. If they are scrolling all the way to the bottom to click on certain links, you could start putting those links at the top of your next email.
• Timing: when is the best time to send an email during the day and during the week? Consider time zone differences, especially when you’re in the US and have to find an optimal time for both east coast and west coast customers. Are your customers checking their email during work hours or on the weekends? Also for holiday promotions, creating a calendar schedule will be helpful; instead of sending a Mother’s Day promotional email the day of, send one a week before so your customers have time to shop.

If you have the time, you should definitely test out what works in the content of your email and what performs poorly. In the same report you pull for open rate, sort by CTR instead and dig into your emails that have high CTRs and compare them with the ones that have a low CTR. The quality of the traffic you’re driving to your site will depend on your CTR.

There are so many resources out there that will help you identify what you can test to improve open rates and CTR. There are also some other KPIs that you should keep in mind, although you don’t necessarily need to optimize for them because open rate and CTR should be able to give you enough information to explain these other metrics.

Conversions = the number of actions/orders a customer has taken/placed on your site.

This one is obviously important in many ways, as you want to make sure you have a good return for all the effort you’re putting into creating emails. Keep an eye out for what types of emails are driving conversions however email marketing is not just about getting your customers to buy. It is about building a relationship with them so if your email is still engaging and useful but doesn’t lead to sales, it can still be considered a win.

Unsubscribes = this is pretty self-explanatory.

Decreasing unsubscribes is not necessarily a good optimization strategy because customers unsubscribe for all sorts of reasons but it’s a good metric to pay attention to for signals on what’s working and more importantly, what’s not. You’ll notice around the holidays your unsubscribes will increase; everyone is sending emails then and your customers might be unsubscribing en masse. Don’t take it personally.

Again, go back to open rates and CTR. Those metrics will help you identify the strengths and weaknesses of your email marketing strategy. Using the metrics I mentioned above, I was able to give some actionable takeaways to the founder and some advice on how to simplify his email marketing strategy so that he can continue to engage with his customers through email.

What’s next?

The most important takeaway from this post is a reminder that using data to drive decision-making can result in high ROI and a better relationship with your customers. Remember to test and look at the performance results of your tests. In the end, you’ll find that email marketing is an effective, easy, and affordable method to develop customer loyalty and increase sales. If you ever have any specific questions about your email marketing strategy, please don’t hesitate to shoot me an email!

Welcome to follow AppWorks fan page, we will continue to share more information about e-marketing.

Photo: Pixbay

Jun Wakabayashi, Analyst (若林純 / 分析師)

Jun is an Analyst covering both AppWorks Accelerator and Greater Southeast Asia. Born and bred in America, Jun brings a wealth of international experience to AppWorks. He spent the last several years before joining AppWorks working for Focus Reports, where he conducted sector-based market research and interviewed high-level government leaders and industry executives across the globe. He’s now lived in 7 countries outside US and Taiwan, while traveling to upwards of 50 for leisure, collectively highlighting his unique propensity for cross-cultural immersion and international business. Jun received his Bachelors in Finance from New York University’s Stern School of Business.

In comparison to the US, China, or UK, Taiwan is not usually the first place that comes to mind when it comes to global AI hotspots. However, recent bold moves by several American tech giants would have you think otherwise.

Earlier this year, the “AI-first” Google announced the plan to recruit 300 new employees in Taiwan, mostly engineers, and at the same time pledged to train over 5,000 Taiwanese in the field of AI. This is on top of the fact that Taiwan already serves as the company’s largest R&D base outside of America, after acquiring 2,000 engineers from HTC last year.

Meanwhile, Microsoft also chose Taiwan over China and India as the location of a US$ 33 million AI R&D hub, which will look to employ 200 researchers in the next five years.

IBM announced the expansion of its Taiwan R&D center to incorporate 100 new hires that will focus on exploring AI, blockchain, and cloud computing technology.

So why the sudden deep interest in an island that has been in recent times characterized as an “Asian tiger that lost its roar”? Well, the parallel explosion of computing power and big data actually did more than just create the perfect storm to bring AI from classroom theory to real-world application. It also turned Taiwan into a fertile breeding ground for the next generation of AI talent.

AI talent is scarce and in demand

A recent study by Element AI suggested that there are only 22,000 qualified researchers that have the abilities to seriously spearhead new AI projects worldwide. With such limited supply, companies are usually left with only two options for building an AI team: poach or cultivate from the bottom up.

However, when starting salaries for even a fresh PhD grad with little to no work experience are reaching north of US$ 300,000 in traditional AI hubs, it actually becomes much clearer why companies are starting to look at places like Taiwan, where the labor is significantly cheaper but no less qualified for training.

Unbeknownst to many, Taiwan has actually made its fair share of contributions when its comes chartering the AI frontier. One of the lead developers behind AlphaGo was actually a Taiwanese computer scientist named Aja Huang, and another Taiwanese native, Tung-Kung Liu, is currently ranked 2nd among the world’s top 10 AI researchers and influencers.

According to the SCImago Journal Rank, the island also boasts the 12th largest number of CS-AI papers published worldwide and places 8th in the number of citations. But more importantly, while Taiwan might not boast the largest concentration of AI experts per se, it certainly exhibits more than enough raw talent that, if properly refined, could fuel a company’s regional AI initiatives well into the distant future.

But how about other countries in Greater Southeast Asia? From a sheer scale standpoint, Taiwan’s population of 23 million cannot compete with any one of its many regional neighbors that boast millions more laborers, readily accessible for pesos on the dollar. However, we’re not talking about sorting beans or sewing buttons here. Effectively employing an AI strategy requires highly skilled workers, equipped at the bare minimum with solid foundations in both mathematics and programming, and ideally with a familiarity—or at least the ability—to quickly pick up common machine learning algorithms and frameworks.

At the top of the talent pipeline, Taiwanese students boast the 4th highest performance in the world when it comes to mathematics and science, according to the OECD, with engineering degrees comprising over 25 percent of all university graduates every year.

Taiwan’s history as a leading OEM/ODM for the world’s leading consumer electronics brands has also spawned an unparalleled talent pool, trained in all facets of electrical engineering. This makes them ripe for capitalizing on the growing application of AI in IoT, Industry 4.0, and smart cities if properly upskilled.

A helping hand

But of course, wooing American money and marketing Taiwan’s potential as a global AI and technology hub has fallen right in line with the government’s long-term ambitions. This past January, the Executive Yuan unveiled the Taiwan AI Action Plan, which has earmarked US$ 1.22 billion over the course of four years to help the proliferation of Taiwan’s AI muscle.  

This includes everything from fostering and attracting top AI talent, facilitating supportive regulations, incubating local AI startups, and bridging academic research with industry applications.

The Taiwan AI Action Plan is actually only one of several initiatives launched under the purview of President Tsai to help Taiwan finally make the leap into the 21st century as a truly innovation-driven economy.

The 5+2 Industrial Innovation Plan focuses on promoting several key industries within the economy where Taiwan displays the most potential to play on the global stage such as IoT, smart machinery, and biomedical. DIGI+ aims to completely digitize Taiwan’s infrastructure and turn it into a Smart Nation.

That said, government-led initiatives that to aim to forcibly spur innovation sometimes fail to amount to anything more than political rhetoric. That is why many intrepid entrepreneurs have taken it upon themselves to showcase Taiwan’s strengths.

One of these individuals is Ethan Tu, the CEO and founder of Taiwan AI Labs—Taiwan’s answer to DeepMind. Most recently serving as Microsoft’s director of AI R&D in APAC, Tu started Taiwan AI Labs to help the country get a foothold in the global AI race by conducting grassroots research and cultivating homegrown AI talent. That mission seemed to have resonated well with a fellow AI heavyweight, Mike Calcagno, who recently joined the project and formerly lead the development of Microsoft’s Cortana.

Unchartered territory

Although AI has been around since 1965, public interest in the technology has only really started taking off in the better part of the last decade with the pivotal breakthroughs in deep learning and neural networks. This is when you started hearing about the IBM Watsons and AlphaGos of the world, when virtual assistants became a ubiquitous feature in every smartphone, when home speakers became smart enough to talk back to you, and when the first self-driving cars started to hit the streets.

Indeed, AI has evolved at an accelerated pace in the last few years, but the technology still remains relatively elusive in emerging regions such as Greater Southeast Asia, which is home to over 650 million people. The region uniquely exhibits a relatively young population that was born straight into the digital era, with more and more coming online every day. That means much more data and a much more receptive demographic when it comes to adopting new technologies.

But having all that data is useless unless you can make sense of it. That’s where Taiwan can step in. Although Taiwan itself is still an infant player in the global AI race, the country has displayed promising initiatives on both the public and private side. In a few years time, multiple waves of qualified and experienced AI talent will likely splinter off and diffuse best practices across the region, and beyond.

If you’re a startup currently or prospectively employing AI, be sure to apply to AW#17—AppWorks Accelerator’s inaugural AI & Blockchain only batch. Final deadline for applications is June 18.

Photo: Visual Hunt

Antony Lee, Communications Master (李欣岳 / 媒體公關總監)

負責媒體與社群溝通相關輔導。加入 AppWorks 前有 18 年媒體經驗，是台灣第一批主跑網路產業的記者，先後任職《數位時代》副總編輯、《Cheers 快樂工作人》資深主編、SmartM 網站總編輯。畢業於交大管科系，長期關注媒體產業變化，熱愛閱讀商業與科技趨勢、企業與人物故事，樂於與人交流分享，期許自己當個「Internet 傳教士」。

Blockchain 正在全球各領域帶來顛覆的力量，是創業者的大好機會。Blockchain 所帶來的衝擊，將如同 80 年代的 PC、90 年代的筆記型電腦、2000 年代的 Internet、2010 年代的 Mobile Internet，是改寫未來 30 年商業與生活的巨型典範轉移。

Blockchain 掀起的第一道巨浪，是加密貨幣 (Crypto) 的風潮。根據交易所 Coindesk 統計，2017 年 Q4 全球加密貨幣市值突破 6,000 億美元，ICO 總值為 32.3 億美元，最具代表性的 Bitcoin，對美元價格，從 2009 年不到 1 美元，一度暴漲突破 17,000 美元。

然而，加密貨幣只是 Blockchain 的其中一項應用。隨著 Ethereum 以及各種新協定、新技術出現，Blockchain 的創新應用正在蓬勃發展。Blockchain 技術是價值交換與儲存的分散式網路，有如 Internet 一樣，將在未來 30 年內顛覆金融、保險、醫療、房產、零售、交通、媒體等許多領域。

面對科技的巨變，為了讓應用 Blockchain 技術的新創團隊，能夠互相交流，共同探索 Blockchain 的新趨勢、新商機，AppWorks 特別在日前舉辦一場 Blockchain 新創團隊專屬的 Open House。

活動由 AppWorks 創辦合夥人林之晨 (圖左一) 擔任主持人，分別邀請發行的加密貨幣 Mith 已在全球 14 個交易所交易、全球排名前 60 的秘銀創辦人黃立成 (圖左二)；提供借記卡／跨境 Blockchain 資產清算服務、服務全球超過 50 國用戶的 WageCan 共同創辦人鄭敬錞 (圖右二)；中信金控區塊鏈實驗室負責人李約 (圖右一 )，一同分享有關 Blockchain 趨勢的觀察，以及創業的機會與挑戰，以下是活動精彩內容整理：

1. Blockchain 創業應挑戰原創主題

黃立成曾經歷過 90 年代的網路泡沫，也累積過不同的創業經驗，他認為不利用這些經驗投入 Blockchain 創業實在太可惜，因此創辦了秘銀。他認為，雖然現在 Blockchain 的發展，仍處於早期階段，風險高、失敗機率也高，但非常鼓勵年輕人要勇於嘗試冒險。

他觀察，目前大眾對於 Blockchain 的討論，仍停留在 ICO 議題，而不少公司都是透過抄襲別人的白皮書、稍做修改就希望做 ICO 來賺錢。這樣的心態並不健康，也非長久之計，他希望選擇 Blockchain 的新創團隊，要更有誠意去思考 idea ，因為這領域在未來幾年，還有非常多有趣的發展，也持續吸引更多優秀團隊投入，因此必須更注重原創性，即使以模仿他人出發，也必須加入新元素。

秘銀創辦人黃立成提醒，Blockchain 創業要更注重原創性。

林之晨也補充，Blockchain 新創團隊可選擇申請加入 AppWorks Accelerator ，尋找更多實驗 idea 的合作機會。AppWorks Accelerator 校友體系，目前共有 300 家新創企業，是大東南亞地區規模最大的加速器網絡，範圍涵蓋各種 B2B 到 B2C 領域，整體年營收突破 570 億新台幣，各校友間緊密的合作與扶持，是 Blockchain 新創團隊進入應用場域的最佳跳板。

2. 把 Blockchain 視為加速建立信任的機制

如何用一句話解釋 Blockchain？李約說：「每個交易透過各個參與方的共識，記錄在各自的電子帳本，並且不能被竄改。」他認為，在實體世界中，當彼此間有利害關係時，必須互相信任對方和對方提供的資訊，例如他是誰、擁有什麼、接下來要做什麼，在過往，信任的機制主要來自於紙本（簽名、身分證）、第三方（律師、會計師）、抵押與擔保品。但這些發展上百年的信任機制，都有可能被 Blockchain 取代，發展出新的信任機器。

李約以音樂產業為例，有越來越多歌手，開始跳過唱片公司，自行與數位串流平台合作發表作品，因此產生非常複雜的版權與版稅手續。因為版稅普遍小額且多元，當串流平台要支付給歌手，經常產生爭議，KKFarm 的 Soundscape 就用 Blockchain 來計算版稅，然而在支付的過程中，必須透過銀行轉帳與審核。因此中國信託就建立節點，做雙方的驗證，確認版稅的事實發生、以及區塊鏈上的運算結果。他更近一步指出，未來不只音樂，遊戲點數、餐券、群眾募資等等，都可以有相關應用。

中信金控區塊鏈實驗室負責人李約指出，Blockchain 是創造信任的機器。

3. 台灣環境仍開放，更適合 Blockchain 生態系發展

黃立成認為，台灣現階段發展 Blockchain 產業，在於「沒有規則」，因為相關法規尚未完善，「這既是困難點，也是機會點，無論你想怎麼發展都可以嘗試，但同時後果也必須自己承擔。」他也指出，在 ICO 領域中國已經明令禁止，不像其他數位領域，中國都發展出強大的龍頭企業，在 ICO 上，台灣不會面臨來自中國的競爭，大家必須把握機會。

鄭敬錞則認為，台灣發展 Blockchain 的優勢，在於擁有許多外語人才，很適合發展全球市場，而 Blockchain 正是全球市場，此外，足夠的人才庫、地理環境、教育環境，也是台灣一大優勢。

李約也提出觀察，他認為在台灣投入 Blockchain 創業或發展，一定要走出台灣。因為台灣集中化的系統太過方便，他舉例，像是日本晚上完全無法匯錢，但台灣卻很方便，很多先進國家甚至沒有自然人憑證。此外，台灣是一個人際間高度信賴的社會，而 Blockchain 本身，則是創造信任的系統，所以大部分的時候，會覺得沒有應用 Blockchain 的必要，所以必須與國際合作來發展 Blockchain ，因為只在台灣內，許多商業活動，並不需要 Blockchain 就能運作了。

對於台灣是否適合發展 Blockchain？林之晨也提出兩點觀察。第一，是台灣消費者很願意嘗試新事物，適合推實驗性的網站服務；第二，則是台灣大部分民眾有信用卡、銀行帳戶，發展加密貨幣相對容易，民眾接受度相對也較高，若像不少東南亞國家，仍以現金為主要支付工具，發展所需的基礎建設與民眾接受度，相對就比較不容易。

4. ICO 再進化，朝向私募化、法人化

同樣的信任問題，也發生在 Token 的募資上。鄭敬錞觀察，雖然 ICO 風潮讓募資更容易，但同時也產生了嚴重問題：許多公司直接抄襲他人的白皮書進行詐騙。這樣的問題，來自於缺少能夠評價團隊的第三方驗證機制，且整個行業缺乏監管。這樣的趨勢下，他預估未來向一般大眾公開募資的 ICO 比例將大幅下降，更多募資開始朝向機構投資人、專業投資人，甚至很多創投也會參與私募。

然而私募仍無法解決「缺乏信任」的問題，依舊會有募資完團隊就解散的風險。此外，募資過程中，需要發送訊息給幾百個投資人，傳遞時容易因人為疏失出現錯誤。因此 WageCan 團隊希望能用智能合約解決這樣的問題，希望在 Ethereum 上，建立連結全球投資人的信任網路。把所有的 offer 放在智能合約上，所有重要參數都能被 Blockchain 紀錄、驗證，並快速發送給投資信任網絡，在整個投資過程中建立信任節點。

WageCan 共同創辦人鄭敬錞預估，ICO 將朝向私募化、法人化。

5. 實名制、非實名制將並存發展

Blockchain 要能與更多商業、生活應用結合，實名制是相當值得關注的趨勢。鄭敬錞認為，實名制其實才是趨勢，且進展得非常快，因為各種 Blockchain 創新應用要真正落地、加密貨幣要換成法幣，最後一定要透過實名制。

李約則補充說：「實名制、非實名制的 Blockchain 應用發展將同時存在。」從銀行的角度來看，當然希望實名制，因為目前很多加密貨幣相關的項目，就因為尚未實名所以無法合法，但從加密貨幣的持有者角度來看，則會認為實名制後，所有的交易都被記錄，其實滿可怕的。所以兩套制度都各有使用需求，將並存發展。

結論

這次 AppWorks Open House for Blockchain Startups 在現場近 400 位創業者的熱情參與下，熱鬧圓滿的結束了。除了以上為大家整理的重點，相信到場的朋友應該都帶走一些屬於自己的啟發。錯過的創業者，期待下次 Open House 可以看到您，歡迎追蹤 AppWorks FB Page 以收到最新活動訊息。

現在就是創辦 Blockchain Startup 最好的時機，歡迎所有區塊鏈創業者加入專為你們服務的 AppWorks Accelerator

「我們當初取名叫 AsiaYo 的最大原因就是，我們從 Day 1 就想做跨區、跨語言的平台，因為我知道，光是靠台灣、靠中文語系，沒辦法長期讓公司活得好。」
—— CK Cheng, Founder & CEO of AsiaYo

數位科技的快速發展，讓共享經濟也跟著熱鬧興盛，源自美國的 Uber、Airbnb 和以東南亞為主要市場的 Grab、Go-Jek 等獨角獸，都是極其成功的例子。只不過，縱使有老大哥領頭帶路，因市場已被快速擴張的先行者佔據，能順利達陣的後發者並不多，而在這些鳳毛麟角中， AsiaYo 用他自己的方式，穩紮穩打地踏上世界級舞台。

許多人用亞洲版 Airbnb 來稱呼 AsiaYo，的確，以服務內容來說，兩者非常相似，但仔細觀察他們的經營策略，AsiaYo 做了很多 Airbnb 沒做的事，對內，他們採取 Centralized Management ，由公司統一提供細緻的真人客服，提升消費者的住宿體驗；對外，他們與日本樂天進行策略結盟，借力使力，創造雙贏，而這些都為 AsiaYo 提煉出不少成長動能。

AsiaYo 在 2014 年上線，2015 年底就決定要開發日本市場，緊接著又向韓國和泰國進軍。如今，他們擁有 25 萬位會員，線上累計有超過 6 萬個房源，海外市場營收也超過整體 6 成，平均每月媒合 15,000 晚的住宿 (業界稱為 Room Nights) ，這樣的成績自然讓許多旅遊同業稱羨不已。然而即使成績一路上揚，過程中，團隊仍是戰戰兢兢、如履薄冰，應該也沒有多少人知道，創辦人 CK 鄭兆剛 (AW#12) 甚至早從十多年前，就在為創業做準備。

二十年內，旅遊業都是亞洲的超級潛力股

「當工程師蠻無聊的。」交大電子工程系畢業的 CK ，用一句話說明了當初沒有走上工程師這條路的原因。

受到多位交大學長的影響，CK 對創業也有一番憧憬，儘管他大學畢業的那個年代，正是台灣電子業最蓬勃的時期，他依舊毅然決然選擇出國唸書。回國之後，不管是進入創投公司、還是後續在里昂證券擔任研究部門主管，也都是他為了創業精進自己的方法。

「里昂證券的分析師工作讓我得到一個結論；創業的重點不是產品，而是產業，你一定要選一個對的產業，因為對的產業在未來二十年內，都可以有一個健康的成長。」證券業的訓練，讓 CK 在分析產業時，非常重視潛能和未來發展。根據他的觀察，亞洲人口的收入剛剛開始起飛，出國旅遊的人勢必越來越多，往後至少還有二、三十年的成長空間。因此，在評估過自己的條件和能力後，他決定了自己的創業題目，那就是打造一個住宿訂房平台。

只是，金融領域的豐富歷練，雖然讓 CK 足以為團隊做最適當的財務規劃與風險控管，一旦要跨足到民宿訂房平台，他深知自己還需要學習更多產業相關知識，團隊的人才也需要補強。因此，平台成立後沒多久，他便加入 AppWorks Accelerator ，希望能透過這個社群，吸收更多經驗，同時也拓展自己的人脈。回想起這半年的加速期， CK 笑稱自己應該是 AppWorks 的最大受益者，因為那些在 AppWorks 認識的夥伴，有些人後來直接加入了 AsiaYo，有些雖然沒有一起合作，但也彼此分享了許多創業過程中所需要的資源，這在極度重視團隊和人才的 CK 眼中，自然也彌足珍貴。

從手動平台到人工智慧，從懷疑到篤定

雖說 AsiaYo 目前很風光的擁有 6 萬個房源，消費者從瀏覽、預訂、付款和後續的客服，走的也是全自動流程。然在兩年前，他們的平台還只是一個很陽春的系統，回想起草創初期，CK 說：「客人感覺已經在我們的網站上訂了房間，但事實是，有人訂房之後，我們才打電話給民宿，匯款也是一樣，一切都靠手動，我們就這麼一路爬爬爬，爬過來。」當時，CK 尚未離開里昂證券，公司就靠太太帶著三名員工，用最克難的方式，試圖找出 MVP。

這麼辛苦，難道沒有猶豫過？CK 笑著說：「我其實一路都在懷疑，但因為有很多事要想，也不能懷疑太久。」而這一路，竟然也走了十八個月才看到一點起色，也才讓 CK 覺得似乎可以繼續。

雖然系統的打造是一步一腳印的緩慢前進，但房源的開發，他們則是一開始就決定不親訪，只透過信件、電話、臉書廣告、房東轉介等方式來搜集，因為他們認為，這樣才有機會規模化到海外。從 2014 到 2017，整整三年的時間，團隊像空軍一樣到處尋覓，一直到後來，訂單成長到一定的程度，房源真的不足了，他們才更積極的進行拜訪，想把台灣所有的房源找齊。當然，這努力並沒有白費，因為到 2017 年底， AsiaYo 在台灣的房源，已經正式超過 Airbnb，這對一個僅僅成立四年的新創團隊來說，絕對是一個莫大鼓舞。

只是，在已成紅海的旅遊產業，如果只是將平台和房源這些基礎建設打理好，不用多久，就會被迎頭趕上，因此，AsiaYo 必須繼續奔跑。

他們透過自家開發的「AsiaYo Sort」演算模型，針對由 6 萬筆房源形成的巨量資料，以及多達 25 萬位會員的使用行為進行分析和追蹤，企圖深入了解訂房市場與使用者行為趨勢，並根據統計結果來調整營運決策，進一步更精準的預測消費者挑選住房的決策影響指標。而這一連串的 AI 應用，不僅在今年第一季讓訂單轉換率提高15% ，也加強了平台用戶的黏著度。

提升住房體驗，雕琢品牌價值

人工智慧技術加速了 AsiaYo 的業績成長，但他們並沒有因此忘記服務業的根本。從成立第一天開始，AsiaYo 就堅持提供細膩的客服，大至重複訂房，小至熱水不熱，都有專職客服人員在線上幫忙排解。 CK 認為，住宿這件事的細節很多，人在海外本來就不方便，再加上極可能產生的語言問題，過程中的確需要平台給予更多關照。事實上，不只是客人，他們也接受來自民宿主人的求助，竭力將訂房平台能提供的功能發揮到極致。而每天都做到晚上 11 點才收工的客服換來的，不只是溫度和品牌形象，還有客人的高度忠誠。

「這樣的做法雖然會讓 AsiaYo 的客服成本比其他同業來得高，但相對的，我們卻能藉此控制服務品質。而且我們相信，未來技術演進的速度會超過我們訂單的成長速度，短時間內我們用人力解決，但之後就可以用技術解決。」CK 想做的不單是藉由優質服務，提高客戶留存率，他更希望在贏得消費者的信任後，可以透過技術，維持相同的服務品質，而不是為了追求快速成長，讓不甚愉悅的住房體驗，趕走自己的客人。

而一連串的佳績與業界的好口碑，也吸引了日本電商巨頭樂天登門尋求合作，也就是說，除了 自己開發的 6 千筆日本房源，AsiaYo 也預計於 6 月中，陸續開放日本樂天提供的合法民宿，初步估計至少有破千筆的獨家新房源上架，供消費者更多住宿選擇。

對於能夠和一個在日本擁有近 9,200 萬會員，同時橫跨銀行、保險、旅遊、電商等多種領域的大企業合作，CK 謙虛的說：「其實我也不知道他們為什麼會看上 AsiaYo，他們第一次來拜訪的時候，我們還非常小，辦公室甚至還在施工。」不管如何，這次合作對 AsiaYo 來說顯然是如虎添翼，因為就長期來看，日本多年來一直是台灣人最愛的旅遊地點，而近在眼前的 2020 年東京奧運，也可望為雙方帶來另一波成長。

AsiaYo 在 2014年第一季末上線，2015 年登陸日本，2016 年正式進軍韓國與泰國市場，今年還要走向新馬港澳，前進步伐之快，讓 AsiaYo 的未來充滿想像空間。「我們當初取名叫 AsiaYo 的最大原因就是，我們從 Day 1 就想做跨區、跨語言的平台，因為我知道，光是靠台灣、靠中文語系，沒辦法長期讓公司活得好。」

目前 AsiaYo 的房源不再以民宿這單一類型為限，青年旅館、文創設計旅店，以及中小型旅館的數量，已逐漸累積至三成。而在今年 2 月，英語系平台也正式上線，預計這個英語系平台，將可在下半年為 AsiaYo 帶來更多外籍用戶。

過去，AsiaYo 致力開發房源，帶台灣、香港人到海外旅行；從現在起，他們的挑戰是開發客源，讓其他國家的旅客可以到台灣，或其他亞洲國家旅行。而想讓客源更國際化，金流、語言、文化，樣樣都是功課。如何持續善用機器學習、深度學習，自動化平台的種種功能，增加運作效率、提昇貼心度和服務品質，也是 CK 不斷在思考的。一如過去這一路上的摸索和嘗試，AsiaYo 這次ㄧ樣會正面出擊，因為從第一天開始，他們的目標就不只是台灣，而是整個亞洲。

AsiaYo 確實不只是亞洲版 Airbnb，他們正在走出一條自己的路。

AI / Blockchain 新創，歡迎申請加入專為你們服務的 AppWorks Accelerator