Efficiency is having a Broadway moment

Everyone is getting AI chatbots, and Musk is getting much more

Increasingly, I can’t tell whether news is news.

Like, those woolly mice scientists made as the first step to bringing back mammoths. Is that news or a new Jurassic Park spin-off?
Those paddles US Democrats were brandishing during Trump’s first State of the Union. Is that news or a new Bargain Hunt special?
That almost-coup in South Korea. Is that news or a new rendition of Harry Potter and the Deathly Hallows?

Does it matter? Does it add up to anything? I can’t tell.

Certainly, when it comes to the fund-management industry, nothing has materially changed in the past year. As far as marketing is concerned, however, there’s been an amazing amount of razzle-dazzle, especially when it comes to AI.

Take this week’s Wired story revealing DOGE Has deployed its GSAi custom chatbot for 1,500 Federal workers, where “Elon Musk’s DOGE team is automating tasks as it continues its purge of the federal workforce.”

It looks like news! Automating tasks? For 1500 employees? I’m all ears:

“Elon Musk’s so-called Department of Government Efficiency has deployed a proprietary chatbot called GSAi to 1,500 federal workers at the General Services Administration”

Banger opening. Except, three paragraphs down we learn it is, in fact, not a proprietary chatbot, but a wrapper over commercial products (made by either Anthropic or Meta):

“The default model is Claude Haiku 3.5, but users can also choose to use Claude Sonnet 3.5 v2 and Meta LLaMa 3.2, depending on the task.”

Never mind the tool, tell me about the tasks being automated! A memo is there to help:

“How can I use the AI-powered chat?” reads an internal memo about the product. “The options are endless, and it will continue to improve as new information is added. You can: draft emails, create talking points, summarize text, write code.”

The memo also includes a warning: “Do not type or paste federal nonpublic information (such as work products, emails, photos, videos, audio, and conversations that are meant to be pre-decisional or internal to GSA) as well as personally identifiable information as inputs.”

Let’s recall what the General Services Administration actually does:

“The GSA’s roughly 12,000 employees are tasked with managing office buildings, contracts, and IT infrastructure across the federal government.”

So between the government’s contracts, its real estate portfolio, and its IT infrastructure, I’d say there’s rather a lot of pre-decisional or non-public information.

Okay, but the efficiencies! According to a story from last month, there was also a push to get GSA to install Microsoft’s Co-Pilot, so maybe ask them?

Microsoft has generously been compiling a list of hundreds of case studies on how real-world businesses are transforming with AI, each with a little sentence summing up the main results.

Most of the value is in purported “time savings”, rather than commercial metrics like revenue, new products or even cost cutting. Here are some of the measured time savings reported as success:

• Honeywell employees are saving 92 minutes per week — that’s 74 hours a year!

• Insight employees using Copilot are seeing four hours of productivity gained per week

• Localiza&Co reduced 8.3 working hours per employee per month.

• Investec is using estimating saving approximately 200 hours annually

Imagine trying to pitch these efficiencies to Elon Musk, back in the days when he was building Tesla or SpaceX. You know this is nonsense.

In short, this story isn’t about efficiencies, or task automation, or AI.
What’s going on? What kind of show is this?

“Musk’s team also seemingly hopes to use the chatbot and other AI tools to analyze huge swaths of contract and procurement data.“ (highlight mine)

There we go.

The US federal government is the largest, most lucrative buyer. Some things only it can purchase. This is typically done through bidding processes, normally designed to be hard for sellers to game.

But if you can barge in and grab aggregate data on what and how the US government pays...having sole access to such data is lucrative to the point it’s beyond currency. It’s a lever on currency.

Often times, adding more words helps explain things. This is clearly not the case here. On occasion, then, it’s more helpful to imagine a language with words taken out.

For example, this story only starts making sense when you imagine a person going through the world speaking almost-English: The only difference being, they’ve never had access to the notion of Having Enough.

Also Happening

• In case you are left wondering what those “other AI tools” might be: consider that all this contract data is likely locked in PDF documents. What you’d really like to do is pull out the meaty stuff (like pricing or payment terms), stick it in a big table, and run some data analysis.

  The trouble used to be turning the PDF back into text. AI has been able to help here. FinText has been doing similar stuff using a relatively new – and free – tool from IBM, and I’ve written up a case study about it.

Treasure Corner: It’s A Wrap

One last thing on efficiency. It’s quite difficult, you see, once you’ve been doing something repeatedly for a long time, to ask whether you should be doing it at all.

This newsletter had two objectives. The first was as a proof-of-concept.

The second was to capture the data sloshing about in academic papers on what gets funds to sell. Some of this data is still really good, like the big five-O.

But, as an approach, I think it’s now less relevant on the whole, because asset management has become so concentrated. All these studies are typically back-tested, and with market fundamentals now this different, it doesn’t feel right to mete out irrelevant advice.

Taken together, I feel keeping up this particular newsletter doesn’t quite make sense anymore. Thus, this is a goodbye, at least for now. Thank you.