•   almost 5 years ago

Coca Cola previous failed attempts

Does anyone have details or documentation of what coca cola tried to do before? It's vague from the devpost description. Cameras, what kind of cameras? weighted sensors, how did it work? What were the costs? Did Coke even try or did they just brainstorm some ideas and gave up?

  • 14 comments

  • Manager   •   almost 5 years ago

    I'll try to gather a case study or at least as much detail as I can find. I do know that they got decently far down the path. With the cameras, once they got a prototype working, they started looking at scaling it and realized the costs were prohibitive. I believe the weighted sensors approach was also built, but was determined to not be "good enough." The weights couldn't tell what was actually on a shelf, just that something was in fact on it. Perhaps the ultimate solution is a combination of things/sensors, etc that collect different types of data which are then combined to make a guess with someone acceptable level of reliability/accuracy.

    In any case, we'll dig up what we can. Thanks for asking.

  •   •   almost 5 years ago

    Hi Jack & Shane,

    I can speak to the weighted mats part of it. We tried to use printed electronics to create mats that would detect weight as well as product profile (think how bottom profile of cans are different from bottles). Was able to determine if the product on it was a can or bottle, but challenge was determining minute weight differences between say a can of two products that use the same type of sweetener (Diet Coke & Diet Sprite are very similar in weight). The cost deterrent was that there was an industry expectation at the time that printed electronics would drastically become a much, much cheaper technology... which has yet to happen.

    Hope that helps! I myself am very interested in knowing what kind of cameras were being used as well and how they were being used. Was it something along the lines of compared different images taken throughout the day?

  •   •   almost 5 years ago

    It would help some of us to know accurate and precise descriptions of the failed attempts. Since this information is not easily available, the participants are probably in the same position as I am in myself. Employees are probably bound by covenants of silence and in-house research is proprietary and probably well protected. That leaves the majority of us to guess and make many assumptions. Unless you have inside information, you are going to be at a disadvantage. I hope any and all public information is shared here on this forum. Without that, the playing field is not level. Thanks, Amanda, for the tip regarding printed electronics. That is an excellent point well taken.

  • Manager   •   almost 5 years ago

    Wayne - I can promise to share any information we gather in response to one participant's request with all participants, here in the public forum. Never will we send something to one individual via email or otherwise that we will then also not post here.

    We are hunting down more specific info from the failed attempts, as well as CAD renderings of the coolers via another participant's request. If there's something else you think would help you and other participants, please point it out.

  •   •   almost 5 years ago

    I simply measured a 12-ounce drink can to gauge it's size, but I am wondering how many different sizes of containers might be in a cooler and the sizes of each. Is this info generally available?

  • Manager   •   almost 5 years ago

    Wayne - here's what the experts told me. There are many (hundreds) of different sizes, shapes and weights of Coke products in many different package types, and they are always changing and will undoubtedly continue to change in unpredictable ways. In order for a solution to be scalable, it mustn't be calibrated for any specific set of items.

    # of times I used "mustn't" above > 0

  •   •   almost 5 years ago

    @Shane

    I was wondering what the issues with cameras were - processing power, occlusion, etc.? What was the cost per module, that made it prohibitive?

  • Manager   •   almost 5 years ago

    I haven't been able to find information on the attempt using cameras yet, although I have been told in order for the solution to be economically viable the cost per cooler at scale must be sub $20.

  •   •   almost 5 years ago

    @Amanda Hui

    How precise were the weight sensors? Would it be possible for the bottling machines to add or subtract a few grams of soda for different types of drinks (330 ml Coke bottle, 333 ml Fanta, 327 ml Dr. Pepper etc.)

  •   •   almost 5 years ago

    @Mohammed Aamir I do not believe so since all packages come in "standard" sizes in the sense that different brands/flavors would all be sold at a 330mL standard. In my opinion, I cannot see it being realistic to have all differing weights to different types (too many brands and consumers would not be happy with getting less because they like a particular brand!) Although that would be a creative approach.

    As for how precise, they were precise enough to detect a solid 1 gram difference, but not any less which would have been required to detect the minute difference between a full-sugar Coca-Cola vs. Diet Coke as an example.

  •   •   almost 5 years ago

    What was the mobile app idea that failed?

  • Manager   •   almost 5 years ago

    @Tom - I'm not aware of a mobile app idea that failed. Did we say that somewhere?

  • Manager   •   almost 5 years ago

    @All - FYI - I've been consolidating all the questions everyone has asked and the answers I've been getting from the Coke experts here:

    https://hackpad.com/Cooler-Hack-n8XCW3i76nE.
    (scroll down to Q&A)

    Some really good stuff in there.

  •   •   almost 5 years ago

    @shane
    Home page of the challenge: http://coolerhack.devpost.com

    More background - i.e. the exact problem we are trying to solve?

    Out-of-Stock is the #1 Strategic Issue at Coca-Cola: This problem has resisted traditional technology solutions; from sensors on coolers to mobile devices in the hands of consumers to cameras. None of the previous approaches achieved the scale or creation of high quality data required to make a meaningful improvement in the system's ability to predict, identify, and address out of stocks.

Comments are closed.