The Curse of Bad Insight

If right is right than left is wrong, right? If smart is well dressed, then a lack of fashion sense must imply stupidity, no? I once went on a diet, swore off drinking and heavy eating, and in fourteen days I lost two weeks.

You get my drift. Too often ‘insight’ just doesn’t deliver because it doesn’t add value, it doesn’t make sense, or even if it does, there seems to be nothing you can do about it. There are so many organisations out there who are in the business of ‘social analytics’ or should I rebrand it ‘transforming big data into actionable insight that drives topline growth’. Not to diminish the incredible value of great analysts, but even monkeys fall from trees. The reality is that the chasm between big data and insight and insight and action is as wide as it was before.

How often do we read how data is ‘gold’, or ‘the new oil’ whilst one of my friends coined it: big data is like garbage, useless on its own, yet priceless when processed. I will add a little build to that: priceless when processed correctly, and in a fast and efficient manner.

What is bad insight?

Assuming there are people who work in research and who can find meaning in even the tiniest correlations - think chaos theory - I would personally classify bad insight as:

1. Stating the obvious,
2. Drawing the wrong conclusion,
3. Not action oriented,
4. Missing the boat.

No shit Sherlock!

The most recognisable form of bad insight is stating the obvious.

If you want to give an over confident sales manager a field day and a renewed appetite for a crusade against insight driven decisions, please do confirm his or her intuition.

I remember clearly when someone proudly presented me two graphs, beautifully correlated. The chart looked like two perfect mountains arising mid-year in the May-September period. One represented the number of mentions on social media of the word ‘barbecue’ and the other represented the hours of sunshine per day. Although there is nothing incorrect with this analysis and the correlation is valid, it did not teach me anything I didn’t know already. I kindly offered free insight on the spot: the number of mentions of the word ‘Halloween’ and retail sales volume for pumpkins, or maybe the number of mentions of the word ‘Thanksgiving’ and the decline in the turkey population in the United States. If your mind is drifting towards further anecdotal examples let me assure you that #bunny and chocolate sales only work when you put in corrective measures for Playboy.

Joking aside, the above are obvious examples we can all relate to. Yet it happens too often a well-meaning analyst makes this basic mistake by presenting a great insight to a group of rolling eyes. A few years ago, I eagerly set up a meeting with the brand director of all cordials in the company I was working for to bring to her attention an issue with that year’s batch – somehow, we messed something up in production and ended up diluting our produce and everyone was up in arms about it on social. I walked away in shame 30 minutes later after I understood that we only use fresh berries. And that not every year is the same, so sometimes the summer is poor and so is the harvest. Which in turn means the berries aren’t as sweet as the year before, giving the impression of dilution. It’s a bit like good and bad wine years. So clearly my ‘revelation’ was something she already knew long beforehand.

Although this was a failure back then, fast forward 8 years where I am privileged to work with an indie juicer who used the above insight to challenge the status quo. After a good harvest year, a portion of the fruit is stored in liquid nitrogen. This preserves all the goodies and the flavour and if next year’s harvest is poor, the frozen batch can be added to prevent a change in taste profile. So instead of making the wrong assumption nearly a decade ago that we had a production fault, I failed to see the real underlying insight of a production improvement opportunity that could mitigate the dilution effect with no reduction of the nutritional value of the fruits used.

Missing the link

Social analytics are highly dependant on correlation, which can then be further dissected with Boolean logic. This is its power, and its Kryptonite. When searching for novel ways to encourage consumers to eat more peanut butter, a thorough social media analysis found a strong correlation between sliced melon on toast and peanut butter. As in the example before, there is nothing wrong with this conclusion per se. The correlation however is not causal, and that’s when we get to call Houston.

There is indeed a strong correlation between toast and peanut butter. Equally there is an interesting seasonal correlation that, in spring and summer, people eat more melon. In turn correlation indicates melon can be eaten on toast, and in turn a spread can be added to enhance the flavour. This can be peanut butter, but equally so almond butter, Nutella chocolate spread, jam or marmalade.

We have a classic case of three correlations going on at once, all significant in isolation, yet not significant when combined:

1.      Any spread (incl. peanut butter) + toast

2.      Melon + spring/summer

3.      Toast + melon

One of the by-products of a human analyst is to fall victim to the lure of deduction and make the subjective jump that there is a correlation between 1 and 3 also, hence melon and peanut butter on toast makes perfect sense. Now if peanut butter was statistically significantly stronger than any other spread with melon that would be correct. In this case, it was not and a lot of money would have been spent on a campaign that would not have yielded a return in measure.

A recent example that most of you will have read about is Pepsi. Again, by doing everything by the book, the marketing team concluded that their target audience is young, they like Kendall Jenner, yet they also want to be relevant and care about big socio-political issues. All these correlations were true in isolation, but when cross-correlated someone should have pulled the plug. You can like Kendall, and politics, but loathe Kendall in politics. And you may like Pepsi independent of being socio-politically active, so blending both has a calamitous effect.

So next time you conclude that your target consumers love puppies, and oceans, and support the survival of endangered marine life, you may want to think twice before sending the Andrex puppy on a longboard into the ocean to be devoured by a great white.

Erh, and now what?

One of the buzzwords that drives business people insane is ‘actionable insight’. In all trueness, it is harder than it seems. Finding an interesting factoid, or a seemingly anomalous correlation is not that difficult. To determine whether this is genuinely interesting enough to warrant a brand director’s attention or to figure out what to do with this insight is a lot harder than it seems.

I spent a lot of my time working in alcoholic beverages, and one of the big issues is that women don’t drink beer (at least not with the same vigour and frequency as men). Countless years have been spent on ‘problem admiration’, so by the time I started considering the issue, there were thousands of pages and graphs explaining why women did not like beer. I could however, not find a single insight that proposed what do to about it, other than trying to “de-beer” beer, or as I learned in my days working for an energy drinks company: “to pink it and shrink it”.

Instead of looking why women do not like beer as much as men, we started to look at what women wanted from a beverage full stop. It came as no surprise that (in certain parts of the world where I was working back then) the main culprit was the alcohol. Ultimately, we had two routes we could take: the painful route of re-educating women to embrace the wonders of alcohol, or to simply manufacture the beverage they wanted. And that was the “actionable” insight: stop trying to morph beer into a “woman friendly version”. Start producing a non-alcoholic, non-beer beverage using all the benefits of fermentation which means a natural production process and no sugar. Seems simple when you read it, yet it took years to get to this place.

Even today, I always cringe when people who should not talk about analytics put this statement on a slide: “American Express can see 6 months in advance when you are going to get divorced based on your spending habits”. Ok, cool. So, unless American Express is going into the marriage counselling business, or miraculously GDPR (Global Data Protection Regulation) stipulates that highly confidential PII (Personally Identifiable Information) can be sold on the black market there is nothing that can be done by American Express to act on, or make money from, this insight.

Every time an insight is uncovered, and it is genuine, one needs to quickly establish if it is actionable and how.

Let’s take this simple example: seaweed is back on the rise. What does that mean? Are people interested in seaweed because of its flavour? Or because it is sustainable? Or because it has certain health benefits? Very quickly the ‘actions’ start to look quite different. If I was working for a food company, a bias towards flavour would mean I could just flavour any odd product to taste like seaweed (e.g. Seaweed Biscuits). If it’s the health benefit, I may want to consider selling dried seaweed in its purest form. Add a sustainability angle and I would make sure that the seaweed I procure is ethically sourced, organically grown with no ocean life cruelty involved and a percentage of the profits going back into a local ‘save the seas’ fund.

Every insight presented to the operational side of the business should end by clearly stating what can be done, and how success will be measured.

I wish you’d told me sooner

Social analytics is reactive by its very nature. We are merely listening to the conversations online, giving a real-time view of what people are talking about. Whatever we do with that information allows us to react, not anticipate.

If only I had known the effect of my campaign/new product/product retirement, then I wouldn’t have ended up having to pull the add, recall my product or reinstate it. Months, even years, of product testing, taste panels and virtual stores are no guarantee that your endeavour will be an instant hit with the masses.

Big decisions for the future have been made based on real-time social information. This is as accurate as playing the roulette and betting ‘black’ three turns from now because it’s ‘black’ right now. Unless you have a 3 week ‘from design onto the shelf’ capability like Boohoo.com or Asos.com, there is always a risk that you miss the boat, or the boat sank by the time you arrive. When social media picked up on kale, Gwyneth and Chris had been sipping it for months, consciously uncoupling.

The insight that is most valuable allows you to make a future decision accessing an opportunity or mitigating a risk. It gives you foresight. Most insight today gives you (real time) hindsight. If you are using social listening to course correct, you probably did something wrong in the first place. I would rather use social listening as an opportunistic tool that scouts the landscape for immediate opportunity in areas where I can interact with lightning speed (e.g. digital marketing, customer care). For everything else, social media is the surprise bearer of good, or bad, news and a painfully effective tool to share success and – especially – failure (think overbooked flights and heavy-handed flight attendants).

And you know what? It will get even harder.

Boys will be boys …

Stop it right there DJ. That was yesterday. Today boys could be girls, girls could be boys or both. LBGT is widely accepted as an extension of past demographics based on sexual orientation. However, after Facebook offering no less than 71 ways to describe your gender, segmentation as we know it is a thing of the past.

I call this demographic fluidity. In segmentation, the demographics were relatively fixed for a while. You remained male or female mostly and moved through the age groups slowly. Today all of this is fluid and depending on the occasion, my gender, or mental maturity, could change.

The same goes for age brackets, and especially labels. I sincerely hope that nobody reading this post still believes you can target ‘millennials’ or ‘Gen Z’. Frankly labelling any young person with either will result instantly in you being reverse labelled somewhere far south on the IQ scale. ‘Millennialism’ is not about age, it is a state of mind.

The demographic axis is no longer fixed and can be bent according to the situation. This makes it particularly hard to segment and target your consumers. And that’s the people, we’re not even talking about their avatars.

Meanwhile, in the US, “a philosophy class had a debate/argument about whether having sex with your clone was masturbation, incest, just homosexual, something else, or a combination. It was inconclusive and people were upset by the end.”

Do you speak emoji?

Social analysis was hard enough 7 years ago when we only had to cope with text. CAPS on is shouting, sick means cool and in South Africa ‘Colgate’ means toothpaste (“my favourite Colgate is Aquafresh”). There are numerous pitfalls that we learned to avoid in social analysis. I remember analysing the tobacco industry and having to make numerous exception rules for the word ‘fag’, which is slang for a cigarette and also a derogatory word for a gay person. A friend of mine looking after the Walkers (crisps/chips) brand has had quite a challenge since both ‘The Walking Dead’ and ‘Game of Thrones’ became hit-shows.

Then we had text speak to deal with: OMG, LOL, IMHO, TL:DR, UG2BK, WTF?

Yet today it has become even more challenging. The continuous rise (and volume) of emoticons is but one evolution, creating an entirely new language of pictures, the ‘hieroglyphs of social’. The arrival of Instagram, Pinterest, Snapchat (to name a few), has augmented the social landscape with imagery. A picture says a thousand words. Literally. So how am I going to capture and analyse these thousand words?

The world we live in is absurd

Maybe this is a strong statement, but hear me out. None of the (traditional) polling efforts predicted a Trump presidency, or Brexit, or the last UK election result. There have been loads of explanations as to why that is and here is mine.

In high school, I learned about logic, and about false logical arguments. My favourite was ‘Reductio Ad Absurdum’ – or reducing the argument to absurdity.

Let me give you an example that I found online: “I am going into surgery tomorrow so please pray for me. If enough people pray for me, God will protect me from harm and see to it that I have a successful surgery and speedy recovery.”

We first assume the premise is true: if “enough” people prayed to God for the patient's successful surgery and speedy recovery, then God would make it so. From this, we can deduce that God responds to popular opinion. However, if God simply granted prayers based on popularity contests, that would be both unjust and absurd. Since God cannot be unjust, then he cannot both respond to popularity and not respond to popularity, the claim is absurd, and thus false. Logical, right?

Now fast forward to today, with social media in full swing.

“I am going to complain about this unfair thing that has been done to me. If enough people like it and share it, the CEO of the company responsible, or even the President of the United States, may notice and ensure this situation is rectified.”

Now you tell me; is this claim still absurd? Surely the President, who remains impartial, cannot both respond and not respond to popularity? Didn’t think so.

Somehow social media has been able to turn the tables upside down and finding insight in an absurd world is even harder than ever before.

Where do we go from here?

I am robot

I am a huge fan of AI and machine learning. One reason is my endless appetite for sarcastic humour. Watching Microsoft’s Tay - their first AI chat-bot – going rogue had me in tears daily. I wasn’t laughing because it was a failure, I was laughing at its success. It became the ultimate caricature of the right-wing, racist, homophobic Twitter rant we are so familiar with. Tay became the uber-troll.

Another bizarre development was when Facebook’s bots began to talk to each other, in an entirely new and made up language, that made communication more effective. Or that priceless moment when Amazon’s Alexa starts having a conversation with Google’s Home.These are but the first steps into the amazing world of AI. These first steps are already showing promise, and since it is machine learning, with emphasis on the latter, it can only get better.

I came across this cool start-up called NextAtlas who use AI in social analytics to shake things up a bit. Their algorithm does two things differently: firstly, the machine hunts for ‘insiders’. Social listening listens to everyone, with an equal voice, so the ‘insight’ is relatively dumbed down. Other tools listen more to ‘influencers’ but these people are typically paid (up to $50,000 per tweet) or sponsored, so hardly objective. Now insiders are ‘in the know’, typically not sponsored, nor paid, and seem to have quite a strong track record of knowing things ahead of the masses. The second part of the algorithm correlates the text and images of this avant-garde group testing infinite combinations seeking anomalies and deducing forward looking patterns, or trends, from it.

For the first time AI can use social to give you something entirely unique: foresight. 

What do you want to DO? Not what do you want to KNOW?

When social analytics entered the workplace, it was a fantastic new toy, and everyone fell in love with it instantaneously. What are people saying about my brand? What is their favourite flavour? Is my campaign being talked about? These were insights that you had to pay a lot of money for before (panels, surveys) and now it’s just all out there for grabbing. This has lost none of its allure, only last week I was talking to Avis who wanted to know what people complain about most when using their services. Easy – get a listening tool and off you go.

Today however, what is being asked from social analytics is far more sophisticated. The analyst’s job is no longer to provide random insights or sentiments, but to solve very real business issues. Examples are: I am going into the flavoured sparkling water category, what is the next big flavour I should go for? A travel agent wants to know what the next big destination is going to be in 2018, given the current landscape where people are avoiding North Africa, the Middle East and even the USA. Another organisation is figuring out the next ingredient, and yet another is pondering to replace their entire core business with a new venture.

The relationship between operational business and analytics teams has fundamentally shifted.

In the past data scientists had to look for a needle in a haystack and figure out how that insight could be found and how best used. It was very much a push model. When the business engaged proactively it was typically with a very specific question they wanted answered (who are my influencers?), limited and informative in nature. This has been gradually shifting to the business partnering with analysts to work in a more outcome driven manner. The business will no longer come with questions for information but with actions they are planning to take. The analyst’s role is to fine tune, enhance and even design what the solution will look like.

The Solution Scientist

The evolution of the insights professional is about to take its next turn. When the tools at our disposal a few years ago were complex, and the algorithms took a lot of human intervention, we needed data scientists. Today, data scientists have become outsourced and they are a lot cheaper than before. The tools are ever more intuitive and programming algorithms is less about code and more about customising an AI engine. The gap that still needed to be bridged was to translate the insight into something the business could action. These new roles have been given various names: the insight scientist, the storyteller, the business link, even the roll-off-the-tongue analytics front office person. Again, with AI pushing forward, these roles are starting to become less and less relevant as the operational business can interact directly with the machine.

What is still lacking is the complete integration of the analyst into the operational business. In most organisations ‘insight’ means a PowerPoint presentation that is thrown over the fence, sometimes with a nice person presenting the findings. The actions taken from there (if any) are hardly ever fed back to the analyst, nor are the outcomes measured. This makes it hard to prove the real bottom line value of the insight. As time moves on, it is inevitable that the analyst will become accountable for the quality of his/her insight. Success means insight that makes money, increases brand equity, improves sustainability, etc. The analyst’s performance will be outcome based, much as anyone else in the operational side of the business. Analytics CoE’s will have a full P&L and not just an L.

This calls for a new type of knowledge worker: The Solution Scientist. He/she will work with a brand, a category, a department to not only deliver an insight, but to co-create the solution to access the opportunity. The job an insights person is doing today will be completely taken over by AI. The job the insights person will do tomorrow is to build winning solutions.