Interview: Michael Brand (Founder, Otzma Analytics)
About today’s guest
In today’s interview I talk with Michael Brand who is a data scientist and the founder of Otzma Analytics. A few weeks ago he released a video (below) showcasing a data scientist view of the Covid-19 pandemic and it was quite interesting to see his take on the current crisis.
Video link: https://www.youtube.com/watch?v=BMzhdoQABFQ
There’s a good set of lessons about understanding data that we can unpack from this video but what better way to discuss this and data science in general than to interview the man directly.
The Interview
Mark: Hi Michael, thanks for giving up your time for this interview.
Michael: Hi, Mark, and thanks for inviting me to this Q&A. I’m hoping this will clarify to the readers both a little about what’s going on with Covid-19 and a little about the kind of data analysis I pursue and advocate for.
Before we start, a quick disclaimer: I am not a medical doctor, I am not an epidemiologist, and in all things Covid-19 the official health sources should be your guides. This is true always, but especially in a case like the Covid-19 pandemic where the situation changes daily, as does our understanding of it.
Mark: Understood. Before we dive in deeper, could you tell us about yourself and your company Otzma Analytics?
Mark: I’ve been a data scientist for almost 30 years now, with my latest corporate job being Chief Data Scientist at Telstra. I’ve done data science and seen data science being done in three continents, and can tell you that despite all the buzz and the hype around analytics, there are only a handful of organisations in the world who have really cracked the formula for how to do it right and reap from analytics true strategic benefits. Otzma Analytics is a boutique data consultancy that I started a year ago with the aim of helping organisations get on that magic road.
I do this with a three-pronged approach. To assist the executive level, I provide training on how to tackle analytics: what its value chain is, what managing it looks like, what it can do for you, and what you need to be wary of. To help data scientists, I work with clients’ existing DS teams, pivoting them to more business-strategic directions and injecting better state-of-the-art techniques into their work. Finally, to help the success of analytics projects, I provide “analytics audits”. These are model reviews where I stress-test an analysis, pinpoint where its weak links are, what can be done about them, and how the project value can, overall, be increased.
If you want to move up the analytics maturity ladder, these are the tools that will get you there.
Mark: Sounds like a good wealth of experience there. So, how does Covid-19 affect your day to day work?
Michae: Surprisingly little! I’ve worked for 8 years remotely, living here, in Australia, while all my work and its clients were in other continents. This, for me, is a return to an old stomping ground. Over the years, many of my co-workers, too, were remote-workers, working from home or other places. We are lucky to be in a fully-digitised industry, where this kind of work mode is possible and increasingly accepted.
I mean, sure, I had many talks and public appearances cancelled because of Covid-19, and conferences I meant to attend are not taking place, but these will all only be delayed by a few months, and in the meantime I reach out using YouTube videos. So, Covid-19 has barely slowed Otzma Analytics down.
Where Covid-19 has affected my day-to-day work most is, perhaps, in two ways. First, there’s a massive leap when it’s not just one or two or three people working from home but an entire office, and every single function needs to figure out how to do its work remotely: VPN lines are stressed, coordination is challenging, and everybody needs to adapt a little. Second, I suddenly find myself working with many people for whom this is now a first time working from home, many struggling with similar challenges to those I had when I first tried it. I’m really glad to have the opportunity to give people a little from my lessons-learnt on this subject, and help them make a smoother transition.
Mark: I hear you there. It’s certainly taken a moment to adjust for some. When did you start looking at the Covid-19 data and decide to analyse it and why?
Michael: For me, there’s no such thing as “deciding to analyse”. Analysing is just my word for ‘thinking’. I can’t read a news article without thinking about the data trace and what it means regarding future decisions.
At the same time, analytics audits are something I’ve been doing for over 10 years now, long before I founded Otzma Analytics, and in the process I’ve developed a keen ear for differences between what the data bears out and what the data is said to bear out. When I read a newspaper headline that tells me that “The epicentre of the Covid-19 pandemic has now moved to X”, I immediately start asking myself a whole slew of follow-up questions: what makes something into an epicentre? Is it because many cases were discovered there? Is it because of deaths? What about all of the ones that haven’t been discovered? What about differences in population densities, lifestyles, etc. that will impact how these numbers are likely to progress into the future? What role does past data — as opposed to present data — play in determining what the pandemic’s true situation is?
There was no decision point, for me, to analyse the data. The only decision point was to make a video about it, and that happened after many conversations with friends and co-workers, where I realised how wide the gulf was between what other people were seeing (often from reading news written by people who are neither medical professionals nor data professionals) to what I was seeing. Some of these conversations made their way to the video, like the question comparing Covid-19 to the flu. Others will have to wait for another video.
Mark: I look forward to those future videos. Could you tell me a bit more about your style of analysis and your process for coming up with your model? E.g. do you have a particular focus area?
Michael: In almost thirty years of doing data science, I was lucky to cover a whole lot of ground. I worked in speech analysis, in machine vision (having been one of the developers of the XBox’s Kinect sensor), in customer prediction (e.g., demand, churn, marketing), in safety analysis… Why, just during my time at Pivotal I was one of the leads in a data science team that dealt with everything from finance to cybersecurity to health analytics to geological models. To me, the beauty of data analytics is that the math is always the same math. It doesn’t matter if you’re counting tomatoes or website hits.
My one area of expertise is the research discipline itself. I am a student of the scientific method, of experimental design, of causal analysis, of Bayesian modelling. Though this was never by choice or design, it turns out that this makes my style of analysis rather unique. I never go into a problem with a definite plan on how I’m going to tackle it. I let the specifics of the problem and the knowledge trawled from the underlying data guide my work. As a result, my solutions are always unique, always in complete alignment to the needs of the project at hand.
There’s a lot of noise, in recent years, from vendors selling off-the-shelf analytics, promising that all you need is to install their product and all your problems will be gone. I don’t subscribe to that style of thinking, and in my YouTube video series “How to Lose Money on Analytics” I show how there is nothing under the hood in these promises. If you want a solution that works for your problem, you need to understand what makes your problem unique.
Mark: Do you see mistakes in how people are analysing Covid-19 data in the news media and general public?
Michael: I don’t know if I would necessarily use the word “mistakes”. When I do an analytics audit it’s not an exercise in ‘gotcha’s. When you look at data — whether you’re a lay person, a subject matter expert or a data professional — it’s always a combination of three things: there are your prior assumptions, there’s the data you gathered, and there’s the logic combining these and leading up to your conclusions. In analytics audits, I check all these things: are your prior assumptions explicit and validated? Is your data fit-for-purpose? Does your logic contain any gaps? Are all your convenience approximations merited?
In the case of Covid-19, a wrong decision can cost lives and ruin economies. This calls for an extra level of robustness in our thinking. Instead, what I’m seeing is people being fed distorted images, missing crucial data, and being driven either into fear or denial. I wanted to be able to have a cool, fact-based discussion on the topic.
Mark: What are some of the key messages you want to convey from your post “A Data Scientist looks at Covid-19”?
Michael: When I started this video, I was thinking of messages that are, to me, timeless ones. These are messages about the importance of rigour in analysis, about how scientific thinking isn’t a thing for geeks in lab coats but rather a general-purpose tool for understanding the world and making better decisions, and about the importance of critical thinking.
Obviously, most of my viewers care more about my specific conclusions regarding Covid-19 than any of these timeless remarks. Regarding Covid-19, I show three main things.
First, I delve into some bad — and maybe wilfully bad — data presentations of the present crisis, that have led many people for much too long to dismiss the crisis or to downplay the need for extreme action to combat it. Using the same tools I use in audits, I demonstrate the faults in their apparent logic.
Second, I demonstrate where scientific thinking leads us by looking at the invisible problem of the untested asymptomatic carriers, and showing how correctly accounting for them changes our view of both how deadly Covid-19 is, and how widespread it has already become. I show, for example, that the testing that is needed to understand and combat the problem epidemiologically is not the same as the testing that is needed (and that is being done) to facilitate medical treatment of infected patients.
Third, I show the power of big data techniques both in helping us fight the virus while minimising the social and economic costs, and in speeding up vaccine development by measuring the virus’s interactions with other factors in the wild.
Science is a powerful tool in determining the best course of action quickly and confidently, even in situations where information is partial and uncertainty is high. If ever there was a time to make full use of it, this is it. I hope my video convinces more people that scientific outlook isn’t just for scientists, but a way of thinking that all of our lives will be better if we adopt it more.
Mark: You mention that we should take a data scientist approach to take a random sample test for Covid-19 to get a baseline assessment. Can you explain how that would be useful for Australia?
Michael: Our best bet in fighting Coronavirus is in prevention: we want less people to catch it in the first place, rather than rely on hospitals to treat the infected. As a result, it is critical to understand how Coronavirus spreads, where it has spread to already, which are the populations that are at risk, during what time window a person is contagious, etc.. Though these may seem like the trivial questions to ask about Covid-19, in fact we don’t know the answer to them, and estimates vary considerably. The reason for this uncertainty is that there are many with Covid-19 who don’t show symptoms. Some present with symptoms, but mild ones or atypical ones that do not get attributed to Coronavirus. Some will have symptoms, but do not present them yet. Some present with symptoms but choose not to get tested. And even among those who are symptomatic and are in hospitals, the present testing strategy misses many Coronavirus patients because they also have other illnesses: in order to save on Coronavirus tests, which are scarce, patients are first tested for other viruses and only if these tests are negative they are given Coronavirus testing.
As a result of all of this, epidemiologists get a highly skewed and unrepresentative view of the true distribution of Coronavirus and of Coronavirus symptoms in the population, and as a result all of our modelling and all of our decision-supporting analyses is likewise skewed.
A random population sample solves this, if it is taken in a population that has a significant enough number of infected individuals. Ironically, the only country I know of that has done a randomised Coronavirus test is South Korea, which is also the country whose tracing of Coronavirus infection paths has been the most extensive. However, what we’re seeing now in the U.S., perhaps due to a difference in the populations and perhaps due to the fact that the U.S. is predominantly under a different strain of Coronavirus than the one that spread in South Korea, is that the statistics from South Korea do not match what we are seeing in the U.S.. That one random population sample isn’t good enough. We need more of them.
Mark: What kind of tools do you use to analyse the information?
Michael: Analytics auditing is rarely about analysing the data. Usually, it’s about analysing the analysis. Luckily, the Internet is right now full of so many analyses of this data, some more raw, some quite involved. The main tools I use in auditing these analyses (and any analyses, really) are the mathematical tools of statistics and probability, and the scientific tools of experimental design. That — and a heavy dose of critical thinking.
It should be pointed out, however, that in all the analyses I looked at, the underlying data is in no way “Big Data”. Population censuses are very small data, even when the data is accumulated globally. There is no Covid-19 analysis that I have seen online that couldn’t have been produced using R or Python or any of the other standard, free, general purpose tools used by data scientists everywhere.
This is yet another bit of evidence that what we need is less obsession with tools and more on honing the correct style of thinking.
Mark: What kind of opportunities do you see will come from this crisis in terms of new ways of working or ways in which the world will be different?
Michael: Well, for one, it looks like all of us will now have to become a lot more comfortable with working from home. I hope that this will be a long-term mindset shift that will far outlast the present crisis. From my personal experience, once you realise how working from home means hours saved on commute, more time with the family, and if you’re really, really lucky also fresh, home-made lunches, why on earth would you want to go back?
There are many other lessons to be learnt, globally, from this crisis, but for these I am less optimistic that the world will heed the lesson. I’m not optimistic our post-truth culture will once again embrace the power of objective fact or of unbiased analysis. I’m even less sure, once the knife is off our throats, that we’ll remember to invest in medical infrastructure or in basic scientific research, medical and otherwise. For some reason, we keep expecting when a crisis hits to have science ready to bail us out, but refuse to spend in order to build up its capabilities at all other times.
Mark: What is Otzma doing to help its customers understand the current pandemic?
Michael: Otzma Analytics is lucky to be working with clients that are technologically mature and can switch to a work-from-home model with minimal pain. Almost by definition, because they’ve hired Otzma Analytics, these are also organisations that believe in data-driven decision-making, making them more likely to ingest the news with a critical eye.
The challenge, as I see it, is not specifically with my clients but rather with the population at large. This is why I decided to release my “A Data Scientist Looks at Covid-19” video, and which is why I made an effort to give a learned response to every question sent to me on the topic whether in YouTube comments or via e-mail. Ultimately, the purpose of data science is to help us see, despite uncertainties, where our actions will lead us. In this particular situation, the salient facts are clear, the actions are self-evident, and yet too many of us are listening to disinformation or succumbing to wilful ignorance.
Mark: What’s your outlook for Australia considering where we are now?
Michael: The data shows that Australia has reacted, in terms of number of Covid-19 cases, faster and more effectively than many other countries. This has always been one of the things I really love about Australia: when we’re backed into a corner, we’re willing to make significant changes in order to adapt and get us out of the rut. It’s a miraculous trait, and one that is undoubtedly going to save us again and again as we’re heading into an increasingly uncertain and turbulent future.
I believe we have this power also when it comes to data. Australia is better positioned, data-wise, to take advantage of the data science revolution than anywhere else in the world, Silicon Valley included. The question is only: will we realise in time that the revolution is now, or will we continue to wait on the sidelines, as we have so far been doing, until all the rest of the world will have passed us by?
With Otzma Analytics, with my videos, with my academic work and in many other ways, I am trying to give Australia the tools it needs to realise its potential in data. But only Australia can decide whether it wants to walk down that path.
Closing remarks and contact
What a great set of insights from Michael. To see more about what he does then check out the links in this article. To learn more about Otzma Analytics then check out their website (https://otzmaanalytics.com/) or follow him on LinkedIn (https://www.linkedin.com/in/michael-brand-b230736/).
Stay tuned for more interviews like this in future and if you would like to get in touch with me about this or other posts on this blog then feel free to reach out below:
Mark Monfort (Head of Data Analytics and Technology)
- Phone: 02 8262 8700
- Email: mmonfort@prosperity.com.au
