How Canadian markets actually work — explained one idea at a time. No stock picks, nothing for sale. Just the machinery.

I'm Amy Xu. My day job is solutions architecture — about twenty years in IT, most of it inside Canadian financial institutions, designing cloud, security and data platforms. So I've been adjacent to the machinery of finance for a long time without being the person who explains it to anyone outside it.

The show is How Canadian Markets Work: a hundred-episode series explaining the Canadian financial system one concept at a time, roughly twenty minutes each. Not what to buy — how the thing is built. How a trade actually clears. Why buying bank shares tomorrow morning doesn't give the bank a cent. What a bond price is really telling you. Why the bid-ask spread is a genuine cost that appears on no statement you will ever receive.

There are two hosts, John and Jane, and their voices are AI-generated. I write and research every script myself. They have different jobs: John lays out how the structure works, and Jane asks the question the listener was already thinking, then pushes back when the tidy answer doesn't hold up. That second role matters more than it sounds. It's what keeps the show from becoming a lecture, and it forces every claim to be earned rather than asserted.

Listeners are roughly three groups: people starting to invest who are tired of being talked down to, people moving into the finance industry, and people studying for Canadian licensing exams who want concepts to make sense rather than just be memorized. Canada is underserved here. Most retail finance content is American, and the structure genuinely differs — different regulators, different account types, different tax treatment, different market plumbing.
 

Two observations, both from years of watching people around me try to learn this material.

First, almost all financial content is selling something — a platform, a course, a newsletter, a fund. Even when it's accurate, the incentive bends it. Second, the material that isn't selling anything is usually written for people who already understand it. Between the sales pitch and the textbook there wasn't much, and there was almost nothing Canadian.

My initial goal was narrow: explain the system I'd spent two decades building infrastructure for, in language my own family could follow, and get the details right. That last part turned out to be the hard constraint. Scripts are human-written and researched from primary sources — regulator publications, official course material, exchange documentation — and the arithmetic in every example gets verified before it's recorded. Fee compounding, bond pricing, option payoffs. If a number is invented to make an example work, the episode says so out loud.

The reason it's a podcast at all is that this material is genuinely better heard than read. A disagreement between two people is easier to follow than a wall of prose, and it lets the objections surface naturally instead of being buried in a footnote.

I do listen to podcasts, mostly long-form interviews and history. What I wanted was the explanatory density of a good textbook with the pacing of a conversation.

I started February 2026 and the first episode took 2 months to get from blank page to published — most of that was building the production pipeline, not writing.
 

I have a full-time architecture job, so this happens in evenings and on weekends. The thing that makes it possible is that I don't produce episodes one at a time. Scripts get written in batches and audio gets generated in batches, so a single sitting produces several episodes rather than part of one.

Writing is the expensive part. A script is around 2,500 words of dialogue, but the research behind it and the fact-checking after it take considerably longer than the writing. Audio generation and assembly are close to free in terms of time terms because they're scripted end to end — I run a command and come back later.

Costs are low and entirely self-funded. Podcast hosting on RSS.com, text-to-speech API usage per episode, and nothing else. No studio, no editor, no microphone, because there's no human recording. It comes to roughly 200 USD per month.

I'd rather be honest that the real cost isn't money. It's that a hundred episodes of researched material is a multi-year commitment, and the failure mode is running out of stamina at episode forty. Batching is the main defense against that.
 

No sponsorship, no revenue. Downloads are 1500 per month — this is an early show and I'd rather give you a real number than an impressive one.

On sponsorship I have a position I intend to hold. The show's entire premise is that most financial content is trying to sell you something and this one isn't. The moment a brokerage, fund company, or trading app sponsors an episode about what trading costs you, that premise is gone and it can't be rebuilt. Editorial independence is the only real asset a show like this has. I'd rather it stay small and keep it.

What I actually gain is sharper understanding. I've worked in and around banking technology for twenty years and I still found gaps in my own knowledge — places where I knew the procedure but not the reason. You cannot write a plain-language explanation of something you only half understand. The script will not close. Writing a hundred of these has been the most rigorous audit of my own knowledge I've ever done.

Beyond that: it's a skill I didn't have. Explaining technical material to people who aren't technical is directly useful in architecture work, where half the job is making a design legible to stakeholders who don't share your vocabulary.
 

It's unusual because there are no microphones and no guests.

The pipeline runs script first. I write the full dialogue in Markdown with section markers and delivery directions, research it from primary sources, and verify the arithmetic in every worked example. Nothing gets generated until the script is settled, because regenerating audio is cheap and rewriting a bad episode after publication is not.

Audio is generated with a text-to-speech API through a Python script I wrote. Each host maps to a fixed voice so they stay recognizable across a hundred episodes, and delivery direction is applied per section rather than per line. Audio is cached per section, so revising one exchange doesn't cost a full re-render of the episode. Assembly and loudness normalization run through ffmpeg so every episode lands at a consistent level.

Video versions are illustrated slide decks derived from the same scripts, which go to YouTube. Show notes come from the scripts too, so the written and audio versions can't drift apart.

The only tools that matter are a text editor, Python, ffmpeg, and a TTS API. No guests means no scheduling, which is the single largest reason a solo show this size is feasible alongside a full-time job.
 

Honestly, this is the part I've done the worst so far, and I'd rather say that than present a strategy I haven't executed.

I have 74 episodes published and no meaningful acquisition breakdown yet. The audience is small enough that any percentage I gave you would be noise.

What I've built rather than measured: a YouTube channel carrying illustrated versions of each episode, which I expect to be the main discovery surface because YouTube search behaves like a search engine and podcast directories mostly don't. Detailed show notes per episode with the concepts named explicitly, because someone searching "what is a bid-ask spread Canada" should be able to land on the right episode. Community posts, and answering every comment personally while that's still possible.

What I've deliberately not done is post promotional material into Canadian investing forums and subreddits. That's precisely my audience, and it's precisely where self-promotion gets you banned and burns the credibility that is the show's only asset.

The lesson so far is that making a hundred good episodes and distributing them are two entirely separate projects, and I badly underestimated the second one.
 

Things that weren't obvious to me beforehand:

The script is the show. Production quality, voices, music, artwork — none of it rescues weak writing, and all of it is cheap compared to the writing. I spend far more time on a script than on everything downstream combined, and that ratio still feels wrong in the right direction.

Automate before you scale, not after. I built the generation and assembly pipeline before episode ten. If I'd been assembling episodes by hand I'd have quit by episode twenty-five.

Audit your own feed early. I found duplicate audio files, missing episode and season tags, and truncated show notes in mine — issues that had been quietly degrading the listening experience for weeks. Run your feed through a validator before you have a back catalogue to fix.

Verify every number you say out loud. Once someone catches an arithmetic error in an episode about fees, nothing else you say about fees carries.

Decide your position on sponsorship before anyone offers you one. It's a much easier decision in the abstract.

And pick a subject you'd keep researching even with no audience. Mine has an audience of very nearly nobody so far, and I'm still writing episode fifty-plus, which is how I know the format was the right one for me.
 

Podcast: How Canadian Markets Work | Podcast on RSS.com
YouTube: https://www.youtube.com/channel/UCcfyyU1xsDnYT45QjwmFM7g
The show is free, has no sponsors, no Patreon, and nothing to buy. If you want to support it, start at Episode 1 and tell someone who's been meaning to figure this out.