Crack Kickstarter in 2021

As I warily approached the cart in the market, the young lady so nicely informed me the cart was disinfected — “I just now sprayed it,” she said cheerily, gesturing to the bottle of foul pink stuff she was wielding like a Tommy gun. I could tell she had just sprayed it; the residue was simmering on the handle of the cart… on the handle of every cart … making me wish for the first time I was wearing gloves to the market. “Thanks,” I said, through the mask…

The next day during a break in a Zoom meeting, I jokingly relayed my frustration about being all-but-drenched in the mysterious pink spray every time I leave the house! Little did I know.

A few minutes later, one of my colleagues sent me an email with the subject “Check these out” — he attached some drawings of several UV-based disinfecting products that he had designed. He had some exciting ideas for projects, one designed for a phone, another for salt shakers in restaurants, another for shopping cart handles. I was super impressed. I asked him why he had never pursued building any of these things, and it ultimately came down to the fact that he didn’t have the funding to take the project to the next level … “. It costs a lot to get a prototype built,” he said.

“Have you considered Kickstarter?” I replied innocently.

“Yeh but would that work for me?” he said.

“I … don’t know” … and that, my friends, is how I fell in the rabbit hole.

Over the next few weeks, we discussed this project at length. The notion of crowdfunding was not new to him; he just wasn’t sure if his projects were viable for Kickstarter or not. First, we needed to know if Kickstarter was still going strong in the pandemic, and if so, which of his projects would be most likely to succeed. Beyond that, is there anything he should do or say with his project to help it succeed, or things to avoid that might doom the project? So I set off to figure out how to predict if a Kickstarter project would be successful.

There are already heaps of studies done on Kickstarter success. It is easy to find numbers about various categories over time with varying degrees of success. Unfortunately, all the previous studies were pre-pandemic, so one couldn’t use that data to predict the success or failure explicitly related to disinfecting household goods, using data from pre-pandemic studies. I also didn’t find anything that would allow me to drill down into the projects and find out if particular keywords attracted (or repelled) success.

Getting the data was easy — webrobots scrapes Kickstarter projects every month, producing a fresh set of data available for download. Unfortunately, due to how the Kickstarter website works, duplicate records are inevitable and, well, every other kind of data error you could imagine, from nulls to high ASCII characters in plain text fields to mismatched countries. So I cleaned all this up with Python (pandas), converted the currencies and dates, detected the language of the project description (langdetect), parsed out the category and subcategories, and grouped the pledged and goal amounts to allow for the creation of mean and median values of projects in specific goal ranges. I also set up filters to create an accurate context to predict the success of a project. The script automatically filters by language, country, and project age based on parameters that can be updated by the analyst when the script is run the next time. I decided to cut off projects older than one year by default for my “filtered” dataset, but it could be modified later.

I set up test visualizations using ggplot2 in R to get a little better look. The first set of bar charts give some notion of what’s happening, but was the graph showing me something that holds up to statistical significance?

At first glance, it’s super promising! For example, 73% of projects in the filtered dataset are successful, with a mean pledged amount of $19,367!

Does that mean my buddy has a 73% chance of being able to pull $19,367 from his Kickstarter project? It just can’t be that easy!? But, of course, it isn’t. The actual numbers were not going to come out without a bit of digging.

The first tip is that sometimes the average (mean) doesn’t tell the whole story; look at both the mean and the median when evaluating a dataset. Some projects achieve extraordinary levels of success, sometimes getting pledges many hundreds of times the goal, but most projects do not achieve this kind of success. A byproduct of having a few super high performers is that calculations of the mean may offer a deceptive appearance of what one should expect from their project. When the data has a skewness of this sort, it is essential to consider both the median and the mean when looking at a particular category. For example, the mean pledge for projects in the technology group is over $27,000, considerably higher than the mean pledge overall of just over $19,000. By contrast, the median pledge for that group is just $1,520, lower than the median pledge amount of $3,095 for all categories. Thus, we have a small number of projects in this category that succeed well beyond their goal. Still, most projects in this category don’t achieve as well as most other categories.

Tip #1 — Choose the right goal amount.

The essential factor in determining the success of a project was the goal. Kickstarter has an “All Or Nothing” model; you only get the pledged funding if your project meets or exceeds its goal. Because many projects are created with a low goal amount to ensure success, this is why such a high percentage of projects are successful; there are heaps of them with goals less than $100.00. The downside is that a lower goal usually means a lower overall pledged amount for a project. For example, there are hundreds of projects with “$1.00” as their goal. These projects are 100% successful but tend to have very low overall pledged amounts and tend to be a relatively low effort in general. On the other end of the spectrum, there are many projects where the goal amount is a million dollars or more. Such projects all fail. Projects with goals over 1,000,000 fail with such consistency that I removed them from the filtered dataset.

Choose the lowest possible goal that still meets the funding needs of the project. Don’t put in an overly low goal and hope the project will catch on unless you’re good picking up just a few hundred dollars. Also, don’t put in a high goal and try to shoot for the moon; you’re just doomed if you try that.

Ok, let’s be clear — there are viral projects that come up on the platform, it’s possible to raise millions of dollars, but that is not the norm. I found no formula that would make a project go viral; you shouldn’t go into a Kickstarter counting on it going viral. Set a realistic goal that is as small as possible while still getting the project going.

As the goal amount exceeds $15,000, the project’s success tapers rapidly, though this does vary by category. The best approach is to analyze projects in the category you are interested in to determine what particular goal amount makes sense.

Tip #2 Make the project duration no more than 30 days

You might think that giving a project more time to succeed would give it more likelihood to succeed, but the data is super clear on this one. I was able to break down the project and compare multiple projects with precisely the same goal amount and found that a more extended project duration correlated to lower success no matter the goal amount. For example, 73% of projects with a goal of $5,000 were successful. But if you look at projects with a goal of $5,000 with a duration longer than 31 days, only 58% of them are successful. The trend persists up to the goal of $100,000.

Tip #3 — You may want to translate the project description to English

There are far more projects in English and pledges, so in terms of reaching the biggest potential audience, that is the best bet. Of course, if you are considering a local project, that doesn’t apply. Still, for projects that might be available to people everywhere, the project will typically have a higher chance of success in English. There are a few languages that chart slightly above the mean (CA (Catalan) and SV (Swedish)), and those particular results are statistically significant, albeit not by much (p< .05 but with low Z scores, it matters but not all that much). Every other language is either not statistically different from the mean or lower. For example, while Japanese appears to be slightly better than the mean because of the relatively low number of projects in Japanese, the ANOVA results do not bear that out to be significant.

There are some languages to avoid. Spanish, French and German don’t fare particularly well. There are a few that appear lower on the graph but but to the low n (count) of projects in those languages in terms of statistically significant results projects in Italian fare the worst of all. Unless you’re working on a project specific to a place that applies to that language, it would make sense to have the project translated to English.

Tip #4 — Give something away with even small pledges, preferably an enamel pin

Many people use Kickstarter to pre-sell a project they are trying to produce at a discount for pre-ordering. This turns out to be a reasonably successful strategy. In the case of expensive products where it’s not feasible to give a product for a low pledge, the projects tend to be more successful if you give something material with the pledge. And the surprise here is it appears that the thing to give away is an enamel pin. Here’s a comparison cloud I made of the descriptions of successful projects.

Comparison Cloud is a fantastic function of the WordCloud library in R. It automatically sizes entries appropriately based on how frequently they appear in the selected corpus relative to the corresponding one. For example, in this case, “enamel pin” does appear in both successful and failed projects, but every time enamel pin appears on the “failed” side, the relative size of that phrase shrinks on the “success” side. Here we see “enamel pin” appears so much more frequently in successful projects that it dominates the “success” side of the cloud. Similarly, “mobile” and “app” appear on the failed side, but there are successful “mobile apps.” If there were no successful “mobile apps,” those words would feature more prominently on the “failed” side of the comparison cloud.

Enamel pin stood out to me, so I made a subset of records that contain that phrase. The “enamel pin” data set included 625 projects with a 97% success rate. All the projects have relatively modest goals; in some cases, the enamel pin is the project’s product; in some cases, an enamel pin is given away as a small donation to a larger project. In either case, the data supports an enamel pin mafia that is trolling Kickstarter looking for enamel pin projects to help.

Tip # 5 Consider your category and match the subcategory

Sometimes you might think the category is a given. My engineer makes technology projects, its hard to put them anywhere other than technology … but just 33% of the projects in the Technology category are successful! Compare that to the Design category at over 90% success but a much lower mean pledged amount. It turns out there is no evidence that Kickstarter staff come along later and put your project into a specific category — its set by the project creator. It’s true my friend is making a technology project, but he’s also doing a design project, designing a technology product… it would be up to him to decide which category he wants to put his project in for optimal success.

The other notable element is the fact that some projects are in the same subcategory as category… in the “design” category, there is also a “design” subcategory. For whatever reason projects where the subcategory matches the category have a much higher chance of success. In the case of the technology category, recall 33% success, the “technology” subcategory has 100% success. Here is the technology group broken out by subcategory.

In the “Design” category, the “Design” subcategory is, again, 100% successful. Here is the design group broken out by subcategory.

The number for the matching subcategory are pretty dramatic; I immediately suspect shenanigans! Perhaps someone is changing the categories for some projects once then have become successful, but I could find no evidence of that type of manipulation.

Overall in the filtered dataset there were 1330 projects where the subcategory matched the category, 1308 of them succeeded. This represents a 98% success rate; by contrast, there were 10,362 projects where the subcategory did not match the category, 7239 of them succeeded, about 70%.

Bonus Tip — Make an interesting write up — get Staff Pick.

An engaging write-up will attract more pledges on its own, but this is compounded if you can get one of the Kickstarter staff to mark your project as a “Staff Pick.” Interviews suggest the staff look at new projects every morning. Interesting projects are marked as “Staff Picks” — this causes the website to feature these projects more prominently, which increases the chance of success and increases both the mean and median pledged amounts. Interestingly the staff picks projects with a higher mean goal and a much higher median goal which should tend toward lower success, but they tend to be much more successful despite that.

Of course the follow up question is — how do you “catch the eye” of Kickstarter staff, future study will include an exhaustive look at “staff pick” to see if there is some common ground there. I’d also have a good solid look at what makes up a “viral” project. I couldn’t find a formula or anything in common for the viral projects, but there might be something there if you look hard enough.

Tip — The Last — Run the Kickstarter Autominer

The software I wrote for this project is called Kickstarter Autominer, which you can download from my Github. It produces a lot more granular analysis than what is discussed in this article. You can use the results of the Autominer to drill down even further into your prospective categories and subcategories. There are too many other variables to discuss them all here but the Python script and R code will break down each and every category and subcategory to give a prospective project coordinator a really good view of the context for their project.

Good luck with your project!