Monday, 8 July 2019

Data-driven is at risk (Bias, horizons, migration part II)


So ok, people are getting visibly more selfish and extreme in their views, so what?

In my previous blog (1) I used some numbers around the elections of Mr Trump and Brexit, and showed that the data was clearly pointing towards Mr Trump winning because of white people, and Brexit happening thanks to older people. I also explained why it is likely that we will see more and more of extreme views being taken and exhibited by people. In this blog, I consider what, as analytics people, we should expect and how we could deal with this new reality.




Data Driven is at risk

People in the analytics field have waited for the aim of becoming data-driven for a long time. Using data to help organisations make the ‘right’/’optimal’ decisions is what we do. 

For many years we were confined to very specific business units such as marketing/customer management – where we increased product take up, decreased churn, increased the value of customers to the organization while decreasing the costs, risk – where we help manage risk and find optimal points to balance risk and return, red flagging potential issues before they happen so that pre-emptive action can be taken, or as they are taken to help relatively objectively re-evaluate cases, operations – where we increase efficiency by making processes smoother, decreasing TAT, optimizing resources and resource utilization, human resource management – where we help select suitable candidates, reduce employee churn, flag potential issues ahead of time and so on.

Finally, with all the buzz around “data science”, ML. AI, DL, the concept of using data to make decisions became mainstream, and the idea of expanding the use of analytics to the organization as a whole started taking hold. So we has Chief Data Officers, Chief Analytics Officers… with associated teams who actually walk the talk, including specialized analytics consultancies such as experfy(2) and AlphaZetta (3). These organisations provide expert support in very specialized roles across the analytics spectrum, although less so on the dev-ops side where the market is very well stocked.

In fact, this emerging gap in the market caused management consultancies to try and create, or buy their own analytics shops, some more interestingly (4) than others (5).

But this new trend of visible extreme runs counter to data driven.

I mentioned the gap in the market above; not only have specialized organisations sprung up to address this gap using human experts, but some organisations, taking advantage of the lack of suitably qualified humans have gone the systems way and codified specific procedures in to generic solutions that may be applied in different cases, for example data robot (6). To me this just makes it easier to pick the answer you like; many algorithms are thrown at the data and a user can get to pick the one that suits them. I am not saying that a human wouldn’t do this – in fact running algorithms is probably the easiest part of the analytics process – but automated systems allow this to be done easier. It’s not the technology, it’s how it is used (and by who).



The second coming of the HiPPOs

In this murky environment, the HiPPOs are making a return. HiPPOs are decisions made by going along with the Highest Paid Person’s Opinion. Despite the hype around “data scientists” remuneration, usually they are not the HiPPOs.

Instead of being driven by data which means that in different circumstances, different positions are optimal and should, in true data-driven fashion, be taken, extreme and inflexible decision makers look for evidence that suits them.

In an environment where it is becoming more acceptable to express and push extreme opinions, where if you really want you can find some interpretation of data that suits your views – usually in the websites/newsfeeds you go regularly to/are recommended to you. After all, we are living in the world of “fake news”(7) and “deep fakes”(8). The fact is that very few people or organisations have the skills and resources to question the data, and in such an environment, it becomes very easy to drive a personal agenda.

A little knowledge

Another related issue is that everyone thinks they are experts, a little knowledge is bad (9). I once was in a meeting when someone said: “It’s easy to become a ‘data scientist’, you can just take a couple of hours course online! I’ve done it myself!”

I think online courses are great. Democratisation of knowledge is good because it allows for healthy informed debates, and eventually less bias/ However, there are all sorts of courses online, and it is very easy to spend a lot of time, effort and money on courses that are not that useful. 

Relying on wisdom of the crowds (rating systems) is not necessarily the right thing, especially for introductory courses which create the foundation upon which further knowledge will be added, simply because the majority of people giving ratings would themselves be newbies, hence the ratings would likely reflect how the learning took place rather than advised opinions on the quality of the learning. There has been some effort to build curate curricula out of the publicly available courses, but not everyone can benefit from them.

Correlation Co-Co-your-head

How many of us have heard the phrase “correlation does not mean causation” in the office or even at social events? Or even worse: how many of us have heard people use the word “correlation’ in everyday conversation? Especially when people use “correlation” to mean a relationship? (The most common measure of correlation in fact, only describes/measures linear relationships(10)).



Hence paradoxically, while more people are believing that being data driven is a good thing, we are getting less data-drive because:


1 It is getting easier to find “supporting evidence” by choosing algorithms after the fact based on their results. The emergence of ‘automated’ algorithm testing software plays a big part in that.

2 The proliferation of “data science” courses combined with the lack of “quality”/”suitability” checks gives the impression that “data science” is all about a couple of catchy algorithms let loose on the data, any data.

3 “Fake news” is becoming more and more common; in some cases, people repost and give credence to stuff without doing research into whether it is true. Not checking is not new. For example, it took 3 years and countless applications to realise that evidence in favour of feeding patients 80% oxygen after operations was flawed, and so was the WHO’s advice based on these paper (11)



So what can analytics people who choose to walk the talk do about this?

Unless you are given the mandate and the power, don’t even think of changing organisations

Most of us end up working for organisations, whether we are actually freelancers or contractors or employees. Depending on the level or role we are engaged in, there are cases when a nice carrot of being given a chance of making an organization become truly data-centric, or at least taking part in an effort to make an organization truly data-centric. Here are a few things to look out for.

1 Industry

The industry you are in makes a huge difference. It is arguable that the degree of exposure to digitization in an industry is a good gauge how likely more people in management would be on board the data-driven train.

A group of people from McKinsey wrote a very interesting article in HBR (12) that uses data and provides possible explanations for the findings. I have reproduced the diagram at the heart of their paper below:
see



One of the interesting things about such surveys is that the financial services industry is one of the most digitized. I think it is necessary to split banking and insurance. 

I have worked in both banking and insurance, and from my experience, insurance has far more dinosaurs alive than banking does. For example, I have met a CIO/CTO of an insurance company who was convinced that the best way to deal with migration of dirty data is to hire students as temps and re-key in the whole database (and I am not talking about only hundreds of records).

But this also means that the rewards for becoming data-driven in the insurance space are huge. And there are start-ups that are trying to bring modernity to insurance. However, there is a long way to go.
In sum, you can use the study by McKinsey as a guide but remember, insurance is much lower down the order.

From a business perspective, it is much easier to ‘sell’ analytics in an industry where it is already quite accepted; in fact many organisations would be looking around for partners; not being left behind is a powerful motivator.

2 strong leadership

If the adoption of digital in an industry, then, if there is to be data driven, the decision and support has to come from the top. Changing people’s mindsets is not easy, and we may disagree on what makes effective leadership, but I am quite sure we would agree that the ability to stay the course is critical in a leader who wants to implement change towards data driven.

You may argue that I am shooting myself in the foot, because I have been arguing that HiPPOs would be winning, and you are very unlikely to have a bigger HiPPO than the CEO. 

However, a HiPPO can easily win any battle, but change is more of a war than a battle. Furthermore, if there are opposing (or even just old-school status quo) voices in the organization – or people who want to become the new HiPPO – then any change will be in a very stop-start/one-step-forwards-two-back fashion. 

Basically without strong leadership, an organization cannot change, or will change at the pace of its slowest/least-willing/biggest bully executive.

In the case of analytics/”data science”, there is simply no point trying to effect change without strong leadership. It will just lead to frustration.

For consultancies on the other hand, doing tactical projects can be a decent source of income, as long as success metrics are defined very clearly at the outset; however any project where delivery cannot be measured objectively should be avoided.

3 Results based culture

It may sound obvious to analytics/”data science’ people, but not all organisations are results based. 
You do not need analytics to be results based, not only in sales but across the organization, such as in operations and even departments like marketing, PR and comms…

An organization who has the mindset of measuring outcomes, setting KPIs and rewarding their people according to these KPIs will quickly adapt to the use of analytics/”data science”.

But if an organization still rewards people subjectively, focuses on ‘effort’ (or worse relationship) rather than outcome, then the first battle for use of analytics will be over the need to measure outcomes. 

That’d be a lot of change to be done, and implementing analytics is hard enough without having to fight that battle.

Furthermore, an organization that does not have a results-based culture is most likely to be personality driven, HiPPOs will fight for their corner of the murky waters… 

In sum, it would not be a good organization for someone into analytics to join immediately.

As for consulting, it may be useful, but focus on short term projects that minimize bruising from HiPPOs, that have at least clear measurable outcomes. These may or may not grow into longer more transformational projects, or even longer term contracts. But as long as the culture does not change, it might be better not to prioritise such opportunities.

Summary 

Nowadays people can find like-minded people easier, and this has led to more visible and unbending extreme stances

Furthermore, the craze around ML/DL/AI/”data science” has increased demand for analysts/”data scientists” but also allowed some not so accurate beliefs to seep into the collective mind.

However, the increased demand, has also increased the supply of training and courses that supposedly prepare one for a career in analytics. But the quality of these is hard to ascertain. 

This sometimes creates people with a little knowledge that may be dangerous. This is worse when that knowledge belongs to executives, or worse HiPPOs.

Therefore, as practitioners, while the goal of “data-driven” organization is very enticing, most of us end-up being disillusioned and frustrated.

In order to minimize chances of this, I have proposed 3 aspects one should be very mindful of before getting involved in a gig/contract/job.

1 Industry – different industries have different maturity I use of analytics/digitization. In a more mature industry, it is easier to find people to support the efforts to be data driven. In others, you may find more blockers. As an employee that can be real frustrating.

2 Leadership – sometimes a good leader can drive change. Finding a strong leader is critical if an organization has to become data-driven. Change has to take place across the organization, and mini HiPPOs, status-quo people will drag the process down, deflect issues and focus on their comfort zone BAU. A leader who cannot cut through this is not someone you should follow as an employee. As a contractor, focus on measurable outcomes, to minimize risk, and report directly to the project sponsor who has to be right at the top.

3 Culture – any organization that wishes to become data-driven has to measure results and reward people accordingly. If an organization is not already doing that, the battle to become data driven will be long and painful. So someone from the analytics field should really reconsider joining such an organization. However, this is good hunting ground for gigs/short contracts as long as you can ensure metrics are used as acceptance criteria; this would help minimize chances of being crushed by HiPPOs.

Conclusion

If you find an opportunity to make an organization data-driven, and this organization is in an industry not traditionally associated with analytics, whose leader is more of a consensus/approval seeking person and where the organization doesn’t focus on results, then it is probably better to look elsewhere if you are looking for a longish work stint.

On the other hand, an organization that is in an industry that has been toying with analytics for a while, has a results based culture and has strong supportive leadership, then lucky you.























No comments:

Post a Comment