Social Media Strategy

6 Potentially Wildly Inaccurate Observations about Tostitos and Social Sentiment

It’s entirely possible you love this commercial. I don’t.

The Tostitos “talking bag” commercial ran extensively during the bowl game extravaganza on ESPN and other channels on January 2. (There are actually at least two spots, but this one ran more often).

Given that I almost never watch television without an iPad nearby, I quickly retweeted displeasure about the commercial.


Thinking I would find other funny tweets criticizing the spot, I did a Twitter search for “Tostitos”. But within seconds, I determined that I was in the minority – the commercial was a hit among many TeleTwitterers.

Sentiment Graph for a Talking Bag

Sentiment Graph for a Talking Bag

Using Sysomos social monitoring software, we found 500+ tweets mentioning the Tostitos commercial on January 2 and 3. Among them, 142 were clearly negative, 190 were neutral, and 180 were clearly positive.

Lessons Learned From a Talking Bag

1. There’s no such thing as universal agreement
Especially in a highly subjective arena like advertising, tastes vary widely. Hell, 12.7% of the American people APPROVE of the job Congress is doing right now. As a marketer, don’t sound the alarm just because you see a few negative tweets or forum comments doesn’t mean all hope is lost. Sentiment may even out in short order. As we wrote about in The NOW Revolution, you have to clearly understand what constitutes a crisis, and a few negative tweets isn’t it.

2. Twitter is a living focus group
Business has wanted to eavesdrop on customer conversations since the time of Pompeii, and now they can. Provided they were listening (and I very much suspect they were), Frito-Lay garnered hundreds of near-instantaneous points of feedback about their new advertising. Powerful mechanism. That does not mean that the people who liked the commercial will buy more snacks, or that the people who hated the commercial will swear off Tostitos. The belief that sentiment and social mentions lead directly to commerce is a dangerous assumption, and requires your own verification and testing. Results can vary wildly based on company and intensity of feeling.

3. Could Twitter be used for on-the-fly ad optimization?
As mentioned, there are at least two of these talking bag commercials. If I recall correctly, the one above ran, then the other one, then the first one again. Based on the volume and sentiment of tweets, could Frito-Lay have determined that the initial spot was better received (and it was, according to the tweets), and then decided to run it again as a result? For a major event like a college BCS bowl game, it’s not inconceivable that Frito-Lay could have been in contact with ESPN in real-time, and told them which spot to run a second time.

This has implications for rotation and optimization of broadcast. What about running a different ending to a scripted show show on the west coast, depending upon the tweets from eastern and central time zones? Hmmm.

4. Sentiment scoring remains an inexact science

Even with the excellent technology in Sysomos, we still had to manually review all tweets and recategorize a few from positive to neutral, neutral to negative, and so forth. Automated sentiment scoring is nearly impossible, and if you or your listening team aren’t at least spot checking sentiment, you’re probably dealing with data that’s not entirely correct.

5. Differences in social listening software are vast

We ran the same exact report in Visible Technologies, and found approximately 150 tweets for the same phrases and date ranges. We like Visible a lot. It’s solid software with great user interface. But the different between 500 results and 140 is pretty vast. Just like with Web analytics software (where the data differential between Webtrends, Omniture, Google Analytics and the rest can be large), there is no objective “truth” in social media listening.

6. Did the agency use a crowdsourced idea to devise the talking bag concept?

On YouTube, this commercial for a talking Doritos bag was uploaded as a response video. The description notes that it was created for the Doritos Crash the Super Bowl crowdsourced video contest in 2010. Interesting that both snacks are Frito-Lay brands, and that entries of the Crash the Super Bowl can be used and modified by Frito-Lay without exception, and in perpetuity.

  • jasonkeath

    Based on some actual conversations I have had with tv ad types, they have been using Twitter to tweak commercials for a while. Just probably not as quickly or as often as we would like.

  • allarminda

    @jaybaer FYI govt wifi denied me access due to inappropriate material. I had no idea Tostitos were so polarizing.

    • jaybaer

      @allarminda Interesting. Not sure why that would be. Hmmm.

  • ShakirahDawud

    I thought the difference in the data you found was intriguing. I wonder how many companies who use social listening software take that into account.

  • 40deuce

    I definitely think that people can use this type of data to tweak their commercials and messaging. I’ve seen commercials that ran one way one week, but the next week there were minor changes to it. I actually wondered at the time if that was due to feedback or they just wanted to make it shorter to save money. I also don’t know if they act as quickly as calling up ESPN and saying “actually, run the first spot again instead of the second one we sent,” but we may not be far off from a time when that happens.

    Also, to your manual editing sentiment statement, I’ll be one of the first people to say that automated sentiment is far from from perfect. However, it will give you a general idea of the mood of the conversation. As well, when you changed the sentiment in Sysomos, the software is actually learning and takes those changes into account down the road when it finds similar content. We’re constantly trying to improve it, but I don’t think it will ever reach 100% accuracy (on ours or anyones software).

    And one last personal point, as someone who lives on the east coast, I’d be pissed off if you west coasters got a better ending to a show than I did.


    Sheldon, community manager for Sysomos

  • sdolukhanov

    Whats even more useful than knowing how many mentions of the commercial are “positive” or “negative”? If I’m Frito-lay, Importing traditional campaign-tracking metrics that tell me how many viewers watched each commercial and the assumed conversion rates in to the same software that gathers and analyzes the social mentions for sentiment on a numerical scale as opposed to positive or negative, and using this information to derive best practices for future ads. Knowing how many mentions something gets is useful aestetically, but if you don’t understand how it correlates to your traditional business metrics it won’t do much for you in the way of increasing sales. I don’t believe this is something Sysomos or Visible can do.

    – Sergei Dolukhanov


  • keithkorneluk

    I thought the commercial was terrible as well but it did elicit an emotional response from you and you blogged about it on a site that has a good amount of reach. Mission accomplished.

    And, let’s be honest, a commercial like this plays to the lowest common denominator. Most marketing people would hate it. It’s pretty clear that the general populace liked it. You were watching football. :)

  • Michael Besson
  • Wittlake

    Jay, yes, ads can be polarizing, but roughly 1/3 negative? If a marketer is pleased with this, the goal is to create conversation, and they are willing to have an ad that is disliked by a sizable portion of viewers in order to create that conversation and increase visibility.

    If I’m right, you just made them more successful! ;)

  • meenrico

    Great post, Jay, and thanks for the mention. I definitely agree that businesses have a lot to gain from listening in on conversations and indeed decisions can be made on the fly by monitoring what is being discussed on social channels. Sentiment will likely never be perfect. Visible’s multi-language sentiment analysis was developed based on over six years of data and the world’s largest collection of human-scored social content for the most advanced automated sentiment analysis. Basing our sentiment algorithm on actual social content ensures the highest level of accuracy.

    In certain instances, volume is a critical measurement. Considering that the primary goal of social media data is to inform business decisions in ways that traditional data is not capable of, it stands to reason that being able to discover channels, authors, and topics which better inform the business is critically important.

    It will be interesting to see this evolution of how well businesses monitor and listen, then react. I can’t help but remember the Gap logo debacle from Fall 2010.

    Thanks for sharing your great insight.

    – Ellen Enrico

    Director, Community @Visible