Candy Japan 2019 Year in Review

Hi, I run a site called Candy Japan, which ships surprise boxes of Japanese candy to subscribers around the world up to twice a month. The site launched in 2011, and as the new year approaches, it's been my tradition to do a yearly review of how things are going.

So here's how 2019 went. Spoiler: it was mostly full of fail.

Key stats

Avg. monthly sales were $6,560 - expenses $5,270 = profit $1,290 (vs. $1,470 last year).

Number of subscribers fell by 34%, going from 314 in January down to 207 in December.

Candy Japan 2019 chart

The site had 130k visits from start of the year to mid-December, down from 145k in the same period last year.

Of these visitors 43k came from organic search (vs. 35k in 2018), 32k from links on other websites, 21k direct, 10k social, and 7k clicked on paid ads.

Cutting costs

The monthly profit of $1,290 was down from last year, but not as much as you might expect from the subscriber drop, because I was able to cut several costs here and there.

For example, about $100 / month was saved by switching from Recurly to Stripe subscriptions. I didn't start with Stripe, because it wasn't available yet when I started the site, and they only added subscription support later on.

Top expenses

Buying candy is unsurprisingly my largest cost at $2,450 / month. This year I spent more time finding the best prices for the candies I ordered, so that should have resulted in some savings as well.

The second largest cost is shipping at $1,800 / month, which doesn't seem possible to improve upon right now (there are volume discounts, but I'm far from getting those).

Do or delegate

I let go of my two freelancers who had been tasked with customer support and newsletter creation, saving $120 / month. With less customers it is easy enough for me to do the support myself. Writing the newsletter and taking photos for it is a bit more work, but I feel good about communicating with people directly, and enjoy the photography.

There is always overhead when having help. Instead of just doing something, you need to explain the task, verify the work and pay for it. For the newsletter I also needed to ship products to the author every time, leading to delays and confusion sometimes.

I think in the future I wouldn't delegate small tasks I am able to do myself, if they only take a few hours each week to do.

YouTube advertising experiments (fail)

I spent $310 / month on ads on average. It's a different type of expense, as I would gladly spend infinite dollars on it if it just had a positive return. Spend is very low now, simply because I have not yet found an advertising channel I could scale.

I'm a bit obsessed at trying to crack paid YouTube ads, because if I could make even a little bit of profit, it would be very scalable. I tried this before, but the cost was always more than I could hope to make back in a reasonable timeframe. I gave YouTube ads two more shots this year.

Attempt #1

Fiddled with targeting settings, streamlined the ordering process, then re-ran the campaign for a bit. Campaign still didn't seem to be profitable.

Attempt #2

Figured that with only one subscription option I might not be making as much as I could from each visitor. Potential customers looking to spend a different amount than the $29 / month option I had wouldn't be able to order at all, leaving money on the table.

Adding multiple tiers meant writing code, as I my homegrown Python subscription system didn't support it. Still, it wasn't as painful to make the change as I had feared, and in just a week I had two new tiers available: get just the first box of each month at $12.95, or the second one at $17.95 / month.

After this I was ready to try the campaign again, and decided to do more careful analysis this time. Sadly I was still a few dollars short from merely breaking even.

Next attempt?

Since I'm close to breakeven now, these ad experiments are no longer expensive. Even if I spend thousands showing my ad unprofitably, I still get most of that back. So I want to keep experimenting, and already know the next thing to try.

YouTube charges for an ad when it gets clicked OR watched for over 30 seconds. My ad happens to be 30 seconds long, mostly because creating the animation was expensive. But that might not be the optimal length.

There could be a large percentage of people who are not really that interested in Candy Japan, but still watch the entire ad, as it's short enough that you might just wait and fiddle with your phone instead of clicking the "skip" button.

I'm curious if making the ad artificially longer would get these uninterested people skip it before I get charged. If someone sees an uninteresting ad begin and see that it is, say 3 minutes long, then they'd be more likely to skip it quickly, as just waiting through it would take too long.

I could extend the ad by adding an unboxing video at the end, and would be curious to see if the added length would make more people skip it during the first 30 seconds. If that made it just 20% cheaper to run, then it would make the campaign profitable. Besides the skip effect, talking more about the service might also convert more of the people who watch the entire thing.

Messed up a major A/B test (fail)

Embarrassingly during 2019 I messed up a major A/B test where I tried to compare two completely different landing page designs against each other. The test ran for 3 months to get enough data, after which I spent a fair bit of effort analyzing and writing up the results. I thought the test was an improvement at 91% confidence.

Except that someone pointed out later that the result was completely invalid due to a stupid little bug: I set the Set-Cookie header in two different places, overwriting the variable that was supposed to control my A/B test! facepalm

The big thing I learned from this was that running what was effectively an A/A test still managed to convince me that I had found an improvement. Without knowing about the bug, following the winner of the test made sense, but not my unwarranted inner confidence of its validity.

I haven't re-run the test yet; it will be interesting to see which design was actually better.

Social media tests (fail)


I didn't have Instagram for Candy Japan before, because honestly I don't really understand it. Maybe because I spend more time reading content, and don't really see the appeal of scrolling through a feed of pictures.

Still, "start posting on Insta" is possibly the top suggestion whenever I mention trying to find ways to promote Candy Japan.

Fine. I started posting.

Six months and 51 posts later… 40 followers. No sales.

Looks like I can't cargo cult my way into Instagram success that easily.


On Flipboard you can curate your own magazine on a topic of your choice and "flip" articles to it. It's a bit like starting a subreddit, except you control it completely. Figured I'd try to start a Flipboard for Japanese candy, to which I flipped articles relating to Japanese candy once a day for a little over a month.

I figured if it gained readers, then I could occasionally post some of my own content there as well.

It didn't. Not a single reader in fact.

Content marketing test (promising)

I was curious if by adding a "news" section to the site and using the proper schema tags I could get Google to consider it to be a news site and surface that in its results. I figured I could summarize candy-related articles that originally appeared in Japanese news sites.

Surprisingly this trial worked, and Google Discover sent me 15,000 clicks, but the article they picked up was about North Korean sweets, so it didn't lead to any new subscribers.

However now I rank #2 in Google for "north korean candy"! Not sure if success.

Side-project marketing test (success?)

Now to wrap things up, here's something that worked.

I was inspired after reading this post about launching side projects as a form of marketing: create small fun / useful things and launch them under your main domain. Hopefully each launch will then get backlinks from sites that wouldn't otherwise have linked to you, and hopefully improve the ranking of the main site.

Seems like a great fit, as I just love short programming projects with a quick payoff. For example I often participate in Ludum Dare just for fun. So I started thinking about what to create, and realized I already had something perfect that I had created just weeks before as just an experiment: a candy animation generator.

I already had it on its own site, but after reading the post I moved the widget under my main domain and submitted that to Product Hunt. Result: 6000 visits, a handful of backlinks and even two sales. Yatta!


It seems hard to grow Candy Japan now, there are many things going against it. There's a general fading of interest towards japanese candy and harsh competition.

The other day my wife wanted to show the site to a friend, but actually couldn't find it because there are so many similar ones.

Personally in 2019 I was in study mode, doing online courses on Swift and deep learning, filling holes in my math background, and continuing my long journey of learning the kanji.


For 2020 I will go back into "exploration mode" again and spend more time launching small projects. During my life I've released around 50 random projects, of which 5 were some level of "successful". Based on that, if I try 10 small new projects I would have a fair (~65%) chance of success with at least one of them.

I will keep running Candy Japan as before and improving it when I come up with things to try, but in my mind this has downgraded from "the project", to just "a project".

Thanks for reading, and do subscribe if you'd like to try some candy for yourself.