Matthew Andrews
Product Manager ・ Software Engineer ・ Executive MBA Graduate ・ Tokyo Resident
IE Executive MBA Review — What I learned

I graduated from IE Business School’s Executive MBA programme in Madrid, Spain in December 2019.

The 15 month programme was conducted mostly online so I took most of the classes crouched over my laptop in the dead of the night from my adopted home in Tokyo, Japan.

Here’s me happily accepting my certificate:

A colleague of mine asked me today:

(How) do you think it has been valuable for you?

Before I answer I want to first write a short disclaimer.

MBAs are very expensive, many are designed for a world that is increasingly out of date and I am certain free or low cost online equivalents of the vast majority of the course content now exist on Udemy, YouTube, Coursera, etc. I genuinely believe a Harvard Business Review subscription, especially reading back through their greatest hits, gives you a third of the value of an MBA with less than 1% of the cost.

For me I was extremely fortunate in three ways that made the full MBA worthwhile:

  • Firstly, my employer chose me as one of the three candidates they put forward each year for full sponsorship.
  • Secondly, I joined IE: an incredibly forward-thinking school who have technology capable of running truly collaborative and engaging classes connecting teachers and students across all the world’s populated continents at once.
  • Finally, almost everything we learnt was new to me. I believe had I already studied business at undergraduate level and complemented it with relevant work experience, I would have got much less out of it.

Your mileage may vary.

(How) has an MBA been valuable for me?

  • Before, I might think of an idea for a business and fantasise it would make me a gazillionaire. Now, I feel I know more clearly than ever what needs to be done to take that idea, evaluate its potential, make decisions about the trade-offs between various ways of getting it funded, and how it might be brought to life. I also realise how much work and risky it is — and now feel almost (but not quite) all of my previous ideas lack the potential to make them worth the effort.
  • I can now read & understand basic financial statements, but despite having a numerical background (BSc Mathematics) I’m not as good at it as I thought I would be.
  • Basic knowledge of accoutancy has given me better understanding about how (and why) budgeting works the way it does at Nikkei and the FT, for example why it isn’t always possible to simply roll over unused budget from previous years into the next.
  • I realised I can write quite well. I guess 8 years working for a newspaper has had some effect!
  • I learnt how to make a marketing plan … and made a basic one for part of my current project at the FT (and it was received well!)
  • I learnt what is considered best practice when Negotiating … and now use this in my daily life …
  • Studying IT/Technology from an academic & business perspective made me re-realise I’m really good at it and how much I love it.
  • On the other side, it’s given me a much greater respect for those ‘from the business side’. Finance, Strategy, Sales, Marketing are crafts in their own right and are in my opinion far more difficult than the likes of IT and operations. Why/what questions are infinitely more difficult to answer than who/how.
  • I learnt what Networking actually is and realised I’m not as bad at it and enjoy it much more than I thought.
  • It will help the next step of my career — to move into product focused technology roles.
  • And finally, I’ve met a bunch of wonderful people doing crazy amazing things around the world!
How to securely configure Cloudflare for static website hosting on AWS S3

Goal: to create a static website hosted on S3 and Cloudflare securely — by which I mean restrict access to the contents of the bucket to Cloudflare only.

Step 1 - Create your bucket

First we’ll create your S3 bucket. Login to the AWS console and head over the S3 service and click the big blue + Create Bucket button to start the wizard that will guide you through creating your bucket.

The first question it’ll ask you is for a name. The name that you will need to enter will need to match the eventual hostname website you plan to host. I’m going to create so I’ve named the bucket that:-

Leave the rest as default and hit Next.

Step 2 - Configure Options

On the Configure options you can leave all the other options as default.

Step 3 - Set permissions

Again leave this as default (Block all public access), which at first might not seem intuitive as we’re making a public website. But don’t worry, in a later step we will grant permission just to Cloudflare to access the content.

Step 4 - Review & Create bucket

Assuming you’ve followed everything correctly so far — go ahead and hit Create Bucket.

Step 5 - Enable static website hosting

Staying in the AWS console for the S3 service, click into your newly created empty bucket, and click Properties.

Then, click the Static website hosting box — and choose Use this bucket to host a website.

You’ll need to enter the Index document, which I’ll set to be their suggestion, index.html.

Make sure to take note of the Endpoint — you’ll need that later! In my case this is

Step 6 - Grant access to your bucket from Cloudflare

Next, click the Permissions tab of your S3 bucket (between Properties and Management) and click Bucket Policy.

Take a copy of the following JSON, take care to replace (my bucketname) with your bucket name, then paste it into the Bucket policy editor. Then click Save.

"Version": "2012-10-17",
"Statement": [
"Sid": "",
"Effect": "Allow",
"Principal": "*",
"Action": "s3:GetObject",
"Resource": "*",
"Condition": {
"IpAddress": {
"aws:SourceIp": [

(This grants Cloudflare’s IP ranges read-only access to your S3 bucket contents)

Extra step - Upload a index.html file to your bucket

Strictly speaking not necessary but useful to check everything is working.

I uploaded an index.html file with the contents <h1>Hello world</h1> to my S3 bucket.

Step 7 - Add the DNS hostname to your Cloudflare DNS console

  • Login to your Cloudflare dashboard
  • Click on the domain that you’ll be setting up a static website on (in my case
  • Then click DNS
  • Click Add record
  • Set the Type to be CNAME
  • Set the Name to be the subdomain (or leave blank if you are creating a static website on the root of your website)
  • Set the Target to be equal to the Endpoint that you were given in step 5 but remove the http:// prefix and / suffix. In my case this is
  • Leave TTL and Proxy Status to be their defaults (Auto and Proxied)
  • Click Save


You’re done — load up your domain to see the fruits of your hard work:

Semver as a Service

I’ve been continuing learning bits and pieces with mini projects… This time: Semver as a Service built with AWS Lambda, AWS API Gateway, AWS CloudFormation and Golang.

What is ‘Semver as a Service’?

Semver as a Service is a simple API that will look at any GitHub repository’s releases/tags, sort them and tell you the highest version or, if you specify a constraint, the highest version that meets a constraint.

Try it out here:-


Well, the main purpose was to learn Go, AWS, etc, but it’s also handy for writing install scripts. For example, this could be a simple script to install the latest version of s3up on your Mac:-

curl -sf \
| xargs -I '{}' curl -sfL{}/s3up_darwin_386 -o /usr/local/bin/s3up \
&& chmod +x /usr/local/bin/s3up
Catching All Errors in AWS Lambda and API Gateway

When building applications with AWS Lambda and API Gateway I’ve found error handling quite difficult to work with.

You first define what status codes your API method is able to serve (200, 404 and 500, for example). You are encouraged to choose 200 as the default. Then you can write regular expressions that match against ‘Lambda Errors’.

According to Amazon’s documentation:-

For Lambda error regex […] type a regular expression to specify which Lambda function error strings (for a Lambda function) […] map to this output mapping.

The error patterns are matched against the errorMessage property in the Lambda response, which is populated by in Node.js or by throw new MyException(errorMessage) in Java.
Be aware of the fact that the .\ pattern will not match any newline (\n).

This seems simple enough.

Lambda functions that have run successfully shouldn’t have errorMessages so I should be able to:-

  1. Set a Lambda Error Regex that looks for .*404 Not Found.* and maps that to 404 errors — this works fine
  2. and then I should be able to map all other errors to 500 with (\n|.)* (note the \n is there because I heeded the warning in the documentation above in case one of my errors has a new line).

Whilst the Lambda Error Regex does indeed now map all errors to 500 responses, unfortunately it also maps all the successful Lambda response to 500s as well.

Lambda Error Regex


So, how do we fix it?

Easy. Whilst the Lambda Error Regex is used to compare against successful Lambda responses, in this case errorMessage is set to something like an empty string.

Just set the Lambda Error Regex that you want to match to your ‘catch all’ error response to (\n|.)+.

Like this:-


I’m really surprised that this is so difficult and that none of the documentation encourages (or helps) developers to write Lambda Error Regexs that match against all possible errors.

If I had to write regular expressions against all the errors I anticipated having to handle I would never feel 100% confident that I got them all and would have needlessly risked returning 200 responses containing errors to users.

Building a half-a-second website

I’ve spent the past couple of weekends rebuilding my website. Previously it was a really old, slow, out-of-date WordPress site running on ridiculously expensive (for what it was) GoDaddy shared hosting. Converting it to a statically generated (Jekyll or similar) site had been on my to-do list for years…

This is it.

Tools and architecture

  • It’s built with (although I swapped out the Sass compilation with one we developed for the Financial Times and removed the client side JavaScript entirely.
  • It’s hosted on S3 (provisioned with CloudFormation).
  • Circle CI runs the builds and pushes to production on green (when linting passes and the pages build).
  • It’s behind a CDN (CloudFlare) who provide SSL for free (thank you CloudFlare <3). They also support HTTP2 and have a nice API that you can use to do some clever cache optimisations with…

Purge on deploy

Currently the CDN in front of is configured to store everything for up to 1 month (and I’m talking to CloudFlare to see if I can increase this to a year) but only instruct users’ browsers to only cache pages for up to 30 minutes. Then, I have set things up to call the CloudFlare API to automatically purge the files that have changed — and only the files that have changed.

Now clearly since Circle CI is already running all my build steps for me and knows what files have changed it could easily coordinate purging of the CDN. Indeed, we use this pattern a lot at the FT. But that was nowhere near over-engineered enough to qualify for a weekend hack project.

Instead, I created a Lambda function that was connected to my website’s S3 bucket’s ObjectRemoved and ObjectCreated streams. Each change in the S3 bucket generates an event that then triggers a Lambda function (written in Go) that purges the CDN for the associated pages. See the code.

Making this change caused the cache hit ratio to jump and even though the website was already fast before making this change, it’s now even faster still. Pages no longer need to travel all the way from Ireland (where my S3 bucket is) to reach every user — it would be as if the site had servers in every one of these cities around the world.

HTTP2 + S3 + CDN make a very fast website

When you add together HTTP2, S3 and smart use of a CDN you get a very performant website.

The above image shows that, occasionally, pages take the almost same amount of time to load in production (right) as they do on my local machine (left). Production isn’t always this quick (a few, very unscientific and statistically invalid spot checks of pages on shows that most of the site loads in about half a second, but is sometimes as slow as 800ms) but it does show that a crazy level of performance is possible.

And there’s so much more left to optimise.