Hacker News new | past | comments | ask | show | jobs | submit login
Nutella 'Hired' an Algorithm to Design New Jars. And It Was a Sell-Out Success (inc.com)
108 points by ALee on June 14, 2017 | hide | past | favorite | 62 comments

I was hoping the algorithm was to design a new shape of a jar rather than the label of the jar. There are tons of variables and constraints to consider for the jar shape. You want something easy for people to scoop out the product and to get all of it without wasting anything (or maybe an acceptable amount of waste so they go buy a new jar sooner but don't feel like they waste money). Also lid shape and size, easy to open and close but as minimal as possible. How efficient can you pack the jars in a box while keeping the box weight under whatever the standard person should carry without needing help. And also how easy it is to actually make the injection molded jar, too complex a design would cause too many jar rejects. And also just overall look, does the jar look like a jar? Human designers usually have a general idea of what a container should look like before adjusting the design to be cost efficient but if you give a computer program just a list of constraints and tell it to go wild, I'm sure you would get some innovative designs that most people wouldn't think of.

Think back to when condiments started coming in the upside down squeeze bottles. No company immediately went from the glass ketchup bottle to the upside down squeeze bottle overnight so it was a long process of interim designs until someone realized what the ideal shape and features the current ketchup bottle should have. Sometimes the most common design practice isn't necessarily the best or cheapest. Bottled water used to be made of very thick plastic with large caps that were sometimes not able to be recycled. The designs have changed to more thinner, ergonomic shapes with better usage of materials both as a cost cutting effort and a green effort. It would be interesting to see genetic algorithms used for things like packaging design in the future.

Me too. The nutella jar is such an awkward and messy shape to get nutella out off, especially as it gets emptier and emptier. I would prefer it in a regular wide-mouth jar if that were available.

Even better, give me nutella in caulk gun form. No wastage, accuracy in application to your toast and you just need to wipe the nozzle to keep it clean. No knife or spoon to clean either. There are tons of condiments I would prefer in caulk gun form. For jams and jellies, there would be no excess air that result in spoilage from when bacteria and fungi get in the jar. Thicker jams with fruit chunks can just use a larger nozzle.

Many jars are designed to increase waste... or so it seem. But in this case, it is probably a good thing... this stuff is about 1/3 fat and 2/3 sugar (https://www.nutella.com/en/us/range).

> until someone realized what the ideal shape and features the current ketchup bottle should have

And then someone had the great idea to create "dripless" bottle caps for both mustard and ketchup. Now my hotdogs always end up with a huge blob of mustard on one end.

Shoving a chop stick into the cap can "fix" this for you.


I have a close relative that works for the company that invented the valves in them. They used to make stuff like keyboard membranes, until China ran the price down too low to compete. An employee figured out how to make a cheap valve using most of the same tooling and the company patented it. Someone in sales got Pepsi to put them in All Sport bottles, and it saved the company.

Now the people in sales try to solve everyone's problem with them.

the nutella jar is rather iconic. it's like the contour bottle or the happy meal box. there are probably efficiency reasons to change it, but i don't know if they will.

> I was hoping the algorithm was to design a new shape of a jar

Me as well. It would have been a fascinating experiment.

agreed. The headline is misleading. If it said "algorithm designs patters on the background of the label" it probably wouldn't have gotten as many hits.

A little disappointing that there doesn't seem to be anything more in-depth than "An algorithm did it" in the article, or it seems in any of the articles I could find on this.

The "algorithm" is called HP Mosaic and is included free in HP SmartStream Designer for HP printers.

It was first used in a Coke marketing campaign in 2014 ...


More about how the algorithm works here: https://www.linkedin.com/pulse/hp-mosaic-20-steven-chow

> HP Mosaic takes the vector PDF file as input (also known as a Seed file), and generates a large number of variations on the file by transforming it — scaling, transposition, and rotation — randomly.

Okay, that's a little disappointing—I was hoping for some sort of image generation, rather than just cropping. But the output still looks good, so I'm not going complain too loudly.

I'm somewhat disappointed that they just used a feature in a consumer program to do this. I expected it to be some "collaboration", ala " ferrero x hp ".

I don't think it detracts from the creative achievement. The designer still has to know how to use the tool properly and make good choices for the patterns and colors.

But yeah, the headline is silly. It's the equivalent of "Vogue hired an algorithm to touch up new photos of models."

I agree. There's a wide range of possibilities here.

In the most basic implementation, you'd have 6 features, each with 14 possibilities, and you choose one option from each and combine them. That gets you 7.5 million combinations.

A more sophisticated implementation might pull from a larger number of features and options and apply complex transforms to produce nonformulaic output.

My guess is that what's actually cutting-edge here is not the algorithm or the designs, but the ability to print custom labels at all. My guess, though, is that it's only cost-effective as a publicity stunt. Printing 7 million identical labels is likely far less expensive.

If your talking fractions of a penny per label even if the costs was 400% higher it might not affect overall profitability of the product.

Identical labels will usually use flexographic printing with cut steel templates which transfer the ink to the plastic. Being able to customize each bottle is pretty cool in its own right.

> "Yes, algorithm. The word you hardly knew until HBO's Silicon Valley focused an entire show about the immense power and responsibility that comes with creating one."

Hard to imagine, but possible.

Surely in 2017 the average person would know what an algorithm is, outside of any tv show. Right?

I feel pretty confident in saying that the average person has no clue what an algorithm is. It's just a fancy word for something they don't understand.

They might have some idea about it being a process you follow to get a result, but they would just say, "I don't know" rather than trying to verbalize that idea.

I think the average person knows algorithms by their close relative - the "recipe"

Most recipes are not algorithms.

From OED:

    ... set of rules to be followed in calculations or other problem-solving operations ...

For me one of the key features of an algorithm is that it contains decisions. Most recipes are simply step by step instructions without any branching or looping.

"Add <spice> to taste."

"Bake for 40-45 minutes."

"If dough is too sticky, add flour."

Many recipes contain decisions.

It was tongue-in-cheek haha. That's why I said "relative", because they are not algorithms.

> Surely in 2017 the average person would know what an algorithm is, outside of any tv show. Right?


It doesn't take much effort to reach the top, say, 10% of the population for knowledge or ability in most areas. For people in the top percent or so that initial, easy set of knowledge or skills are so trivial and were learned so long ago that it's easy for them to assume everyone knows or can do those things, but in fact it's just a small fraction of the population—most people having devoted nearly zero time to any given thing, or having done so only very long ago and with little interest and since having spent no effort retaining whatever capability they did once have.

Basically if it's not part of the set of things most people have to know/do to get by, day to day (so, most math past ~6th grade is excluded) and you know/can-do it at all, you're better at it than most people.

As a high school student, I heard the word "algorithm" for the first time I could recall in the movie "A Beautiful Mind". I had been playing with computers and writing small BASIC programs for years, but had never been formally introduced to the concept.

I wasn't able to infer the meaning of the word from the film, and it wasn't until a couple years through my CS degree that I remembered the use of the word in the movie and the meaning of the scene in which it was used clicked for me.

No way. You know only 25% of Americans have college degrees right? And few know how things like search engines work. Maybe 25% of the population has a vague idea of that word meaning, 10% could accurately define it.

> You know only 25% of Americans have college degrees right?

And most college degrees [0] don't require anything beyond basic math (e.g., intro to statistics or calc) where "algorithm" might not even be said.

Might be overcompensating because I go the entire day only talking to people with PhDs (where 100% could define it?), but I really think << 10% could accurately define it.

[0] https://blog.suny.edu/2014/05/most-popular-college-majors-ov...

>intro to statistics or calc

Most degrees don't even get that far. You can satisfy the math requirement for most degrees with college algebra. In many cases a generic math class that's considered a step below college algebra works.

FYI, according to the Census Bureau, as of 2014, for Americans aged 25 or older: 31.96% have a bachelor degree, 41.89% have at least an associate degree, 58.57% have attended some college.

That said, I'd bet your 25%/10% guesses are fairly accurate. It also depends on what answers are considered accurate. An algorithm is really just a set of steps or rules for solving a problem. It doesn't have to involve math or computers, but I'd bet "something to do with computers" would be a common response.

Nah. <anecdata>My wife is very technically minded (and has been married to me, a software developer, for over 10 years) and I just taught her (with more effort than I expected) what an algorithm is.</anecdata>

IDK. I knew an "aspiring" programmer a few years ago who knew how to do things like write to text file and make a "for" loop. He decided he needed to study "algorithms" because he read somewhere. I walked by and he was looking at web pages on "logarithms". So not so sure everyone knows.

I think for most people "algorithm" is a fancy word having to do with advanced math formula or something like that.

I'm sure most people would have a very basic understanding (e.g. they know what the word means but can't define/explain it to another person) but anything deeper than that I would doubt. I know it sounds ridiculous but as a self-taught developer I had released popular software and was still in the bracket above before actually taking more of an interest in CS.

thats like saying the average person knew what an app was before the iPhone.

So, they randomized the creative and made "unique versions". Whee. A real win would be "Nutella 'hires' algo to optimize design aspects to both increase sales and improve label perception both emotionally and rationally". Aka, an optimization. For example, their current label is very busy with poor use of whitespace; what if they experimented with a variety of placements, size, and overall design of various elements, with algo driven optimization and letting the "best" compete for sales and user feedback...

As long as they don't change the flavor, of course. Love that Nutella.

The label appears to have increased sales. That's a real win.

Some people complain the novelty effect biases all A/B testing results against the status quo. This one weird trick flips bias on its head!

The ad cMpaign around the new labels seems to have increased sales.

New labels.

Here I was hoping that they had finally made a jar you could actually get all the product out of.

That's how you get horrible one-off utensils like https://www.kickstarter.com/projects/1508505428/spoonani

I think I see the problem: they didn't make a left handed model as well. Tut-tut.

This is so awesome. I hope he posted the design to thingiverse so people can print their own spoonanis and enjoy waste-minimized yogurt consumption.

Or one with better ingredients.

Misleading title. It was about the label, not the jar.

This is very similar to a campaign Absolut Vodka did where they used spray guns to create unique bottles. http://www.huffingtonpost.com/2012/09/13/absolut-unique-vodk...

This is genius marketing. They flipped "random labels" into "robotic algorithm of success". There isn't much to the story because that is all they did. But if the customers love the bottles, sales are up, and the brand refreshed their image, then this was a 10/10 execution by the folks responsible for it.

They put an awful lot of emphasis on "algorithm" like it's like pulling the sword from the stone.

I saw this in the social network too, where Zuckerberg demands "that algorithm" from Saverin and he proceeds to chalk it up on the window like that's how we communicate algorithms.

Makes me cringe a little bit!

It wasn't even an algorithm either, it was the basic Elo formula.

I do love that movie but it falls to a lot of the Hollywood Hacking stereotypes.

They should focus more on the UX. I never been able to remove seal correctly.

Given how I keep seeing ever larger Nutella containers I half expect them to begin selling it in 30 litre barrels.

This title made it seem like the algorithm was a proprietary piece of software that Nutella licensed for a limited time. It might have been interesting to discuss what it means to "hire" and algorithm vs a person.

But I think it just means they (the ad agency) wrote an algorithm.

It's even worse than that. "The "algorithm" is called HP Mosaic and is included free in HP SmartStream Designer for HP printers."

Lots of detractor comments. This kind of thing gets me very excited for the future of unique products.

I bought a pair of Camo shorts the other day and wondered if every pair was identical, or if they were "random". Unique products can present some fun new challenges for online retail where the customer may want to buy a specific item. Not just a jar of Nutella, but "that" jar of nutella. Maybe they'll even bid on them, because "pink stripes are my favorite" or whatever may be the value proposition.

To me this looks like a world of fun, not a chance to complain about it being the wrong tweak.

I think the biggest oversight is the actual feedback portion of sales to train the model on which features sold the best.

I was coming here looking for something like this -- a little more elaborate than this, but like this nonetheless.

The algorithm could probably be nearly anything as long as it output a mix of complementary, or aesthetically pleasing colours and simple patterns.

Psychology might do a better job designing labels than this algorithm.

I understand the work of the algorithm is to produce the actual designs, so it accounts for the colours and patterns, but from a higher order all we are really seeing is:

>> People like colour, make lots of colour. Simple is memorable, make simple. Punchy is eye-catching, make punchy.

I would be more taken if there was indeed a feedback loop, or further testing on people to see exactly which patterns/colours had the highest number of takers, and a poll to corroborate whether or not they chose consciously or unconsciously.

They could measure which designs sold faster: which pastel designs, vibrant colors, dots, shapes, etc., and get a quick overview of the preferences.

I think that's what would inform an interesting algorithm... but maybe I'm just conflating ideas. I think I'd be more smitten with a machine learning model that at first presented designs, then took as feedback sales data that informed its design progression.

In the end, sure the patterns are catchy. I guess that's all people really care for in a Nutella label. :P

Yeah, for the second batch of jars take sales data and do A/B testing with a new round of designs!

I agree there could definitely be a graduated approach to the granularity of features that might matter, but I think it would be possible to discover emergent multi-modal signals that matter.

I only hope the algorithm knows how to not output "wrong" shapes, such as a swastika.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact