8 stories I read this week and what I took from them.
Klarna AI assistant – Klarna
- Klarna AI assistant handles two-thirds of customer service chats in its first month
- After 1 month, the AI assistant has handled 2/3 of customer service chats
- It’s doing the work of 700 people
- On par with human agents on customer satisfaction
- It is more accurate than humans: 25% drop in repeat enquiries
- It’s faster than humans: customer errands are resolved in 2mins vs 11mins for humans
- It’s online 24/7 and speaks 35 languages
- Klarna estimates the AI assistant will drive a $40m profit improvement
- Bottom line: Another example of AI efficiency gains in the real world
Ukraine remains stronger than you might think – Michael O’Hanlon
Opinion | Ukraine remains stronger than you might think - The Washington Post
O’Hanlon paints a picture of a continued stalemate, which differs from the consensus narrative of momentum in the war having shifted to Russia’s advantage.
‘Defense is stronger than offense’ and the war has effectively been in stalemate for well over a year already.
The arguments:
Current artillery should be enough for a stalemate
Ukraine’s army will not run out of soldiers
Ukraine’s economy has stabilised and even shown modest growth recently
Zelensky’s approval rating remains high (~77%)
Defenses against drones and cruise missiles are good
Ukraine made it through the winter with most power and heating systems intact
The majority of Americans (even Republicans) still support aid for Ukraine
But there is also a warning: like bankruptcy, defeat could occur gradually, then suddenly.
Bottom line: After 16 months of stalemate, the most likely scenario is more stalemate
Project 2025 – Heritage Foundation
The Heritage Foundation's Project 2025 wants to reshape America under Trump | The Week UK
If Trump wins a second term, the market’s focus will be on short-term economic impacts from issues like tax cuts and tariffs. The media’s focus will be on social issues.
Slower-moving themes like the strength and role of institutions can be less exciting in the short-term, but the hugely important in a crisis.
We are getting more hints at what Institutional Degradation 2.0 could look like:
The Heritage Foundation’s Project 2025 aims to ‘deconstruct the administrative state’ with a ‘sweeping expansion of presidential power over the machinery of government’. The plan would be to consolidate, shrink and restrain federal agencies.
The Center for American Renewal, amongst others, suggests invoking the Insurrection Act, which would allow Trump to deploy the military against civil demonstrations.
The Washington Post reports plans to bring regulators (e.g. FCC, FTC) under direct presidential control.
Also, plans to revive the practice of ‘impounding’ funds, refusing to spend authorised congressional funds on unwanted projects.
An overriding theme: Trump’s team has learnt lessons from the last time. They will be better prepared, more loyal and more competent. A dangerous mix.
Bottom line: Trump 2.0 could mean institutional degradation on steroids with a risk of dystopia, but markets will focus on the shorter-term effects like tax cuts and tariffs.
Material ESG Alpha: A Fundamentals-Based Perspective – Ahn et al
- https://doi.org/10.2308/TAR-2022-0256
- Do companies with improving ESG scores outperform? Does ESG add alpha?
- This paper argues that the answer is ‘yes, but’.
- They find that when you control for fundamental drivers of stock returns (e.g. profitability and growth factors), the alpha disappears. Turns out the alpha did not come from ESG, but from characteristics that are correlated with but not caused by ESG scores.
- Bottom line: There are many reasons for ESG, but superior returns may not be one of them, at least not with current scoring methodologies.
Canadian pension plan dumps stake in NYC commercial real estate project for just $1 – Fortune
- Commercial real estate: Canadian pension plan dumps Manhattan project for $1 | Fortune
- The headline sale price of $1 is eye-catching, but the buyer also assumed CPP’s share of the mortgage.
- Real estate analysts always talk about ‘transactional evidence’ as the true sign of where markets are. There haven’t been many transactions, which is why deals like this are interesting…and scary.
- A sign of capitulation? Max bearishness? That’s still a difficult call to make. Office real estate has been under pressure for 4 years and you could have found dozens of high-profile stories that looked like capitulation, but weren’t.
- Bottom line: Without a positive catalyst Deep Value is just a Value Trap.
When will OpenAI announce GPT-5? – Metaculus
Metaculus expert predictions puts the release date in October 2024. The estimate has been for autumn 2024 very consistently for many months.
This would be another high profile news flow event that feeds the narrative and keeps the focus on AI progress.
What I’d like to know from a market perspective:
How much better than GPT-4 will GPT-5 be? Enough to keep the market excitement of exponential growth alive?
Will GPT-5 shift the AGI timeline? Shift probabilities towards ‘AGI is near with current technologies’ or towards ‘we need more fundamental innovations for the next leap’?
Is there progress on hallucinations? This would make a huge difference for adoption in many professional use cases.
Bottom line: GPT-5 seems likely this year and will be an important milestone for the AI investment case.
Do Analysts and Investors Efficiently Respond to Managerial Linguistic Complexity during Conference Calls? – Bushee et al
- https://doi.org/10.2308/TAR-2019-0358
- Filtering out real insights from analyst earnings calls has always been an art rather than a science, but AI should be able to take it much closer to a science.
- We know that managers have private information on their company’s future earnings. This paper shows that the complexity of the language used on conference calls provides a signal about future earnings.
- Managers with positive private information are incentivised to actively engage with questions and give detailed technical answers. Managers who have negative private information obfuscate, discouraging detailed questions.
- To my surprise, analysts seem to be a little better at deciphering the management speak than investors. The latter seem to be better at recognising bad news than good news, unless they are actively participating on the earnings call.
- Bottom line: There is hidden value in earnings calls. Someone should build an AI tool for conference call analysis based on this.
Wars and elections - European Council on Foreign Relations
Wars and elections: How European leaders can maintain public support for Ukraine | ECFR
A public opinion poll on Ukraine across European countries.
Only 10% of Europeans think Ukraine will win the war. Twice as many think Russia will win. Whatever ‘winning’ means.
More Europeans support pushing Ukraine towards a peace deal (41%) than helping to retake territories (31%).
Bottom line: Europeans, across the board, are pessimistic on the outcome of the war. Support for Ukraine remains strong, but there are some signs of war fatigue.