Generative AI is poised to transform how we extract insights from financial data. Systems like Claude can read and comprehend financial statements, filings, and reports to provide useful summaries and analysis. This technology can revolutionize how individuals and organizations evaluate companies, industries, and markets. This post is intended to provide an example of the capabilities of generative A.I., using the tool Claude.ai created by Anthropic. I wanted to assess if Claude would be able to perform an analysis of Tesla’s most recent quarterly 10-Q (Q2 2023) based on a blog post written by Calcbench, “Parsing Tesla’s Latest Numbers“
Summary of CalcBench Post
In the post Calcbench describe Tesla’s disclosures as frustrating for analysts seeking operational and financial insights. But Calcbench’s tools can still help with analysis using available data. In the end, Calcbench was able to show the trend of Tesla’s Revenue by market segment from Q1 2020 to Q2 2023. Calcbench utilized Tesla’s XBRL filings for the analysis and observed:
That’s some mighty sharp and steady growth in the U.S. region (blue line) with rest of the world (yellow line) following a similar trajectory except for one dip in mid-2022. China (red line) is poking along at a somewhat lower pace.
The rest of the post switches to the need to manually identify important information in order to ascertain the average per vehicle sold of, $43,804 for Q2 2023.
What knowledge is needed to to analyze publicly traded company financial information?
I asked both Claude and ChatGPT, What areas of expertise would a person need to have to use Calcbench in order to analyze the financial report of a publicly traded company? List them in bullet form. Below are the results:
From Claude:
Here are some areas of expertise a person would need to effectively analyze a company’s financial report using Calcbench:
- Accounting knowledge – An understanding of accounting principles, standards, and methods is critical for analyzing financial reports. This includes knowledge of the income statement, balance sheet, cash flow statement, footnotes, revenue recognition, depreciation, goodwill, etc.
- Financial statement analysis – The ability to read and interpret financial statements is key. This involves analyzing profitability, liquidity, leverage, efficiency, and other metrics based on the reported financial data.
- Industry knowledge – Understanding the dynamics and metrics of the industry the company operates in provides context for analyzing performance. Knowledge of competitive forces, trends, and norms is helpful.
- Understanding of accounting regulations – An awareness of GAAP, IFRS, and SEC reporting requirements helps ensure proper analysis that complies with regulations. Keeps analyst updated on regulation changes.
- Financial modeling skills – The ability to build models using historical data to project future financial performance provides additional insight from reports. Requires knowledge of Excel, valuation, discounted cash flows, etc.
- Research skills – Can utilize various resources to supplement analysis such as economic data, industry reports, competitor analysis, news, and management commentary. Provides fuller picture.
- Critical thinking – Ability to interpret the numbers, spot inconsistencies, raise questions, assess quality of earnings, and draw insights from financial reports. Requires skepticism and attention to detail.
From ChatGPT:
To effectively use Calcbench for financial analysis of a publicly traded company, a person would need expertise in the following areas:
- Financial Statement Analysis: Understanding how to read and interpret financial statements such as balance sheets, income statements, and cash flow statements.
- Accounting Principles: Familiarity with generally accepted accounting principles (GAAP) and International Financial Reporting Standards (IFRS) to accurately assess financial data.
- Financial Metrics: Knowledge of key financial metrics and ratios like earnings per share (EPS), price-to-earnings (P/E) ratio, return on equity (ROE), etc.
- Industry Knowledge: Awareness of the specific industry the company operates in, as financial performance can vary significantly between industries.
- Securities Regulation: Understanding of securities laws and regulations that govern the reporting and disclosure requirements of publicly traded companies.
- Data Analysis: Proficiency in data analysis techniques to extract relevant information from the large datasets available on Calcbench.
- Corporate Finance: Knowledge of corporate finance concepts, such as cost of capital, capital structure, and valuation methods.
- Economic Factors: Awareness of macroeconomic factors that can impact a company’s performance, such as interest rates, inflation, and economic growth.
- Market Research: Ability to conduct market research to gain insights into the company’s competitive landscape and market trends.
- Forecasting: Skills in financial forecasting and modeling to project future financial performance based on historical data.
- Software Proficiency: Familiarity with using financial analysis tools and platforms, including Calcbench, to efficiently navigate and analyze financial data.
Keep in mind that while expertise in these areas is valuable, Calcbench is designed to provide a user-friendly interface, making it accessible to a broad range of users with varying levels of financial expertise.
Claude’s Analysis of Tesla
The remainder of this post will examine how Claude analyzed Tesla’s data. A few things to keep in mind. I only used Tesla’s Q2 2023 10Q, so the trend analysis was only done for the two quarters reported Q1 2023 and Q2 2023. Other than that I wanted to see if Claude could produce the same results as Calcbench, albeit with less of a need for the above areas of expertise. All that was needed was for me to use the correct prompt.
My first step was to save the 10-Q as a .pdf file, which I submitted to Claude and prompted with Tesla Q2 Earnings. I did not ask any questions, Claude understood the data to be a 10Q and provided the following analysis (I’ve highlighted the values that match those from Calcbench):
Based on my review of Tesla’s 10-Q filing for Q2 2023, here are some key takeaways:
- Tesla reported total revenues of $24.9 billion in Q2 2023, up 47% from $16.9 billion in Q2 2022. Automotive sales revenue was $20.4 billion, up 49% year-over-year.
- Net income attributable to common shareholders was $2.7 billion in Q2 2023, compared to $2.3 billion in Q2 2022.
- Automotive gross margin declined from 27.9% in Q2 2022 to 19.2% in Q2 2023, primarily due to lower average selling prices and underutilization of new factories.
- Operating expenses were $2.1 billion in Q2 2023, up 41% from $1.8 billion in Q2 2022, driven by higher R&D and SG&A expenses.
- Cash and cash equivalents grew to $15.3 billion at the end of Q2 2023, compared to $16.3 billion at the end of 2022.
- Capital expenditures were $4.1 billion in the first half of 2023, mainly related to expansions at Tesla’s factories. The company expects capex of $7-9 billion in 2023.
- Vehicle deliveries were 889,015 in Q2 2023, up 57% from 564,743 in Q2 2022. Model 3/Y deliveries grew 61% year-over-year.
- Tesla reiterated its goal to grow vehicle deliveries by 50% per year over a multi-year period.
In summary, Tesla delivered strong top-line growth but lower automotive margins in Q2. Cash generation remained robust even with high capex spending. Vehicle demand and production continued to grow rapidly year-over-year.
Next I wanted to see if Claude could calculate Sales by geographic region, which it was able to do:
I noticed that Claude was only providing Q2 data and I also wanted Q1 data so I asked it to include this results as well. I’m not certain this was necessary but note I included instructions on how to calculate Q1 data based on the headings used by Tesla in their 10Q:
Next I simply asked if Claude could put the data in table format and to do some simple analysis:
Having determined that up to this point Claude was providing correct data I moved on to what was the heat of the Calcbench blog post, determining average vehicle cost. This is non-trivial because of the way that Tesla’s reports needed information, or maybe more to the point to the way that Tesla has in obfuscating the needed data. Calcbench had to manually find the data (units produced and delivered) in order to retrieve the needed data for the calculation.
Could Claude do the same just by prompting?
Notice that Claude produced the correct number of vehicles produced and delivered for 2023, and estimated quarterly production. These estimates are not part of the Caclbench calculations and I wasn’t sure how Claude was calculated them so I asked.
At this point it needs to be pointed out that the estimated numbers do not appear in the 10Q but that Claude was using estimated US deliveries as one of the main assumptions so I asked about that and it is interesting and to me important to highlight Claude’s ability to explain itself.
I could not validate any of the numbers or the quoted remarks in the 10Q, so as I approached the end of this experiment, I simply asked Claude what the average vehicle revenue is for 2023 to compare that figure with the value computed by Calcbench.
This average vehicle cost is slightly higher than the $43,804 calculated by CalcBench but it reflects the total 2023 average vehicle revenue vs. the Q2 2023 average vehicle revenue and does indeed reflect the final findings from CalcBench, a decline in average vehicle revenue.
To summarize Claude was able to analyze a Tesla earnings report to provide key financial insights:
- Claude reads and understands Tesla’s quarterly 10-Q filing, including details about revenue, costs, profits.
- It can identify key takeaways from the filing, like 49% annual growth in automotive revenue.
- Claude can break down Tesla’s sales by geographic region and quarter when asked. It computes sales growth rates in each region.
- Using reported production/delivery numbers, Claude estimates quarterly deliveries by region.
- With these estimates, Claude calculates average sales price per vehicle by region and globally.
- Finally, Claude summarizes each analysis step in a comprehensive overview.
This example highlights Claude’s capabilities in:
- Reading complex financial filings
- Extracting key facts and figures
- Performing calculations to derive insights
- Answering followup questions and adding analysis
- Summarizing the process in easy-to-understand format
By automating analysis of public filings, generative AI can unlock huge time savings and benefits for investors, analysts, academics, and journalists. This technology is poised to become an indispensable tool for financial analysis in the near future.
Second Life in a Browser
It’s been a while since I’ve had a reason to post and a lot has occurred since my last post including the departure, reemergence and second departure of the founder and CEO Phillip Rosedale. An unexpected price increase or more precisely the ending of non-profit discounts saw many educators scurrying for alternative virtual worlds.
But with all this upheaval Linden Lab continues rolling out new and exciting features that are now in beta including Mesh imports and most recently the ability to access SL via a browser. Browser access is made possible by streaming SL via Gakai’s cloud-rendering.
I tried it today and with a few exceptions it worked flawlessly, see the video below:
Why is this important? After all since I’ve first began using sl I’ve seen flexi prims, mega prims, sculptie prims, voice chat, mono, display names, shared media, mesh imports, and now browser access. In a word it’s (or hopefully will be) greater and easier access. Now when students need to use sl the learning curve should be greatly reduced. Just provide them a URL (perhaps even avatars to use) and they can gain immediate entrance to the learning platform you’ve designed. If a new version of the Second Life client comes out students and most importantly campus labs will not need to re-download or reimage their computers. Â While we don’t know all the details yet, I expect this to also reduce some of the hardware barriers students sometimes face due to graphic card requirements. Â If this turns out to be the case than even more students will be able to access content created for Second Life. Â In fact Gakai’s website provides support for hoping this indeed will be the case:
Gaikai requires no special hardware or extra computing power. Rather, it can run premium 3D games on something as lightweight as a netbook or smartphone.
These are indeed exciting times for anyone interested in the benefits of virtual worlds.
Visualization of Data
A video showing the almost infinite possibilities for data visualization within Second Life by Alchemy Sims.
As you know I’m always on the lookout for new ways to create engaging material for my accounting students.  Well ZooBurst is a digital storytelling tool that is designed to let anyone easily create their own customized 3D pop-up books.  It is also an augmented reality program allowing you or anyone else to view the book you’ve created in the palm of your hands.
As you’ll see in the video below, you need pretty steady hands so this will work best for the surgeons among us.
The interface is simple enough to use so creating a story is as straight-forward as 1) having a story to tell; 2) finding the correct clip art, 3) moving, sizing, rotating the clip art on a page; 4) adding the text to he clip art callouts and to the page. Â After that you have a 3-D pop up book, viewable via the web, embedded in a web page (see below) or as mentioned earlier in the palm of your hands.
So, here is my story below, please let me know what you think of this tool as a way of engaging students with accounting content, what suggestions you have for improvement and if you try, please share the stories you’ve come up with.
Related articles by Zemanta
- Alpha.ZooBurst.com – 3D Books For Everybody (domainmacher.com)
- ZooBurst: Augmented Reality 3-D Pop-up Books (prathambooks.org)
- AR Stories (ya cant do this on a kindle or an ipad) (cogdogblog.com)
SL Pro! and Viewer 2.0
Today marks the first day of the SL Pro Conference . It started with a key note by Tom Hale and I’ve recapped some of his major points below.
Alluding to the new SL Viewer 2.0, he talked about taking the wraps off the new user experience work that Linden Lab has been working on during 2009!  He started with a summary of some 2009 statistics:
- 2009 User-user transactions grew by 65%.
- Xstreet Sales grew 74% 2008-2009.
- User hours grew 21% to 481 million per year.
- Active users grew by 5%
- Announced Linden Endowment for the Arts, designed to support, encourage and highlight Second Life artists. Recipients of an arts endowment will include a land grant; the creation of a dedicated Second Life arts committee; and the capability for Linden Lab to store art creations for future generations.
But so much of the content has been in a walled-gardent, because of “thick” downloads and viewers. Â “We must make it easier to amplify your awesome“.
In order to bring in new users, creators, business’s, organizers, etc., and because Linden Lab knows that new users become consumers the only way to do this (to increase the 5% user growth noted above) is to make the platform/ecosystem successful. In doing this Linden Lab looked to lessons learned from Facebook and iPhone -“experience matters”; the Quality of the Experience!  The eco-system (Second Life) creates the experience.
For 2009 the focus has been on the user experience and how it is mediated via the viewer.  The complexity of the viewer made it hard to adjust to Second Life, creating a feeling for new users of being overwhelmed. To address this issue Viewer 2.0 is being rolled out as a Beta today.
Viewer 2.0 - is the most ambitious project by Linden Lab to date (hmmm I’m thinking the Second Life platform had to be pretty abmitious for them). Some of the Viewer 2.0 capabilities are: shared media (support for Flash – finally), better search, easier navigation, alpha masks, and most importantly it is easier to use!
Search:  Powered by Google Search Appliance, my understand is that search will be an html experience allowing for search to be desegregated from the client.
Communication:Â Separate public from private chats. Made an effort to make voice more easily accessible. Â Messaging can now be managed (just like an in-box).
Home Panel – 100% html, will be (or can be) account and location aware. Â Mesh is not part of 2.0 but is coming later this year.
Shared Media (game changer??) – place any web URL onto a surface and allows you to interact with it just like a web browser, as you click a part of the web page the viewer will animate the camera to view the media – it breaks down the walled garden of Second Life including access to flash.  Opens up Second Life to web-standard type tools and technologies.  Supposedly this feature is drag and drop!
For 2010 Roadmap Linde Labs will be looking at asynchronous and synchronous interactions and how that works. Â But more broadly thinking how to improve the overall experience for the full life cycle, not just the newbie experience. Â Social experience, creator and developer experience will be targeted. Â Introducing new capabilities Media, Mesh, investing in platform (more scalable, modular), virtual goods, etc.
Intellectual Property – new tool for IP, called the viewer directory that lists 3rd party viewers that respect the rights of content developers. Â Viewer developers must comply with SL policies and SL terms of service. Â IP complaint process, IP takedown tool, Viewer Directory, Seller Register, and a new Commerce System. Â The openess of SL can not harm integrity of SL, SL will protect creators.
SnowGlobe 2.0 was also announced.
Tom’s biggest message seemed to be Linden Labs supporting “Amplifying the Awesome” by creating a platform that will help developers and through the new viewer allow new users to enter Second Life and not be overwhelmed with the “what do I do now” question. The attempt is to make the functioning of Second Life more intuitive – lets see how things shake out.
As some might know I’ve been using Second Life in my financial accounting class at the University of Central Florida since the fall 2007 semester. Since that time my class size has gone from an average of 200+ to 800+. I have tweaked the learning objects each semester to try to improve the students learning experience, from the way the students interact with the 3-D accounting equation and interactive T-account to how they submit their homework, now employing automated homework boxes that examine the formatting of the data and the naming convention of the notecards that the students use. The homework on the notecards is now automatically read and sent to a MySql database and a confirmation e-mail is sent (well usually this works) to students so that they know they’ve submitted things properly.
As the occasional reader of this blog will know, well firstly I only occasionally post, but secondly I’m interested in using Second Life to engage my students in what surprisingly I’ve been told is a rather un-engaging topic. So I’m always listening to feedback from my students about my use of Second Life and the problems some of them have. In fact when asked how my students feel I usually reply that anecdotally the students fall into thirds: 1/3 really like it, 1/3 really hate it, and 1/3 just use it because it’s something that is required.
But as I prepare for the start of the Spring 2010 semester I thought I should look at some actual data to see how students feel. I am always cognizant of the cost/benefit of using Second Life for my students and for ME (as it takes up most of my time, answering orientation questions, grading non-automated HW assignments, dealing with students whose laptops aren’t powerful enough, etc.). And I wanted to see if the “noise” from the 1/3 “I hate it group” is causing me to actually inflate the size of that group – the old squeaky wheel thing going on.
So I looked at data I collect at the end of each semester. I ask my students to take a survey in which I ask for feedback on all the technology I use in the course: Second Life, Twitter, Concept Mapping software, My Accounting Lab, and Meebo (yes I know its a lot of technology for an accounting course). I’m not going to go over all the results to that survey in this post but want to examine one of the questions that I ask:
For each of the Second Life objects listed below, use the scale below to indicate your agreement with the following: I felt that the Second Life objects helped my understanding of the underlying accounting concepts in acg2021
These are the results:
For Fall 2009 40.7% of the students agreed or strongly agreed that Second Life helped them understand the underlying concepts of financial accounting! And it’s actually better then that. Included in the average % is data related to students who used Second Life to watch my lectures (as opposed to using a website – though 15% of students did report they watched lectures at least some of the time via Second Life), removing that and just examining how the students felt the learning objects helped them to understand the underlying concepts of financial accounting we see that 47% of the students agreed or strongly agreed that Second Life helped them to learn accounting! Wow! So how do previous semesters stack up, here they are:
Spring 2009
Fall 2008
Spring 2008
You will notice that when class sizes are relatively smaller more students felt that the use of Second Life helped them to understand the accounting concepts, those students who agreed or strongly agreed overall represented 55.5% and 58.6% when using just the learning objects.
Fall 2007
The number’s here are comparatively low, but during this semester Second Life was voluntary and not a required component of the course.
So (and BTW if you’re still reading this far down, thanks!) I would say that any tool in which 40% or more students agree helps them to understand accounting is well worth the effort.
Second Life does Translation
I’ve just finished playing with a new release of the Second Lifeâ„¢ Snow Globe viewer. In this release, chat is translated automatically. The chat translation is not the default though, you need to select it from your Preferences, Text Chat and select the language the chat is translated to/from.

Once you’ve done this, you can now chat with a group of avatars and they will see your chat translated into the language you’ve selected in your prefences (see Chat Language above) and their chat will also be translated.
I teach accounting so you might wonder how this could affect me. Well not only does the chat translator work with avatar chat but it also works with object chat. That is, with the text an educational object is scripted to “say”. This could be objects used in an orientation for your students, or in learning objects. Now if a student is more comfortable in a language other then English, they can use the same objects you create and read the chat in their own language. For example see this image of my accounting equation providing feedback and notice that the text is shown in both English and Spanish.
Now I don’t know Spanish so I can’t vouch for how good (or not good) they are, but I imagine they will only improve over time. Certainly if I was teaching a foreign language this would be a huge boon, but I can also see other possibilities, what if there were an accounting translation version? Accounting after all is the language of business, and many first time students struggle more with all the new terminology then with the actual number crunching.
To try it out yourself you need to download the beta Snow Globe Viewer, the links are provided below (Note: I was using the Mac version and it worked just fine).
On Sunday I’ll be doing a workshop as part of the American Accounting Association‘s annual conference. My workshop will be given prior to the officaial start of the AAA as part of the Conference on Teaching and Learning in Accounting (CTLA).
I struggled to figure out how to do a workshop about Second Life in a limited amount of time, with attendees who have limited experience in Second Life if any, and who will have limited if any internet connections. I wanted to keep away from PowerPoint and find something that would be compelling enough to keep the groups attention, my argument is that Second Life is engaging so my workshop should be too. Thankfully I stumbled upon a wonderful new presentation tool called Prezi.
I see Prezi as a tool that helps you to tell a story and while PowerPoint lends oneself to use a lot of text, Prezi lends oneself to use more media – flash movies and pictures along with text to help tell the story. You don’t have to worry about transitions, Prezi zooms into each object in your presentation which turns out to be a really nice way to show emphasis. Prezi is fairly new and there are certainly some things I’d like to see. For example, a way to include sound so that I could create a narrated presentation or at least some background music. It can be more flexible in terms of themes and color choices as well, but all in all I think it did a nice job of forcing me to think a bit outside of the box. Will see what the workshop thinks about it.
Below is an embedded version of the prezi (you can also see a larger version that is hosted on their site).  One thing that might not be inuitive at first is the need to click through the presentation and that is accomplished by using the controls on the bottom right corner of the prezi. You can view it full screen, click on any object to zoom in for a closer look, and click the links to open up a web page. One of the nice features is that when an object is a flash movie, it automatically starts playing nice and seamless. Let me know what you think in the comments.
I’ll also be giving a talk about Twitter and how it can be leveraged for classroom use in which I’ll be doing a basic run down of what twitter is – for anyone in attendance living under their green eye shades. I’ll also cover some of the relatively newer ways of interfacing with Twitter including TweetDeck, Seesmic, Spreadtweet, Mishnet, and Tunein. For managing multiple accounts something most anyone using twitter will end up doing, one for personal use and one for the classroom, I’ll be talking about two products, HootSuite and CoTweet. And finally I’ll be discussion a neat tool from Poll Everywhere that allows you to take polls with responses coming from various sources – the web, text messages, mobile web and of course twitter.
http://maxping.org/business/news/meerkat-may-impact-second-inventory-market-share.aspx
I haven’t tried this yet but will later and will update the post with my impressions, but as the video demonstrates at the link above, this tools could greatly enhance the ability to, of course, back up your work (I wonder what it does with the scripts contained in the objects you backup???), but also serve as a semmingly very easy tool for distributing content to users – either paying or in an opensource model. Will post more as I learn.
Wordle
After just receiving a tweet from @carolrainbow about Wordle which is a
is a toy for generating “word clouds†from text that you provide. The clouds give greater prominence to words that appear more frequently in the source text. You can tweak your clouds with different fonts, layouts, and color schemes. The images you create with Wordle are yours to use however you like. You can print them out, or save them to the Wordle gallery to share with your friends.
I checked out the presentation by Tom Barrett, and then made my own Wordle of this blog, very nice way to visualize text and the whole process to tweet, to prezo view, to Worldle blog about 20 minutes.