Ecomm volumes, profits (or lack thereof) in GenAI and Temu's philosophy

15 October 2023 | Issue #11 - Mentions $UPS, $FDX, $AMZN, $MSFT, $ADBE, $ZM, $TEAM, $PDD, $3690

Oct 15, 2023

Welcome to the eleventh edition of Tech takes from the cheap seats. This will be my public journal, where I aim to write weekly on tech and consumer news and trends that I thought were interesting.

Let’s dig in.

Shipping and Ecommerce

From the WSJ

Parcel carriers are giving customers a break on prices ahead of what is set to be a weak holiday shipping season, the latest sign of a slowdown in demand for goods that is hitting every link in the global supply chain.
United Parcel Service and FedEx, which in recent years have raised prices in lockstep as volume swelled, have been offering discounts and other forms of cost relief to their customers, according to some businesses that have met with sales representatives from the two carriers. The U.S. Postal Service said last month that it wouldn’t impose a holiday surcharge this year.
“It’s not going to be a good peak. People are spending money on things that we don’t transport,” said Satish Jindel, president of ShipMatrix, which analyzes parcel-shipping data.
ShipMatrix estimates that carriers will deliver 82 million parcels a day during the peak holiday season that starts around Thanksgiving and ends in mid-January. Last year the industry delivered an average of 90 million parcels a day during that period.
Signs of a lackluster peak season have been flashing across the supply chain. Makers and sellers of clothing, electronics and other consumer goods are seeing demand cool as shoppers cut back or shift spending to travel and services. The rates that containerships charge to transport freight across the Pacific Ocean or from Asia to Europe are plunging, as are rates to move goods by truck from U.S. ports to warehouses and stores.

The article also highlights the increasing competition among logistics providers in recent years. As shipping capacity became strained due to heightened demand, businesses started embracing regional and alternative carriers, such as DHL and OnTrac. Additionally, Amazon Shipping extended its network to include external deliveries.

UPS and FedEx have significantly benefitted since the onset of the pandemic. Demand for shipping surged as consumers shifted to online shopping while restrictions disrupted global supply chains. Their power over customers got even stronger. This resulted in the introduction of holiday surcharges and price hikes exceeding 30% from 2019 to 2024. Now, we're witnessing a projected return of shipping volumes for the holiday season to pre-pandemic levels, as consumers allocate more of their disposable income to services and you have greater competition going after a shrinking pie. At least from a volume perspective the normalisation in ecommerce has yet to trough. It’s going to be tough for UPS and FedEx especially as Amazon tries to fill up its excess logistics capacity that it vastly overbuilt during covid.

I wrote the other week that it seemed as though Amazon was re-investing for growth when it announced plans to hire 250,000 new employees in preparation for the holiday season, which is up 66% yoy. This could imply that the company is going to gain decent share this year. Indeed, the company announced on Thursday that its fall Prime day “outpaced” last year’s event and its Prime members ordered more than 150 million items from third-party sellers, up from 100 million items in 2022.

GenAI is expensive

I’ve written in the past about how companies are thinking about pricing models with GenAI as well as cloud service providers insourcing chips. This article from the WSJ offers valuable insights, shedding light on the rationale behind these strategic choices. Below are some intriguing excerpts.

Microsoft, Google, Adobe and other tech companies investing in AI are experimenting with an array of tactics to make, market and charge for it.
AI often doesn’t have the economies of scale of standard software because it can require intense new calculations for each query. The more customers use the products, the more expensive it is to cover the infrastructure bills. These running costs expose companies charging flat fees for AI to potential losses.
Microsoft used AI from its partner OpenAI to launch GitHub Copilot, a service that helps programmers create, fix and translate code. It has been popular with coders—more than 1.5 million people have used it and it is helping build nearly half of Copilot users’ code—because it slashes the time and effort needed to program.
It has also been a money loser because it is so expensive to run.
Individuals pay $10 a month for the AI assistant. In the first few months of this year, the company was losing on average more than $20 a month per user, according to a person familiar with the figures, who said some users were costing the company as much as $80 a month.
Microsoft has been exploring using less powerful and cheaper AI tools for its Bing search engine, including some built with Meta Platforms’ open-source AI software, said people familiar with the discussions.
Zoom has developed a smaller, cheaper software for its AI assistant, powered by multiple models, including those made by OpenAI and Meta. It only uses the most powerful AI for the most difficult tasks.
Zoom doesn’t have to charge extra for the tool, which can summarize meetings and compose chat messages, because it avoids expensive AI when it can, said Smita Hashim, Zoom’s head of product, who has also worked at Microsoft and Google.
“We are pretty amazed and shocked at the big price tickets we are seeing from some of the competitors,” she said.
Adobe uses a system of credits to help ensure its AI image generator, Firefly, won’t put it into the red. Once Adobe customers go over their allotted monthly credits, the company slows down the speed of the service to discourage overuse.
“We are trying to provide great value but also protect ourselves on the cost side,” Adobe CEO Shantanu Narayen said.

The astonishing cost incurred by Microsoft for some users in compute usage is notable, with losses reaching as high as $80 per month! This cost dilemma sheds light on why the company is keen on pursuing vertical integration by developing its own chips and maintaining control over large language models (LLMs). In order for co-pilot to be profit-margin neutral for the company it is going to need to cost less than $15/user/month in compute. I suspect the company is not going to use GPT-4 as the default LLM for Office 365 co-pilot, or if they do, they may impose limitations, such as a monthly query quota, similar to the approach used by ChatGPT Plus and Adobe. The fact that Zoom is offering its AI assistant for free gives me some hope that software margins aren’t going to get squeezed from the extra computing costs. It’s actually pretty positive for those who can get customers to pay for it.

Atlassian acquires Loom

From TechCrunch

Atlassian announced this morning that it is acquiring video messaging service Loom for $975 million, the same company that had a $1.53 billion valuation in May 2021 when it announced a $130 million Series C. That was when companies were still thinking about all work being cloud-based and the future looked oh so bright.
As times have changed, so has the value of the company, but Atlassian still sees Loom and its 25 million customers, and more than 5 million video conversations per month, as a valuable asset. The company believes that it can be a useful collaboration tool for its platform, especially Jira and Confluence.
“Async video is the next evolution of team collaboration, and teaming up with Loom helps distributed teams communicate in deeply human ways,” Mike Cannon-Brookes, Atlassian co-founder and co-CEO said in a statement.

I saw a lot of takes on twitter, with many people expressing skepticism about the price (which appears to be approximately 20x ARR as per online sources). While I don’t have in-depth knowledge about the product, this thread below from @BucknSF suggests there’s strategic fit. The software itself seems to be loved by customers and is considered the best of breed for its use case. If all these assertations hold true, Atlassian should have the potential to generate added value beyond the roughly $2 Average Revenue Per User (ARPU) it currently earns from its existing customer base. It may not look so expensive 3-5 years out after integrating into Atlassian’s broader ecosystem.

Team behind Temu

I thought this article by Momentum Works on how Temu’s team was built was interesting. It provides insights into the operational philosophy of Pinduoduo (PDD), the parent company, and the strategies it employs when incubating new businesses. Notably, in late Q3 2020, PDD ventured into the online grocery sector with Duoduomaicai, and it has since navigated the fiercely competitive market to secure the top position alongside Meituan Select, Meituan's grocery business, in China. Now it’s mobilised core personnel from Duoduomaicai to run Temu. I’ve put some fascinating snippets below.

Promotions and Demotions

“Promotions and demotions” was a management strategy applied at both Pinduoduo’s main platform and in the grocery business. Before leading the grocery business, “Grape” had served as a senior executive. Later, due to a mistake, he was demoted. However, “Grape” was eventually promoted again due to outstanding performance.
Regarding his demotion, reliable sources claimed that “Grape” had artificially inflated the performance of the advertising platform, Duoduo Jin Bao (多多进宝), by using methods like phone credit top-ups. Some mid-level Pinduoduo employees familiar with “Grape” defended him, stating that for businesses like Duoduo Jin Bao, using such methods to inflate performance was necessary to meet targets.
Performance had to be summarised and reported to Abu {Portsea: PDD’s COO} every quarter, and it almost never met the standards, necessitating the redefinition of goals. If neither met the targets, then it came down to who performed slightly better, and that person would be ranked higher. Abu consistently maintained very high expectations for the team.
“Promotions and demotions” also meant there were no limits. Pinduoduo’s primary principle was never to cause losses to the company. However, if you were to ask where the line was drawn, there was none; it was not clearly defined. Essentially, you could do anything, as long as you didn’t cause losses to the company, as emphasised by Li Yang, a former Pinduoduo tech employee.

Interviewing prospective employees

Unlike other companies where HR asks questions related to one’s profession and past experiences, Pinduoduo’s HR asks extremely detailed, privacy-related questions, such as, “How many family members do you have? What do your parents do for a living? Do you have a mortgage or car loan? Are you a local resident?”
These questions are asked because through data analysis Pinduoduo realised that local residents in cities like Beijing and Shanghai, particularly those with family financial support, tend to resist working overtime. On the other hand, individuals from rural areas or those burdened with mortgage or car loans, with strong financial needs, often display greater drive and efficiency, and hence are more likely to produce results.

Management models in grocery

According to Wang Yuan, a person closely related to Pinduoduo, “ecommerce and grocery businesses have inherently different models, so the strategies from Pinduoduo’s main platform don’t necessarily apply. That’s why Pinduoduo’s management implemented “small central and large local” (“小中央大地方”) management model, where provincial leaders have a high degree of autonomy and can flexibly mobilise resources from multiple parties.
Under this model, regional leaders are like “local lords,” wielding significant power, while middle and top-level supervisors have the flexibility to mobilise resources as needed. Typically, when Duoduomaicai supervisors take on local roles, the first thing they do is select a warehouse. “Once you’ve made your choice, you have full authority to get things done.”
In a way, the Amoeba management model, which allows for mistakes and flexibility, is better suited for a new business like Duoduomaicai {Portsea: as opposed to a centralised management model}

…

In comparison to Duoduomaicai, Meituan Select is notably less efficient. Before making a decision, Meituan Select’s local managers typically need data support from the group’s business analysis team, and only then can they proceed with execution.
For instance, back when Meituan Select and Duoduomaicai were competing in the Guangdong market, just to decide whether to establish one warehouse in Guangzhou or have separate warehouses with one in Guangxi and one in Guangdong, Meituan would need to hold numerous meetings and discussions internally for such a minor issue. Pinduoduo’s approach was straightforward: one warehouse in Guangxi and one in Guangdong, each managed by “Yingjun” and “Grape” respectively.
A more direct example is the choice of products used to attract customers (引流品). Fresh pork is one of the biggest sources of traffic in the community group buying business (used to channel traffic to other products).
Pinduoduo’s provincial managers directly used pork to drive user traffic, while Meituan had to write reports to prove that pork could bring traffic. The data in the report even had to be detailed down to the type of pork and how the cold chain logistics would be managed. In contrast, Duoduomaicai took a simpler and more straightforward approach, using plastic bags and ice packs.
As a result, most of the time, decisions made by the Meituan Select team, from the group level to the local level, often take a long time to be executed.
Zhang Wei, a former middle manager at Duoduomaicai, stated, “Meituan Select lacks clear performance metrics, and its organisational efficiency is noticeably lower. Pinduoduo’s execution efficiency is highly focused, with just one metric — GMV and growth rate. Take parallel imports as an example; this is not a decision made by the leaders but a result of independent procurement decisions. After all, parallel imports can lower procurement costs and rapidly bring in users and GMV.”

On cost control and frugality

The excessive concentration of power in various provinces and regions of Duoduomaicai was bound to lead to corruption issues. Pinduoduo foresaw this and introduced the “three-pronged approach” (三板斧) to curb corruption in procurement and other aspects that have the potential to drive up operating costs.
The frugal and simple approach to expenses has existed since the early days of Pinduoduo. At that time, a few colleagues hired a housekeeper to cook for them and wanted the company to reimburse the expenses. However, when the boss found out, they were called to the office for questioning.
Even today, invoices over a hundred yuan at Pinduoduo are still subjected to strict approval. Currently, Abu is the last person to approve expenses in the reimbursement system. In the past, it was Colin Huang {Portsea: founder and former CEO of PDD}. There are 2 deep reasons behind this approval process: one is to prevent corruption, and the other is to save costs and create a psychological deterrent.

It appears to be a prevailing trend among leading retailers such as Amazon, Walmart, and PDD that a culture of frugality is essential. This aligns with their core value proposition of offering competitive pricing.

That’s all for this week. If you’ve made it this far, thanks for reading. If you’ve enjoyed this newsletter, consider subscribing or sharing with a friend

Share Tech takes from the cheap seats

I welcome any thoughts or feedback, feel free to shoot me an email at portseacapital@gmail.com. None of this is investment advice, do your own due diligence.

Tickers: UPS 0.00%↑ FDX 0.00%↑, AMZN 0.00%↑, MSFT 0.00%↑, ADBE 0.00%↑, ZM 0.00%↑, TEAM 0.00%↑, PDD 0.00%↑, $3690

Tech takes from the cheap seats