Blog

Twitter-data---sorting-the-problems-from-the-positive-feedback

Twitter data - sorting the problems from the positive feedback

We explored social listening in a previous blog and covered the differences between generic models, and tuneable, granular models when looking for customer insights from Twitter and other unstructured text-based feedback.

To bring this to life, let’s look at packaging. It’s a timely example to use as it is a growing topic as consumers voice more and more environmental concerns, as well as the usual quality issues relating to packaging.

In a generic text analysis model, you might be lucky to pick up packaging issues at all, or at best to pick them up and assign positive or negative sentiment. However, this doesn’t help to action anything, not without further reading and coding. One must also figure out what the themes are to code in the first place.

With ‘human-in-the-loop’ software like PrediCX and the S4RB model specific to grocery retailer, the new signals are referred to a human as they appear so that nothing is missed, nor does it have to be guessed. The data truly speaks for itself!

Let’s take some publicly available examples from Twitter which are also representative of other queries in private data sets e.g. contact centres.

  1. Theres child proof. and then there's human proof. I swear @tesco own #bleach is human proof. You need muscles AND Phys strength to squeeze, push down and turn the top. I've neither. Anyone else struggle
  2. Lesson learnt from buying cheap @Morrisons bleach... Clearly don't spend money on the packaging as it just leaked all over the weekly shop

These two Tweets are interesting as they provide direct contrast on comparable products – own brand bleach. The first one is too difficult to open and the second one is too easy. The customers’ views are that these are design issues. This may or may not be correct, i.e. they could be quality issues but it doesn’t come across that way, nor is there any hint at damage in transportation.

So, you might label them: “Package Design – Not Easy to Open” and “Packaging Malfunction” respectively along with the product identifier, i.e. in this case “Bleach” for both. It might also be appropriate to state “Ruined something else” or “Packaging Leaks” for the second one, particularly given the potential cost and health impact. Indeed “Health and Safety” might well be a flag for the second one.

  1. Hi @Tesco, just got back from one of your stores. Whilst walking around the store it turns out a bottle of soy sauce I had been carrying was leaking as I walked around the shop. This resulted in a train of soy sauce on the floor but more annoyingly my lovely suede coat, dress,
  2. @Morrisons just received my online shop and the lasagne has arrived with the cellophane open, please can I have a refund for this item? Thanks

Tweets 3 and 4 above are comments about packaging malfunctions, but with hints to packaging quality rather than design. Therefore they might be labelled “Packaging Malfunction” although the first one also might be labelled as “Ruined something else” or “Packaging Leaks” as well as the product mention itself. The second one is also from a delivery so a label with “Home Delivery” would potentially assist in case it is a transit problem (i.e. it could have been packed up something heavy).

Beyond sentiment – which is clearly negative, this allows you to see common themes across categories.

New call-to-action


  1. Nearly sliced my finger open trying to get into a packet of @waitrose figs
  2. @Tesco check this out! Bacon I can open one handed. Unlike your packaging that I have to attack with a knife because for years the pull tab has not worked once. @AldiUK #voodoomagic #bacon

Tweets 5 and 6 are both to do with packaging that is “Package Design – Not Easy to Open” similar to Tweet 1, but there are other labels that might be appropriate such as “Health & Safety – Near Miss” for Tweet 5 and “Mentions Competitor Package Better”.

Again, beyond sentiment this richness allows brand owners to understand competitive advantage or disadvantages, which can feed into either marketing or product development.  The ease of opening on Aldi’s product will be for more than just bacon!

Also, for Tweet 6 there’s a hint of long-standing Tesco customer so can add label: “loyal customer”. These tags can both be used to help improve packaging, avoid serious issues, and also improve the brand’s standing to competitors in terms of the features that customers mention. What’s interesting in Tweet 6 is that a longstanding, loyal Tesco shopper has made an unsolicited comment to Tesco about a competitor. Have they switched? Imploring their favoured brand to improve?

  1. @waitrose Do you allow bringing your own containers to fill at the fish/cheese counter? #reducewaste
  2. @waitrose If Aldi can use cardboard why can’t Waitrose?!
  3. Hello @coopuk as members, how do we share opinion: your waste policies are great, but I'm not the only customer who would appreciate less packaging on fresh produce ...
  4. Don’t usually shop in @asda but popped in today - shocked that 99% of fruit and veg is in plastic packaging with hardly any loose options. We walked out and won’t be back @AsdaServiceTeam #plasticfree #plasticpollution

Tweets 7 to 10 are all to do with environmental concerns with packaging and could all be labelled with “Packaging – Environmental Concerns” per se. However, there are subtle additional noteworthy comments:

  • Tweet 7 could be labelled “Customer Suggestion – Reducing Packaging Waste” as the customer has an interesting suggestion rather than just a complaint or observation.
  • Tweet 8 could reuse the label mentioned in Tweet 6: “Mentions Competitor Package Better” and again the intent of the customer seems to be to support Waitrose rather than a Parthian Shot.
  • Tweet 9 is a positive message from a seemingly loyal customer expressing concern with packaging wastage on particular items (so add “Loyal Customer” and “Fresh Produce”).
  • Tweet 10 is a stronger version of Tweet 9, so strong in fact that a non-loyal customer has said they won’t ever shop at Asda again. The tags: “Non-Loyal Customer” and “Churn – Won’t ever Shop with Brand Again”) could be added to measure the trend and strength of this signal.

In conclusion, generic text models are OK for picking up general topics. However, if you want to pick up truly actionable, granular signals then the combination of AI plus 'human-in-the-loop' will help you uncover those unknown unknowns. All within the context of an industry specific model, which is why Warwick and S4RB have collaborated on an industry specific machine learning model. 

Dan Somers, Warwick Analytics

Share this blog:
 

Leave a reply

Subscribe to our monthly blog digest

Recent posts