The Broken World of Marketing Data Sciences

Do you know why the world of marketing has a response rate of 2-3% and is happy with it? Even lesser? It is because the world of marketing data and marketing data sciences is broken. And marketers don’t know a way around it.

Here’s what the data looks like typically-

• Market Research & Panel data – For decades, marketers have been happy dealing with panel data that fulfills what they believe in. Like the huge TG for their product and innovation. So much so, that it takes years of business disasters in a row for a Nokia to understand how things have changed. But Panel Data is at best a coarse approximation. The sampling is flawed 9 out of 10 times. Flawed. I am not even talking about biases. Or, fraudulent data entry.
• Focus Groups – Less said the better. The FGDs are not representative (by design), the insights are more often than not an accident, and there is a serious dearth of good marketers capable of running half-decent and probing focus groups. What you can usually depend on here, though, is something which is unanimously panned or applauded. Like the Kala Bandar (Black Monkey) of Delhi 6 (the movie).
• Test & Control Groups – Several practitioners have lost their voices screaming about the need for quality T&CG setup. The only real deal here, to some extent, was the world of database marketing. Unfortunately, DBMs believed a little too much in their potential, and focused less on their improvement areas, a problem that often accompanies success.
• POS/Warehouse Data – Logged in/ loyalty customers is what we are capable of tracking, they’d tell you. And can they track you across your properties? Their properties? Households? Don’t they have the data already? And what about the other data elements? How good are they? What about that graph model thing?
• Big Data – Lets mine social media, twitter sentiments, facebook posts, and let us use key opinion leaders and influencers, and viral content. And we will have machine learning for real time recommendations and offers. And… I know. I believe in the possibilities. But do you really understand it? Where does it even begin?

What this has, sequentially, led to, is –

• Lack of trust in the data – an obvious side effect, won’t you think?
• Rise of Witch Doctors –People with creative halos, learned expressions, snob behavior, and/or a “heightened sense” of what the consumers want. The primary reason why you’d trust them – because they tell you so. And, because you have no alternate version.
• A Kodak Moment – All traditional data providers are getting destroyed by Google, Facebook, Inmobi and the likes. These companies have brought “attribution” and measurement to the center of the discussion.

There has been, therefore, a dearth of high quality and effective use of marketing data. Take for instance, Amazon. It’s a data-science influenced company. Yet, what you get to see on the landing page tells you how broken the listings/ recommendation algorithms are. And they’ve been playing with it for a while. Flipkart fares much worse. The others are not even talking about it or worth talking about. Offline retailers have little good data. Brand managers work with incorrect data most of the time (or trade data). Yet, conversations about analytics and big data makes for good headlines.

Nevertheless, there are few philosophical departures, none too easy, to keep pressing forward.

• Invest in data – Most companies need a multichannel data play, common denominators across platforms, and investments in “data collection”. It’s long, tough and they don’t have the right focus or resources to do it.
• Improve data quality – Data needs to become a constant companion in every discussion. Across HR, Finance, Legal, Admin, Marketing, Sales, Strategy. Any investment in that direction would, today, mean creating hundreds of cross-walks, resolving inconsistencies, requiring layers of data cleansing, and finally, standardizing the definitions. It would be, for most companies, a journey of at least 12 months if they are smart, ambitious and ruthless. Ruthlessness is the key.
• Data in the front office – One recurring funny moment in every consultants’ and analytics professionals’ life is walking into a cross-functional meeting, with at least 40% of the attendees reacting at some point– “I’ve never seen this data before” OR “this data is wrong, because I checked last week and it was something else”. That’s because data and analytics sits in the back-office, and reacts to distributed demands and definitions. If it were in the front office, like revenue and profits, it would be reported correctly 8 out of 10 times.
• CEO’s Mandate – All professionals, in time, will have basic understanding of data, like they (should) have of English (as a proxy for language of the business) and Mathematics. Till then, however, it needs to be driven like a CEO’s mandate. Will that happen?
• Data Monetization – All of this pain becomes easier if start monetizing the data itself. Data and Analytics are a cost center. They can be a P&L. Especially, if companies stop being scared of how priceless their data is. It is not. The only thing that makes your data priceless is how it is used. Governance around the usage of data might help companies more than sitting on it and hatching it.

You know one thing that would really help? Knowing when you need help and being candid about it! And – I’ll not get into the psychological reasons (I know better than data), infrastructural reason (our technology and analytics team is not ready for it) or business reasons (it is not a priority at this point) why marketers don’t ask for help.

<First published on linkedin>

 

Advertisement

Are you satisfied with your analytics training?

Lets assume that you’re in a third party analytics services/ consulting organization. I assume that your organization started with a core of like minded people with the relevant analytics credentials. And I would like to assume one of the following two things – that you’ve grown somewhat, or are planning to scale soon.

Source: The Cartoon Guide to Statistics (from Avinash Kaushik’s blog)

You must have hired or are planning to hire fresh talent. Recent undergrads and graduates. When you hire them, do you put them on a project on day 1? If not, do you train them? What is the nature of this training? The duration? What did you assume about their knowledge base when you started training them? Do you conduct the training internally using consultants’ time? Or, do you bring in trainers? What about the continuing education – understanding Digital in 2010, for instance?

Now, some more assumptions. You assumed that the quants have the knowledge or appreciation of strong technical and statistical principles behind analytic problems. You also assumed that the engineers are good programmers and problem solvers. And lastly, you assumed that the MBAs will be able to manage projects, learn enough technical stuff, support pre-sales and sales activities, etc. Somewhere, you also assumed that each of these groups will acquire the skill-sets of the other groups.

Pan the camera towards India analytics talent now. There are three distinct talent sources right now. Quants (Masters in Statistics/ Economics/ Computational Mathematics), Engineers/ Problem Solvers (just smart B.Techs), and Managers (MBAs). Think about the questions  for these three groups separately.

More questions now.

• Did you train them on a mix of SAS/R, Excel, Powerpoint, Access, SQL? Something about messaging, charts, dashboards?
• Did you spend 2-4+ weeks training these smart professionals?
• After you trained them, were you satisfied (completely) with where they are? Were they really ready to become independent analytics professionals?
• Did you question your training methods? Assuming, they are not delivering over and over?
• Did you ever question the foundational assumptions made? That a quant is not really appreciative of what goes into all this? Or, a manager who is not quant enough to start with, can never really manage?

Here is what I am thinking –

1. Can the same level of satisfaction with training be achieved in just 3-5 days? Can the remaining training duration be used for cutting edge?
2. What if you have two different sets of training? One that equips you for delivering projects, and another for turning you into a data scientist? Is there a preference for one over the other?

Source: http://www.aepconvert.com/web-analytics-training.php

Chances are that you are too busy or have too many things going on. Based on several senior folks in the industry that I have met recerntly, and over the years, this is an industry wide problem. An analyst with three years into the industry may not know anything beyond basic programming methods – not analytic thinking, not methodology depth, and not the ability to have an engaging conversation. By itself, it may seem OK, since you’re focusing on the core programming skills in the earlier years. However, what if you’ve damaged the raw material that you started with? That out of 100 people that you hired, in the hope of creating those data scientists, only about 1 or 2 really set on the path of becoming one?  What I fear for is the quintessential mediocrity that we bring to a particular field of work. IT is a great example of how we’ve treated talent. Politics is another one.

Like in almost everything else, the dependence on the aptitude of the new hire, and their given interest in the field of work is what everyone is depending on. It’s the model that has been used by strategy consulting. And by being so myopic towards project delivery, the industry is missing the knowledge component of this knowledge economy. The recent curriculum innovation in certain universities, such as NYU Stern, Northwestern, UNC, etc. gives me hope. India needs to move on that track too.

P.S.  a question – have you every wondered why the core team’s talent/ skillset is a lot higher/superior than the people you hire when you’re scaling? Is it OK to be like that?

Note: The reason I did not talk about the in-house analytics teams – they rarely have a culture of training their analytics professionals. They usually just deploy them. While I am a fan of BYOL (bring your own learning), I don’t think its applicable everywhere.

Data Democratization To Analytics Democratization

One of the more recent buzzword in the world of analytics is data democratization, used in the context of public data, govt. data or just organizational data. In layman terms, it means that everyone in the organization should have access to consistent (and preferably correct and comprehensive) information presented to them in easily consumable formats.

I am a huge fan of the possibilities that data democratization would lead to. However, the real tipping point of the world moving to an analytics led marketplace is analytics democratization, and not just data democratization.

Questions before I go further –

• Do you think it takes a lot of time for your team to get a particular report/analysis/data slice that you need for, say, an urgent meeting/review?
• Do you think that there are far too many people  involved in this process of getting the data to you? DW/ BI/ Analytics/ Product/ BU/…?
• If you are an analyst, do you think you get too many ad-hoc requests in a day? Are some of these what you’d call a re-hash of a previous request? And are you scared of telling the requester that you’ve already given it to them albeit in a different format/ it can be derived from the earlier data dump?
• If you are a business manager -Do you always get that feeling that you do not have enough information? Do you feel that you need something more? Do you also get frustrated with how long it takes to get the analysis that should have been  a cakewalk? Do you also feel that the analyst has just dumped a whole lot of data without stopping to think about the insights?

If your answer to most of the questions is yes – then your organization is most likely miles away from reaping the benefits of data democratization.

Barrier to adoption

Merely by giving access to data, access and consumption of data cannot be guaranteed. For instance, open source and freely available Linux got the appreciation of a lot of people, but remained the geek’s preferred OS. Masses stuck to their user friendly iOS and Windows systems. Even when Linux became more user friendly, the half-baked OS democratization had already created more barriers for the commoners. And the whole idea of migrating from their existing “warranty led OS” to an open source OS is something that masses (and democracy) were not ready for.

The barrier is not data, but understanding

Most stakeholders who do not advocate data driven thinking, and hence analytics, are the ones who have a limited view of what analytics is. Providing data and reports to them never solves the issue of analytics adoption. A closer look at the real leaders in this industry, e.g. CapOne, P&G, Amazon, Walmart, etc. would tell you that “analytics thinking” is ingrained in the way these organizations function and think. In these organizations, analytics has been democratized.

Data by itself is only half the value

Data is almost like a blunt weapon that can be used to club the enemy, or carved into a knife for the kitchen, or a sword for the battlefield. It can also be converted into a home equipment like needle or farm equipment like spade. Without questioning the merit of democratizing data, the real value of data can only be unleashed by making customizable adoption formats. Most end users would still look for flexible thin applications that help them understand data as per their need, or provide enough reports/dashboards that makes the need to be creative redundant. The latter will lead to the generation and maintenance of hundreds of reports that never get used, or whose existence is never ROI justified.

A report is as the reporter does

The number of dashboards and reports that I have seen in my career is an insignificant number. The significant aspect though is the “narrow’ application of most of these reports. Or, the demand-supply gap. Demand of the user, and supply from the report designer. Also, a lot of these reports are so rigid that a change in them requires raising a service request of sorts that goes through three layers of approvals and few weeks of analyst time. The number of consulting engagements that get delivered on the basis of inconsistent reports that different parts of the organizations use is a different story altogether. I believe that once analytics thinking is democratized, the products/reports/dashboards will start evolving to the needs of the masses, have flexibility to adapt on tap. And that’d be an interesting place!

****

So, even if you’ve democratized data sufficiently, you need people to understand the tools and skills that enable their conversation with data. I have seen too many “smart” people get scared when you throw too much data at them at a high speed. And whoever tells you that you cannot train the whole world what multicollinearity and mixed models are is likely trying to fool you. They need to tell you that you don’t need to.

Let’s aim for a culture where a quant(?) can communicate in a language that business can understand and the business can communicate in a language the quants can understand. And to that extent, organizations need to invest not just in building Big and Bigger Data capabilities, but also an analytics thinking mindset second.

And that’s an area where I see a significant gap. There are few people capable of educating analytics teams, and fewer capable of educating executives.

Image Credits:
1. Infolytics, here

2. Here

3. Here

The Great Analyst Series– The 9-pointers

What makes a great analyst? It’s a question that plagues every manager and every analyst who’s working in third party analytics organizations/teams. Less so in the case of captive teams, because the model works more on the lines of individual contributors rather than large managed teams, hence dramatically modifying the equations. Though some of these apply universally.

The adjectives I have heard fall in these categories –

1. Keep it error free. The first and the biggest. This one’s a killer. An analyst falls from being great to not so great in that one short span where a client or the manager’s detects an error. Especially, if its a client. Akin it to your confidence in betting your salary on your analysis. An interesting practice would be – Keep aside 100 bucks for every time an error is found on a file or an email that leaves your inbox/machine. Even if you did not create the file. Two errors is 200 bucks. Even if they are on the same email/file attachment. At the end of the week/month, see how much you’ve put in the kitty. Donate that money to a charity. Or, if you’re not keen on social stuff, then order sandwiches and samosas for the office. If you’re a good tracker, you might see yourself make more mistakes at 5:30 in the evening to start with (because you sent out an analysis too close to a client call or to the end of the day).
2. Plan and communicate. Managers struggle with the effort estimation done by their analysts. They’re never too sure if the buffer they’ve budgeted for is appropriate. So, it’s a trial and error process. You can make it easier for them and for yourself by planning, adding buffer, communicating the plan to the manager, and if you see a risk, communicating upfront about not just the delay, but also the reason for it and the planned remediation. It’s good to know that you’ve got it in control, even if there is a problem. My love for analysts with a solid personal project management method might be a bias, but I have always been wary of geniuses who do everything in their head. Maybe because I am not smart enough to reside in their head. But if I had to take a bet, I’d bet on most managers being as dumb as me.
3. Engage. An analyst that takes a task as given, and comes back with the output against that task is not a bad analyst. After all, the baseline expectations are met. But, when you’re discussing the top tier analysts, you’re focusing on the “exceeds expectations” kinds. And they engage more. With the engagement, with the team, with the problem and with the analysis. They come back with a checklist of having delivered on the task, and some more, and a few questions or ideas to develop it further. Sometimes, they come back with questions and ideas about why the earlier discussions/paths were not the best.
4. Question, but with a pause. Sometimes, you come across a glaring “situation”. And you want to question it. And you question it rightly. No problems. The point where it becomes a potential problem is when you question it too long. Sometimes, and rather most of the times, it’s a smarter move to demonstrate the problem you have with something tangible. So, if the analysis is flawed and the manager ain’t gettin’ it, then go ahead and do it the manager’s way first, show the reconciliation/ triangulation errors, propose an alternate, do it the alternate way, and show why your method is better. It’s a much easier conversation than telling the manager that they’ve got a few bolts missing. And you make your point, earn your brownie points (assuming you don’t offend the manager with your in-the-moment superiority complex). What definitely never works out (I have seen it true for at least some of the analysts I’ve worked with) is – I told you so!
5. Be visually delightful. Don’t dance, please! An analyst that transcends the analysis and can communicate it effectively always gets more brownie points. So, if you have two analysts where one is a better thinker, and the other is a better presenter (assuming baseline attributes are not at risk), the chances of the better presented succeeding more often in their professional career is higher. And when I say communication – it’s all forms – the written, the oral and the visual. And the visual has two components again – the visual aesthetics of the output, and your own visual aesthetics. A shabby looking person with some irritable personal habits can find it difficult to be role model, despite their technical brilliance. I feel flimsy saying this, but sadly, this is true.
6. Adapt. Be flexible. Managerial styles are different. There are micro-managers and there are delegators. And there are indifferents. I know it’s a lot to expect the youngest one in the team to be the most mature, but never hurts to know, right? So, monitor the manager carefully in the first couple of days of an engagement, and alter your style to have the most fruitful relationship. The best of the managers can teach you a lot. The worst of the managers can teach you how they became a manager in the first place (which might be a career goal of sorts)!
7. Provide leverage! A good analyst provides good leverage to his/her manager. In common language, I translate it to the ability to preempt and take up several mundane tasks that your manager is spending time on, without compromising on the quality of the work you’re expected to be doing. Doing the manager’s job so that the manager can focus on their next set of priorities. It’s an attitude that I cannot over-emphasize the importance of.
8. Don’t just earn your salary. Not everyday. When you look around, you will notice them. They are prone to saying – this much money means this much work, or “I can work harder if there is more money to be made”, or “whats the point, I am not going to make more because of that”, or “today was a productive day and I managed to get my work done”, or some version thereof. They think that “earning your salary” equivalents to being meaningfully busy for 8-9 hours a day. When you’re just trying to be meaningfully busy, you forget trying to be the best you can be. Make the system feel bad about paying you only your salary. Make the system think about how best it can value you. By better projects, better career, better opportunities, and in the process, hopefully, better salary.
9. Read. A. Lot. I have experientially seen it to be true. I am not sure other managers are aligned with me on this one. In my view, a well-read analyst always finds more dimensions to analyze in a project than a walk-in-walk-out analyst. The well read could nerdy (e.g. reading about the cutting edge work like Watson or In-memory databases) or business savvy (e.g. mobile payments and NFCs) or socially aware (literary works that provide insight into how people and societies think and act). But it’s a strong hypotheses in my experiential learning that a good reader usually becomes a good thinker.

 

Why a 9 pointer? Because a perfect ten should be all this customized to your own inimitable style, as and when you find it.

The Pit, The Fall and The Reichenbach Retreival

The world is abuzz with words like Big Data, Cloud, Efficiency, Real-Time, Analytics, Power, Complex, etc. Most companies are being implored to think about these words. By organizations that are building “capabilities” around it.

Somehow, I think, we are missing the point.

The problem can never be smaller than the tool that solves it. And that to me, has been the bane of the world of analytics/big data/ <fit the new key word> professionals. They create a world where the tool becomes bigger than the end goal it serves. Right from poorly implemented data warehouse solutions, to point-in-time  dashboards that answer a question relevant 5 years back. Do you need to solve your “Big Data” problem? Or, do you need to solve for your business issues?

• Nevertheless, over close to a decade of trying to be an analytics professional, I have seen more examples of organizations trying to latch on to a fad rather than focus on what they have. Social media is just an example.

You’re walking towards a pit, and you should do it only if you’re fond of them and the treks and the views they offer. Not to fall.

If you are just about venturing into the web world, a free plugin of google analytics can reveal a lot to you, to get you started. Instead of a million dollar investment in sophisticated tools and dashboards. Secondly, a good looking dashboard does not always reveal something additional. It reveals in a palatable way. If you’re focused on what you want, that is, if you’ve figured out your business problem, you don’t always need that sexy solution.

Reminds me of that debate we had about two attractive girls – the difference between the two was that the first was sexier, but the second more marriageable. The second had more suitors while the first evoked desire a lot more often.

• A lot of these giant ideas fail. You do fall in those pits. Investments that don’t seem to be worth it. Models which stop being predictive unless fresh blood is pumped into them at regular intervals. Technology that becomes obsolete faster than your ability to eat French fries. You will make mistakes, and I guess that’s not that bad an idea, but the least you need to do is be aware of the costs.

For instance, internet was the in-thing. It still is. But the world is already looking at mobile as the next big thing. Right from iOS to Android enabled devices to extremely interesting content delivered in real-time. For every dotcom that succeeded, there are many that got bust because the fundamental idea itself was not thought through.

Microsoft Excel could handle 65k rows of data with great difficulty at some point. Today, it can handle a million rows with lesser amount of difficulty! And people have started talking about billions of data points. However, a lot of beautiful insights are driven sometimes out one unbiased and strong validated hypotheses. Which often, smartly done, does not take a million rows.

• Last point – it’s difficult to recover from a large investment gone wrong. And I do not always mean monetary investment. Most firms fail to look at executive time investment on half-baked investments as a loss. The opportunity cost of such falls may be significant.

EndNote: The environment today has given us more power in our own hands than most of us are capable of handling. It overwhelms us. We run hither tither and grope around to latch on to the keywords that supposedly enlightened souls are spewing at venomous speeds. No-one is wrong. Yet, the right question for you needs to be the same as it has always been – Is this what the business needs? No. Not wants!

In the Indian mythology, the tale of Bhasmasur is fairly popular. Bhasmasur was a demon who practiced austerity of several years in extreme conditions to please the gods. Lord Shiva appeared in front of him and agreed to grant him a boon in return for his perseverance and dedication. Bhasmasur requested for a power that allowed him to burn anything down to ashes (bhasma) that he’d put his hands on. As the powers, and hence, the tyranny of Bhasmasur reached its pinnacle, the other Gods implored Lord Vishnu to save them. Vishnu took the form of a beautiful danseuse (Mohini Avatar), and tricked Bhasmasur into copying his (or her) dance steps. However, as Mohini put her hand on her own  head while dancing, so did Bhasmasur on his head. And thus, Bhasmasur burnt himself down.  Bhasmasur forgot the reason why he aspired for that powerful hand. The hand that burnt him down.