Editorial
Editorial standards.
Almost every claim about AI in finance comes from someone with a reason to make it. The vendor wants its tool to look indispensable, the bank wants its rollout to look fast, and a headline wants the analyst job to look finished. So the first question asked of any claim here is whose it is.
Four kinds of source, kept apart
What a bank says about its own tools
A bank's own announcement, technology blog or earnings call is the best record there is that a tool exists, when it launched and what it is for. It is a weaker record of how much the tool gets used or what it changed, because the bank is marking its own homework. Guides lean on it for the first kind of fact and say so when it is the only source for the second.
What a vendor says about its product
A customer count, an accuracy score or a figure for hours saved is the vendor's claim, and it appears here as exactly that, with the vendor's name on it. A lab's benchmark on spreadsheet tasks tells you what the lab chose to measure, not what a first-year analyst will get out of the tool the night before a deadline.
What the press reports
Reporting on an internal memo or an interview with an executive is often the only public account of a rollout, so it gets used, with the outlet named. A number that exists only in blog posts citing nobody does not get used, however often it is repeated. At least one widely shared figure about AI in finance traces back to exactly that.
What people in the job say
Forum posts from candidates and junior staff are the best evidence of what a tool is like to use and the weakest evidence for anything with a number in it. They appear as what people report, and never as the only support for a figure.
What goes stale fastest
Numbers that keep moving
How many staff use a bank's assistant, how many AI use cases it has in production, how many job ads carry a new title: these are live counters, and they rise between one draft and the next. One public tracker of forward deployed engineer postings rose by more than a third in ten days in September 2026. A figure like that is either left out or given as what a named source said at a named time, never as a standing fact.
Product names
Microsoft has at least five products with Copilot in the name, and they do different jobs. BloombergGPT, the Bloomberg name most coverage reaches for, was a model working behind the Terminal's search rather than a product anyone opened by that name. So a guide names the capability first and the brand second, uses the name the maker uses today, and mentions an older name only where readers will still search for it.
Corrections
When a claim here turns out to be wrong, the sentence is fixed in the article itself, and the byline switches from the publication date to the date of the update, so a returning reader can see the page has changed. A tool that has been renamed or retired since a guide was written counts as an error to fix, not as background.
How AI is used to make this site
A site about AI in finance is written with AI tools, and it should say so here rather than in small print. Language models do much of the reading and the first drafting. People decide what is published, check every claim that names a bank, a product, a person or a number against its source, and cut whatever does not hold up. It is the same split between software and junior that the guides describe inside the banks.
What the tools do
- Find and read long filings, call transcripts and product pages
- Turn a sourced research file into a first draft
- Suggest quiz questions and point out where a guide contradicts itself
What a person does
- Opens the source behind every name, quote and figure
- Decides what is published, what is cut and what waits
- Answers for every sentence once it is live
There is one job the tools are never given. A model knows only what it read before its training stopped, and the products it gets asked about are usually newer than that: renamed, repriced or relaunched since. Ask one about a bank's assistant and it will describe an older version with complete confidence. So nothing about a current product comes from a model's memory. It comes from the maker's own page or a dated report, opened when the guide is written.
Who is behind it
Lena Arkwright is the pen name the site is written under, and the company that publishes it is named in the Impressum. No bank, AI lab or software vendor pays to appear in these guides, and none of them sees a page before it goes up. The one paid product linked from the site is Finance CV Check, a CV review for finance applicants, and it appears on a handful of pages.