Tuesday, 3 January 2017

Why you don't need a full-time "data scientist"



Most organisations today do not need a data scientist, at least not as a full time employee. The reason is very simple, the RoI on a “data scientist” for most organisations is way too low (you actually might be losing $). Why is that and what can you do about it?

HBR came up with an interesting article “most industries are nowhere close to realizing the potential of analytics [1]. The reasons pointed out by HBR is issues with finding people who can apply “data science” to business questions, issues with embedding analytical output into business operations, and required business transformation.

All these points are definitely valid. However, it seems that they are focused on what I’d term ‘mature’ markets and organisations. However, not all organisations are mature enough. I will break down maturity into 2 aspects.

First is competitiveness. For an organisation to take advantage of “data science” there needs to be a will to do so; the organisation must be geared towards extracting every piece of advantage they can. It has nothing to do with the size of the company; in fact many ‘disruptors’ are small but highly competitive. “Data science” can give them the comparative advantage.

Competitiveness gives the will.

Second is the IT maturity of the organisation. For an organisation to take full advantage of analytics/“data science” it must be enabled with adequate IT infrastructure. The IT maturity is important since it enables the “data scientist” to access all sorts of relevant data, analyse it, find patterns and make predictions, which are then tested and made part of normal operations.

IT maturity gives the way.

In my view, the HBR article applies best to organisations that are both competitively and IT mature. Not all organisations are. And especially for these, in order to fully take advantage of “data science”, an organisation has to go through some transformation, a change in mindsets, in tools used, and organisation structure.




The matrix above shows shows 4 situations depending on whether an organisation is above or below average in IT and competitive maturity.

Note that these measures are not absolute, but relative to the organisation and the market it plays in. There is no one size fits all, especially in terms of IT maturity; not everyone needs hadoop, not everyone needs real time monitoring, not everyone needs image recognition.

Ostrich





Some organisations are laggards both in terms of their competitiveness and their IT maturity. Frankly these organisations are in trouble, and simply employing a “data scientist” will not help. The organisation has to be transformed.

Wolf




On the opposite end of the spectrum, organisations that have IT maturity and use it to be competitive are the best set-up. These are organisations that are agile and able to maximise the advantages of “data science”, they probably already have a team of “data scientists” who work collaboratively with the business and IT and are kept really busy influencing the business on a daily basis.

It is not that easy for an organisation to take a big bang approach to transformation, form an ostrich with its head in the sand to an agile and hungry wolf.

Most of the time, either the business will be competitive and push the IT to get support (transform via competitiveness first), or the CIO/IT management will try to create competitiveness by pushing for the latest technology, hoping to make the business change.

Monkey






In the last few months, I have seen many cases of the latter. In many organisations today, “data scientists” belong to the IT group and report eventually into the CIO. I believe that this approach is likely to end in extremely underwhelming results, if not outright failure. Some of the reasons I have discussed earlier [2][3].

The organisation simply does not understand what “data science”/analytics can and should be doing for them. This type of organisation does not need a full time “data scientist”.

What they actually require is people with the ability to convince the business and prove to them how to exploit data and improve results. 

Early last year, a bank went back to the vendor who helped implement some of the latest IT equipment “I have followed your advice for the last couple of years and have the best implementation of your solution worldwide, where are the returns on my investment?”

Basically the implementation was taken as a pure IT project assuming the value if the data would be obvious. Well, the business was running fine without the latest developments, so why should they change what they are successful at? Nobody made a compelling case to the business of the benefits of the latest capabilities provided to them by the IT group. If you build it, they may not come if they are quite happy with the games at their local ball park.

I was involved in a case of an organisation that used to have huge independent IT budgets approved and the IT worked relatively independently of the business. It was quite amusing. The IT had, on its own initiative, purchased an algorithmic trading platform and repurposed it into an engine for the organisation to react in real time to customer behaviour. The problem was nobody from the business side knew about it or how to make use of it. And this was a few years ago, when the idea of real time offers was in its infancy. Some group IT had a vision, but just couldn’t “sell” it to the business or even to the other groups in IT who owned various pieces of data.

What the organisation needed was not a “data scientist” embedded in the IT department (they had and still have talented technical people), but rather someone familiar with both the business and the IT department to bridge the gap. 

The first step is to work with the business to build use cases and implement experiments to prove the value that the use of technology and data can bring. 

In such cases, more often than not, the business is not geared to run experiments. Most of the time, from conception to execution, campaigns in these organisations take weeks, if not months. And what would a “data scientist” do while the organisation has reached its limit of concurrent campaigns/experiments? 

Rather than a full time “data scientist” (especially chosen by and embedded in the IT department), these organisations first require “data scientists” with good domain expertise, to create a roadmap of relevant use cases in collaboration with the business, rapidly build analytical/”big data” prototypes and educate the business analysts to own and refine them, run the experiments and track the results, to obtain measurable success with the technology from the IT department. These “data scientists” do not have to be “full time” employees. The organisation is not ready for using this expertise yet.

These organisations lack the ability to gain a competitive edge out of their IT maturity, and, at the start, it makes much more sense to engage with consultants to bridge the gap while training internal staff.

Once the business is convinced of the usefulness of IT and the collaborative environment has been established, then the organisation can transform itself into an above average in IT maturity and competitiveness. This may then require the inputs of consultants in the field of “data strategy”, who can ensure that the IT and business work in concert.

But these guys are even more needed when the IT is lagging behind.

Komodo Dragon



Many organisations today are very competitive based on the knowhow of their people, organisational knowhow and culture. Whatever CIOs and geeks might tell you, you can still do very well with yesterday’s technology.

These companies however may find it more and more difficult to respond quickly to changes in the market place, let alone changes in the regulatory environment (Basel II, Basel III, Solvency II... in the Financial Industry for example).

Here too, a full time “data scientist” is likely not to be fully utilised.

The type of skills needed to improve the IT maturity of these organisations depends on whether the business recognises the importance of investing in IT or not.

If they don’t, then the first step has to be to exploit the current IT infrastructure and show how value can be squeezed out of it. As in the previous case, this would involve a “data scientist” with domain knowledge to rapidly build prototypes and prove use cases purely based on the current infrastructure, but also to sketch what else can be done. 

I was once engaged with a very competitive organisation that was using very old technology in the insurance industry.

While the management had decided to scan their paper forms to speed up the processing and make storage easier, they did not see any value in investing in anything more than capturing pictures of the forms. Underwriters looked at the pictures and did what was required, and the picture went into storage. 

On the other hand they had the largest market share and had good cross-sell into their base using traditional methods of their agents.

So in order to convince them of the value of the information in the pictures, we had to make use of what information they did capture, generate experiments with more progressive agencies, and prove the value of data. Only then did the management consider capturing more data at source. 

And before these changes were made, there was simply not enough to keep a full time “data scientist” utilised and generating sufficient RoI. 

But if the organisation does see the need to improve the IT side of things, or has been convinced to, then what is needed is proper advice on how to grow the IT capabilities and increase business returns in tandem. 

The skill-sets required to achieve such a balanced strategy are very unlikely to be found in a single individual. I believe a team effort is required; people with “data science” skills will be required, but also people from data engineering, IT architecture (and geeks), and people with consultancy experience.

For example, a typical range of activities would include uncovering use cases by consulting with the business, understanding the data that is required (including the frequency and how it should be kept and readied for use – do you really need real time capabilities or does a slight delay not matter much, or even an end of day update – mapping out the journey forward and knowing the most appropriate technologies required. The idea is to transform the picture so as to leverage the data in line with the business goals; technology enabling business goals attained with the help of data science embedded into operations.

A nice introduction to this type of transformational activity can be found in this relatively old article by McKinsey [4] or this more recent one by O’Reilly "when building an enterprise data strategy, consider ‘why’ before ‘how’"[5]

A good enterprise data strategy helps develop the IT maturity and the business results at the same time. Hence it can be used by organisations who are below average in both IT maturity and competitiveness to leap into the top quadrant.

Ostriches revisited, they can run fast, and jump




One of the great things about the technological changes in the last few years is that they have reduced barriers to entry on many industries. Hence you hear of disruptors entering and transforming whole industries.

(It is very important to realise that these changes affect organisations of all sorts of sizes. Not everyone needs to adopt heavy stuff like hadoop, sometimes moving to cloud can be cost effective without having to engage with the elephants. There is no “one size fits all”, different situations will call for different solutions.)

Nobody is stopping enterprises from disrupting themselves, and the most likely to have the incentive to do this are precisely those who are lagging behind.

For these organisations, it is even more critical to get the team of people who will help develop the data strategy right. Again, at least one “data scientist” would have to be involved, but it is very unlikely that the transformation needs one on a full time basis, but only as and when required.
So should no one put data scientists on the payroll?

Not really; I was being dramatic. Basically most organisations, at the moment, are not ready to keep a “data scientist” full time engaged in creating value. An organisation that fully utilises “data scientists” has to be agile and designed as such. Hence old habits and processes, will have to have changed so that multiple experiments are carried simultaneously and their impact studied. Some will be learnings, some will be successes that should be expanded, replicated and monitored.

“Data science” can be a competitive edge for many organisations, and to the extent that certain practices are often peculiar to organisations, it makes sense to employ a team of full time “data scientists”. But before reaching that stage, an organisation has to transform itself and put data at the centre of what it does.

Not many organisations are willing to do that. Yet.

Hence I stick to my guns, not many organisations should be hiring “full time” data scientists.



[1]https://hbr.org/2016/12/most-industries-are-nowhere-close-to-realizing-the-potential-of-analytics
[2]http://thegatesofbabylon.blogspot.com/2016/12/why-returns-on-data-science-have-been.html
[3]http://thegatesofbabylon.blogspot.com/2016/12/why-returns-on-data-science-have-been_22.html
[5]https://www.oreilly.com/ideas/when-building-an-enterprise-data-strategy-consider-why-before-how




No comments:

Post a Comment