Gordon Irlam is a semi-retired software engineer with an interest in retirement planning. He has developed three websites that solve the retirement portfolio problem in three different ways: via Merton's portfolio problem, stochastic dynamic programming, and reinforcement learning. He is also an effective altruist with an interest in mitigating the risks of future advanced artificial intelligence.
The evergreen subject of retirement planning is something that we prioritize here at the Rational Reminder Podcast, and today we have a very interesting conversation in which we explore the topic from a slightly different perspective. We are joined by Gordon Irlam, who is a notable researcher with a wealth of experience from the world of tech and beyond. We have the chance to ask Gordon about bonds, annuities, and optimal allocations for different outlooks, and also get his perspective on charitable giving, effective altruism, and different spending plans. Gordon has conducted some amazing research and even developed his own tools to help investors calculate the variables of their situations. This episode is a great gateway for listeners to explore these concepts, as well as make use of Gordon's resources. Our guest's personal story is equally fascinating, after working with Google early on, and subsequently starting a company that was then acquired by Google, Gordon has leveraged his experience and finances in order to continue asking questions that interest him and will definitely interest our listeners. So for this standout conversation with a great mind, be sure to take a listen.
Key Points From This Episode:
Looking back on the Google equity that Gordon sold and how he feels about the decision now. [0:03:00.8]
Google’s acquisition of a company that Gordon started and the impact of this financial windfall. [0:04:33.1]
Gordon's explanation of effective altruism and how he utilizes the idea. [0:06:25.3]
Approaches to asset allocation for foundations and how this differs from personal funds. [0:10:28.7]
Comparing practitioner and economist approaches to financial planning. [0:15:59.7]
An explanation of stochastic dynamic programming and its strengths. [0:17:45.5]
Why Gordon now favors reinforcement learning over stochastic dynamic programming. [0:20:12.6]
Considering the role of annuities in Gordon's optimal model for retirement planning. [0:25:05.3]
Constant spending versus variable spending in the optimal retirement plan. [0:27:55.2]
Gordon's practical advice for entering retirement and tracking spending. [0:29:35.8]
Exploring mean reversion in stock returns for tactical planning. [0:32:16.6]
A message from Gordon about fixed guaranteed income and the value of long-duration inflation index bonds. [0:35:18.7]
Advice to younger individuals and investors; the importance of saving. [0:36:18.9]
Thoughts on possible future innovations for the problem of better portfolio building. [0:37:45.3]
Gordon's definition of success: the ability to work on interesting and important problems. [0:40:26.2]
Read The Transcript:
So Gordon, you worked at Google prior to Google's IPO and given your clearly data-driven approach that I have started to understand from doing research for this interview, I've got to ask, what did you do with your Google equity after the IPO?
I followed the recommendations of financial planners which was to diversify and sell all my equity as soon as it vested. It makes a lot of sense but there's also a lot of hard take associated with that, in that if I had held on to that equity of my employer, it would be worth, I don't know, $10 or $20 million today.
Wow. So looking back, how do you feel about the decision?
I have mixed feelings obviously. I still feel if it were like a rational thing, back when no one knew what Google was going to become.
It could've been the next Yahoo, and so I'm comfortable with that decision but it's like decisions in finance where it's easy to look back and say this is what you should've done but it's very hard to know what you should do at the moment.
So if you met someone today who currently is in a similar situation to what you were back then, instead of getting advice from a financial planner, what would your advice be to them?
I'd still say you should sell your equity. You don't want to know the concentration in your employer's funds.
That's definitely the rational answer, and we would tend to agree. Now your story just gets more and more interesting. After you left Google, and then you left some equity on the table then, they actually ended up acquiring a company that you had started prior to joining them for a pretty significant sum, that it was for $625 million cash, as I understand it. Did the windfall from that transaction change your life in any ways that you did not expect it to?
It didn't change it a whole lot. It made things more comfortable and secure outside. I previously had enough funds so I was able to fund most of my lifestyle but now, instead of having to worry about the stock market on a day-to-day basis, I could feel very confident and secure.
I'm curious if anything that you expected to change actually stayed the same after that event?
I'm not sure it did. I'd say however, one thing that it would be nice if it changed but didn't is the sense of when I was young, I had so many opportunities before me and now I'm at my current age, I've got a limited number of opportunities and I've got to be careful about choosing between them but since I got older, I'll only have one or two opportunities left that I can really pursue full time and so it would be nice if becoming wealthy could reduce the risks of that and give you more time in which to live but over wealth in the world won't change how much time you've got on this planet.
That's really interesting point. One of the things that you've done, a lot of writing and clearly thinking about is giving back after the Postini transaction where you sold the company you've started. You were able to start a foundation with those proceeds and the mission of the foundation on its website, which I love the careful choice of language, seeks to do the greatest expected good and so it's taking inspiration from the effective altruism movement. Can you describe what effective altruism is and how you're implementing it through your foundation?
Okay. Effective altruism's just for idea in our lives, we should seek to do the greatest good we can but it hasn't received that much attention in the historical past. Foundations were set up just ...they're allowed to do whatever they feel like rather than any... I think the financial world has a huge focus on trying to make as much profit as you can and that big companies focused on that but in the charitable world, there isn't the same focus on what are the returns you get from the investments you made. So effective altruism's a movement that started in the last, maybe 15 or 20 years, trying to think about what are the returns you get.
So an example would be something like bed nets for treating malaria and traditionally, foundations might have gone in and distribute bed nets and feel good about doing that but the effective altruism would say how much does a bed net cost? What the chance that that distribution of a bed net will prevent one case of malaria, even put a value on the value of a human life. From that, calculate for each dollar you spend on bed nets, you get $10 of social return or whatever the number might be and so that's what effective altruism movement seeks to do as compared to these different charitable causes to figure out which are the most effective ones.
That's so interesting. People watching this on YouTube could see I'm just thinking, there's so many questions that come to mind because so many people are passionate about certain charities based on perhaps their personal family experience but it sounds like, and I find this whole idea very appealing, it sounds like you're suggesting that people should focus more on the legacy, efficacy of the endeavor as opposed to the almost spiritual motivation behind it. Is that true?
Yes, it is. The effective altruism movement tend to identify if there's a number of cause areas that seem to have the biggest impact. One of them is global health because very cheaply, you can do something like vaccinate or provide bed nets and save lives. Another big area is animal welfare, and then the third area is the effect of the future, artificial intelligence and global catastrophic risks and trying to minimize virus.
Based on your evaluation, and then you've got another website just for people that are listening, called back of the envelope where you go through your calculations to figure out which causes have the most impact or the most expected impact. You just mentioned three there. Can you talk about which one has the greatest expected benefit?
It's interesting, with the Back of the Envelope Guide to Philanthropy that I created, there's some areas like global health that do seem to have a very high definite benefit but for some of the areas like the risks of artificial intelligence on the future, that probably have a higher expected benefit but there's much more uncertainty regarding what the actual value is.
Interesting. One of the other things on this topic that you've written about very well, and I really appreciated your thoughts, is on asset allocation for foundations or for donor-advised funds. Do you think if someone has set up a foundation or if they stuck money into a donor-advised fund, should they be thinking about asset allocation for their giving money in a way that is different from their personal consumption money?
I think the risks go down. There isn't as much concern you need to take about risks so you can invest more aggressively but the way I do it is I treat my personal assets and the money I've got set aside for going to charitable causes is one big bucket and try to optimize the overall outcome of that large bucket, then only at the end when I got a spending recommendation factor back and say how much of this should come out of the actual donor-advised fund or charitable foundation versus how much of this is personal expenditures.
Can you expand on why people can take more risk with the foundation money than they can with the personal consumption money?
It has to do with if the foundation loses money, there isn't as much harm to this social good or whatever it is you're trying to achieve with that money as if your personal funds. An example of this might be like the sensitivity of investors is higher to the stock market, like the stock market goes down by a factor of two and then the ropes gone down by a factor of two but the stock market doesn't have as much impact on the wealth of someone living in Africa on $2 a day.
Okay. That's fascinating. So your consumption varies is more sensitive to changes in the stock market but because each dollar is worth more to somebody who you maybe doing social good to by donating, the sensitivity is not as high to changes in the stock market. Is that right?
Yes. They've got other resources besides the stock market that are keeping them alive so they're less concerned about a drop in the value of the stock market and your charitable investment in them.
Wow. Fascinating. It is really interesting thinking. Is it possible to optimize the timing of spending from a foundation?
Well, in theory, the answer's yes. It's very easy. You just plug in the numbers and run some calculations and you'll get an answer but that answer depends hugely on what I call the charitable discount rate, which is whether $1 spent today can do more or less good than $1 spent in the future, and unfortunately, it's very hard to pin down the charitable discount rate, which makes it very difficult to know so just try and spend all your money today because you've been spending it today, that will allow the recipients of money to grow their wealth or should you hold on to it and defer spending into the future.
That is really interesting. How do you determine the charitable discount rate?
It's in an island. I don't have any good answers. I've plugged in my best testaments and unfortunately, they're all over the map but not precise enough to give you an answer to your original question which was how to use time for spending from a foundation.
I want to touch just on the asset allocation for foundations again. If the sensitivity of those assets to change in the stock market is less than our personal consumption assets, does it follow that you would want to be much more aggressive in your foundation money?
Yes.
To what extent? Are we talking about using leverage? Are we talking about small cap value tilts inside the endowment money?
Certainly a small cap value tilt but I think if it then gets into regulatory issues or if you want to get more aggressive in a foundation, there's issues like the regulator saying you're taking too much risk but I don't have a good understanding of the exact regulatory environment but something I've had is in concern of when you are running a foundation of that, of how much risk you can take.
Interesting. So would it be safe to summarize as theoretically, if you have a foundation or a donor-advised fund where you got this money that's for altruistic purposes, you want to take as much compensated risk as possible within the confines of whatever regulatory or otherwise, is that the sensible way to think about it?
I think that's close to the truth. I'm not sure. I think that does have to be an upper bound but that upper bound is certainly high up for your personal consumption.
Okay. Yeah, it's fascinating to think about. It really is.
So maybe we just shift to some questions on retirement planning. I'm curious if you can describe some of the big differences between, say, a typical practitioner approach to financial planning and the economist approach to financial planning.
Okay. So I think practitioners typically tend to focus on growing wealth and focus on how much assets you have and putting the dollar value on what your net worth is. As for economists, it's more concerned of what good is having that wealth? What are you going to do with it, and that is spend it and consume it. So economists are more concerned with utility of consumption. Utility just means how much good you can do or achieve for each dollar of spending. As you spend more each year, the utility doesn't increase linearly but it starts to saturate.
So an example of this is if I'm spending or consuming $100,000 a year, and then certainly, my consumption drums to $200,000 a year, then my utility that I receive from that consumption isn't going to double. It might be, say one and a half times as much just because I don't get the same value for money. And so that rate at which consumption tends to flatten is what economists call risk aversion or relative risk aversion.
Yeah. That's definitely different from just using something like a fixed withdraw rate. If we start getting into the utility of consumption. One of the things you've written quite a bit about and built some models for to capture the economists' view of financial planning is a stochastic dynamic programming. Can you describe what that is and how it can be used to solve for the optimal asset allocation and consumption using these economic models?
A stochastic dynamic programming is based on the idea that if you knew, for example, you are going to live to age 100, you could figure out what the optimal thing to do at age 100 is, which is you spend all your money so then if you'd step back to age 99 and say knowing what I'm going to do at age 100 if I've got, say, $200,000, what would I do at age 99 given the different possible returns I might get? And you'll look at all the different returns and the knowledge of what you'll do at age 100 and you'll work out what the optimal thing to do at age 99 is given by stochastic uncertainty in the returns.
And then you can keep stepping back to age 98 and 97 and each year, compute what to do based on what you know you'd do in future years. So that, in a nutshell, is stochastic dynamic programming.
So how is that better than a typical approach to financial planning? What does it mean to the listeners here that they may have done a typical financial plan?
I think the difference is it's based on utility of consumption, so it gives you results in terms of how much to expect to consume and it tells you how your consumption should change over time in response to changes in the market.
Interesting keeping in mind the margin of utility of consumption where you're much worse off as it drops below a certain point and you're not much better off if it goes above a certain point. Yeah, it is very interesting. And you've done a lot of work on this and you even built a tool that people can find online, A-Calc, which uses stochastic dynamic programming to give people results that they can use. It is using US taxes, I believe. Those are, for our Canadian listeners, just a note of caution if you do check out the tool. You've since then though moved on to reinforcement learning as opposed to stochastic dynamic programming.
Can you describe what that is and why you've decided that it is better than stochastic dynamic programming?
Reinforcement learning is a machine learning technique that unlike stochastic dynamic programming which will give you the precise mathematic and optimal solution, reinforcement learning learns by just doing millions and millions of simulated years in trying different things and seeing what gives a better result. And so gradually touching this matter of how you should behave closer and closer to the actual optimal value but never gives a precise optimum. Well, the reason for switching from stochastic dynamic programming that does have a precise solution and the precise optimum to reinforcement learning is reinforcement learning's able to handle a lot of real world complicating factors that stochastic dynamic programming can't.
So anywhere where you have a state variable, it gets very expensive to use stochastic dynamic programming. So an example of that state would be the cost basis of your investments or the value of a SPIA in addition to the value of your portfolio and so there's a lot of those things. And then the other issue is stochastic dynamic programming, you have to almost run things backwards in time, which doesn't sound difficult but it's a lot easier if you can run them up forwards in time and optimize that.
So did any of your beliefs or assumptions around financial planning change when you went through this reinforcement learning exercise?
I think it solidified some of the beliefs or suspicions I had. One of those has to do with the value of long duration inflation-indexed bonds, which the traditional financial planner, they seem dangerous because the long duration means they're very volatile but to the economists' perspective is you're like buying your retirement consumption and locking it in even though it may change in nominal terms if you've bought a 30-year inflation-indexed bond and so what that's going to give you for your time and consumption.
Generally speaking, how will you say that the nearer optimal, because as you said, it doesn't quite get to the optimal result but the nearer optimal generated results of the reinforcement learning process, how do they compare to things like, I don't know, the 4% rule or their typical retirement planning rules of thumb?
I did some analysis of this and I found that in every scenario are considered reinforcement learning outperform the rules of thumb and they did that from anywhere from 6% to 32%. That was after I had carefully tuned those rules of thumb because they'll have different parameters of how long you should assume you're going to live and what the stop on the asset allocation is. So in each of those rules of thumb are tuned then to find the optimal value which is a process, and then reinforcement learning was able to beat them straight off the bat.
When we say beat, so say beat them by 32%, is that measuring it in terms of consumption or utility? What is it measuring?
Something called certainty equivalent consumption and certainty equivalent is where constant consumption that would have the same value to you as an uncertain variable consumption that you might be expected to experience.
Wow. So just the anchor on the 4% rule is an example and in that case, you're spending 4% just for inflation of the starting value portfolio every year. If someone's using reinforcement learning for their retirement plan, what is their path going to look like? Are they spending a different amount every year?
Yes. You'll be spending a different amount every year and your asset allocation will be varying from one year to the next, not by huge amounts but by some more significant amounts.
How prevalent are annuities in the optimal model?
Annuities might create a significant improvement. That is SPIAs, Single Premium Immediate Annuities.
Is there a certain timing when annuities become more appealing?
I've done some work of this outside of the asset allocation models I've been evaluating and annuities are very similar to bonds and that they provide a fixed payment each year.
So based on that, you'd think you could substitute annuities for part of your bond portfolio but the surprising discrepancy with annuity is that makes it more advantageous to delay that purchase into later in life. That pricing discrepancy has to do with the mortality tables that annuity providers use. They lump everyone whose patches in the annuity into the same bucket, so at the age of 18, their mortality's based on people who bought an annuity 20 years ago might not now be in such good health and people who bought an annuity just last year are still in good health. And so because of that, you can actually do better by delaying your purchase of annuities.
I don't want to use the word rule of thumb here but I'm going to use it. Is there a rule of thumb in terms of the age or maybe in your results, what is the typical age where the annuities do start to make the most sense?
I'd say from 70 to 80 but this comes back to what I was saying before about reinforcement learning going off in an approximate solution. It doesn't give you the precise sense so it's a bit fuzzy. The nice thing about reinforcement learning is if something's significantly going to improve your welfare, it will tell you but when things are less clear, you get a fuzzy and uncertain answer and you can train the model once and get one recommendation and when you train it a second time and you get a different recommendation. So unfortunately, the precise timing of annuities is one of those things of.
It probably doesn't make a lot of different recommendations you get out of using reinforcement learning are a bit different every time you learn it.
That's fascinating, which is probably a closer reflection of reality. If someone were to give a point testament of this is the exact optimal age, it's unlikely to be accurate anyway. When we're talking about reinforcement learning, something just popped into my head when you're speaking there. What are the big assumptions that are going into this model? Are we having to define expected returns and life expectancy? What are the inputs that are going to change, most materially, the outputs?
Yes. You have to estimate both of those things. So, life expectancy is an important and I think under-discussed issue in retirement planning. So life expectancy, I have a stochastic model which is based on the Social Security Cohort Life Tables but they not let people apply to have their own adjustment based on how healthy they actually feel they are.
So Ben mentioned the 4% rule earlier. Is constant spending ever the optimal solution?
I don't believe so. Constant spending is trying to apply a fixed constant model to a dynamic varying situation and so the consequences are probably not going to end up very well. If we had a constant stock market then a constant spending rule would make sense but we don't.
I'm going to ask the inverse of the question and maybe the answer's obvious based on what you just said but is variable spending always what people should be aiming to do if they want to have an optimal retirement spending path?
Up to a point, variable spending makes sense but at some certain age, you get so old that you should probably call all your money into SPIAs and so at that point, you'd have constant consumption.
How do you practically recommend people that are entering retirement think about their spending? Do you think that they're best to really itemize their spending and find out what that basic need is every month? And then the discretionary via travel vehicles, luxury items, may be that excess amount that would be the variable part of their spending? Do you have any practical advice for people?
I know I do keep track of my spending based on categories and refer to that. I don't know what you'd do if you didn't keep track of your spending. I guess you can look up blank statements and get a sense of what you're spending each year.
But how much in percentage terms might you're spending vary year to year? Is it significant?
For me, it is just because probably the bulk of my spending goes to charitable causes. So there isn't that same sensitivity to variations in spending comes back to being able to take more risk than the charitable side of things and that fact fall over stock market more than personal spending.
So you're personal spending is relatively steady, stable, whereas if you have a good year in your portfolio, your charitable donations may be larger.
Wow. Is the variation in spending on the giving side dictated more so by the stock market or by the charitable discount rate that you talked about earlier?
More by the stock market. The charitable discount rate, so emphasized, it doesn't give a good prediction as to what to do. So what I did is I set the charitable discount rate to a value that may spend the charitable assets over my lifetime and be done with it.
Interesting. And then your variable giving, is it dictated by the same reinforcement learning model that your consumption would be?
Yes.
Okay. Fascinating. So you mentioned a couple of things earlier that I want to come back to. We've talked about variable spending, we've talked about variable asset allocation, which as some people may recognize as tactical asset allocation.
Again, I didn't mention you have another tool, AIPlanner, which again, people can use on this website. We can link to it in our show notes. And that's a reinforcement learning, financial planning tool that you've created online, which is very cool. It has the option to turn off and on tactical asset allocation. Now I want to ask, and you have a note on the website saying that in a tactical case, it gives you a slightly better outcome than the not tactical or strategic asset allocation or planning case. For tactical to be slightly optimal, how important is mean reversion in stock returns?
Mean reversion is one of the two big factors that go into the tactical planning. So the other factor is locking up the stock market volatility and taking advantage of when it's more volatile, you want to hold less stocks and vice versa.
So both mean reversion and volatility clustering better dictating the tactical allocations?
Yes.
Interesting. Then I'm asking you because this is something that we spend a lot of time digging into recently. I agree, volatility clustering is a thing. There's no question there. That one's fairly obvious. How confident are you in mean reversion in stock returns?
I've done some analysis of it myself and I'm not at all confident that it exists but if I take a view of, it might or might not exist then what level of mean reversion would you expect? They come out to something like, roughly 10% the stock market's overvalue with a 1% to just decline in returns going forward. So it's a bit like if you've got a bunch of numbers and they add up to 0.5 and you're trying to work out is the average equal to zero up to 93% confidence level. You might not be sure of that. It might be a borderline call but if you're going to do some work in mathematics based on the average, you're going to take the best testament to 0.5.
On the volatility clustering, we're talking about tactical planning changes. So in consumption and in asset allocation, the volatility clustering, they're pretty short-term phenomenons. So practically speaking, if someone's using the reinforcement learning model to adjust their spending and asset allocation, what would be a trigger? I guess is the question. Would you do it when markets are volatile, you change your asset allocation?
That's what it recommends, and this is doing asset allocation on a once a year basis. So you probably don't get the full advantage of volatility and targeting but you'd probably get some of the benefits, and this is ignoring a lot of the tax considerations and complications that would come from moving the portfolio around that much that frequently.
Right. Yeah. That's one of the, from our perspective, definitely one of the big deterrence of doing more tactical stuff is transaction cost and taxes, and especially if we say turn off mean reversion in the model, then I think it probably gets even less compelling.
So given all this work and all that you've learned in this area of personal financial planning and research, what's the key message that you want to get across to someone who maybe using traditional planning approaches?
One of them, I suggest, has to do with value of long duration inflation-indexed bonds. Then probably another important message is you can't do any of this planning, the economic planning, without knowing about pensions and social security and other forms of fixed guaranteed income, and so that makes a huge difference to things like asset allocation and consumption. And so I think that needs to be emphasized more than planned. It's meaningless to talk about what an asset allocation should be for a 65-year old without knowing, do they have $100,000 of pension income a year? Well, not.
If I was to sit you down with my 22-year old son, what would be your main piece of advice to him?
I think for him, it's focus on saving and I think your saving rate is more important that your asset allocation or some of these other things we talked about.
I don't know if I've seen academic papers on that but definitely some interesting blog posts showing the relative importance of saving versus asset allocation, so that's a very good point.
I want to come back on the long dated inflation-protected bonds for a minute because that is one that's the second thing that you said there, basically taking a holistic approach. 100% agree, and that's something that we do as well but the long dated inflation-protected bonds is not something that we've looked at, really, for clients in our financial planning practice. Can you talk a little bit more, practically speaking, about how that looks? Is that similar to annuities, where they're being purchased later in life?
Not necessarily. I think with inflation-indexed bonds, you can patch it from early in life and that's what I've done with my personal portfolio is it's a mixture of stocks and long duration inflation-indexed bonds.
Fascinating. And again on that same topic, it that something that you would be tactically allocating to based on the reinforcement learning model?
Yes, it is.
Okay. Very interesting. You had a paper in 2020, the Journal of Retirement on reinforcement learning and why you have transitioned to using it, so covering a lot of the points we've talked about in our conversation so far.
And you mentioned in that paper that this is the first new approach to the portfolio problem in over 50 years, which is a very interesting comment. Do you have any sense for what might be next? What's the next big innovation in the portfolio problem going to be?
I think there's two areas. One has to do with right now, you really require a lot of computing resources to solve using reinforcement learning and so what I do on AIPlanner is I compute a generic model and then I try and force each individual into that generically-trained model and get the results from that. As computers get faster and faster, I think it will become possible to train reinforcement learning to each individual scenario and so that will be one big improvement. And the second area is, I think, there probably needs to be some understanding, some theoretical understanding of the relationship between stochastic dynamic programming and reinforcement learning and what other links and relationship there because they're both trying to solve Markov decision processes.
And so there's some commonality that might be fruitful for both areas that could come from that theoretical understanding.
Fascinating. When you talk about training the machine learning process for a situation and that being an expensive process, what does that look like? How expensive are we talking?
We're talking about, for AIPlanner, I trained for 200 million years and then I trained 10 models for 200 million years in taking that so that's 2 billion simulated years of financial life.
Wow. Okay, because we often run 500 trials of our Monte Carlo simulations, so that's a little... A few more trials. Wow. I wasn't expecting the number to be that large.
Which leads us to our final question, Gordon. How do you define success in your life? You've had an incredible life and career in research and philanthropy. I'm really curious, how do you define success?
I think for me, I define success as having sufficient resources to be able to work on the problems that are both interesting and important.
Beautiful answer. This has been a real pleasure to get a chance to meet you and introduce you to our listeners. Gordon, thanks so much.
Thank you.
Links From Today’s Episode:
Rational Reminder on iTunes — https://itunes.apple.com/ca/podcast/the-rational-reminder-podcast/id1426530582.
Rational Reminder Website — https://rationalreminder.ca/
Shop Merch - https://shop.rationalreminder.ca/
Join the Community - https://community.rationalreminder.ca/
Follow us on Twitter - https://twitter.com/RationalRemind
Follow us on Instagram - @rationalreminder
Benjamin on Twitter — https://twitter.com/benjaminwfelix
Cameron on Twitter — https://twitter.com/CameronPassmore
Gordon Irlam — https://www.gordoni.com/