Domo Arigato Mr. Roboto – Utilizing AI Chatbots for an preliminary Inventory High quality scoring

As I discussed in my 2025 Q2 Efficiency overview, my central investing device is now a dynamic watchlist that prioritizes Firms based mostly on High quality, Valuation and Momentum.

I’ve already mentioned my preliminary method to momentum the final time.

Now the right way to method the standard of an organization in an environment friendly method ?

The primary problem right here is that I will be unable to do a “full stack” evaluation for each firm I come throughout. However I wish to have at the very least a fast “place to begin” which I might use to resolve if it is smart to dig deeper or not.

I outlined 10 standards that give me a primary perception on how an organization might match to my “Beuteschama” or not. If I might wish to verify these standards manually, I would wish to make use of a number of sources, comparable to TIKR, some inventory web sites, the businesses’’s IR web site and the annual and quarterly stories. A course of that on common takes me at the very least 60-90 minutes to get to a conclusion.

So I made a decision to check out LLMs in an effort to “outsource” this primary step screening and scoring.

I name this the “10 Issue Mannequin”. One comment: Once I say attention-grabbing to me, which means that in precept I’m on the lookout for boring, household/administration managed corporations which might be taking part in in comparatively enticing markets,  rising over the long run and are conservatively financed.

That is enticing for me as it fits my funding model and threat choice. Different buyers may need very completely different standards. I’ve not backtested these standards in any method, I simply assume that corporations that rating nicely will do nicely over the mid- to long run.

At this early stage within the funding course of, the mannequin additionally contains momentum, each, inventory worth and basic momentum to get a primary glimpse how the corporate is doing on this regard.

When an organization is attention-grabbing sufficient, I’ll attempt to substantiate the components by further evaluation in a second step earlier than I then add it to my “Prime 100 watchlist” (or not).

The ten components (in no specific order)

1. Shareholder construction

I do like corporations which have at the very least one vital (or a bunch of shareholders, ideally a household) that ideally enable the corporate to make use of and execute a long run technique. Ideally, Administration itself is a part of that Group. As a threshold, I exploit 25% of voting rights.

2. Working margin

Right here I exploit simplistically a threshold of 10% EBIT margin within the final FY. I’m absolutely conscious that sure enterprise fashions have larger or decrease common working margins (COSTCO). However as I discussed earlier than: That is only a first step.

3. Return on Fairness

12,5% is the hurdle which I contemplate to be a good ROE. After all, that is additionally a perform of leverage, however at this stage I wish to maintain issues easy.

4. Long run development

Right here, I outlined a threshold of 6% CAGR for the EPS over the latest 10 years or, if not accessible, 5 years. Why 6% ? As a result of that is barely larger than nominal GDP over the long run. 

5. Web Debt

A web Debt to EBITDA ration of beneath 1 provides us 1, the whole lot else is 0.

6. Cyclicality

This issue requires extra judgement. I both use the final sector or assume if the corporate has a really low beta that it’s not very cyclical. Non-cyclical is sweet and a rating of 1.

7. EPS Momentum

This one checks if within the final accessible monetary reporting interval, EPS has gone up. If sure, then 1, in any other case 0. It is a very crude proxy for basic momentum

8. Capital allocation high quality

That is as soon as once more a standards, the place the mannequin wants to guage info. Within the immediate I give 7 examples of which at the very least 5 ought to be fulfilled to get to a rating of 1. I additionally give the LLM some KO standards.

9. Market place & Market attractiveness

Right here, I boiled it all the way down to market development, diversification throughout nations and high 3 place in a market. I had extra advanced definitions however they didn’t work nicely.

10. 6 and 12 month worth momentum

A fast verify to have a look at inventory worth momentum. I exploit easy 6 and 12 month worth efficiency. If each are optimistic, the rating is one.

Closing rating:

On the finish of the train, the LLM ought to then add the person scores with a max of 10. I additionally ask for some KPIs (Market cap, P/E and so forth.) , info on administration, the newest information, a inventory chart and a strengths and weaknesses abstract. 

The present model of the immediate is embedded right here:

https://drive.google.com/file/d/1ZMIf7Nh__6frI8nsyWtcTZOaZfaxO_RF/view?usp=sharing

I might be very happy to obtain options for enhancements and trade “immediate secrets and techniques” 😉

Basic impression & Comparability of various AI fashions. 

Some remarks upfront: My aim is right here to get a fast overview with out requiring an excessive amount of effort and time (no uploads and so forth.). For this train I used ChatGPT 5 Professional, Gemini 2.5 professional, Claude Professional and Mistral & Perplexity with the free variations.

Listed below are some observations:

1. outcomes for a similar immediate can change considerably from sooner or later to a different and even one chat to a different
2. Inside a chat, outcomes are likely to worsen if you immediate a couple of occasions. It’s higher to create a brand new chat each time.
3. Additionally the output codecs can change from immediate to immediate for no specific purpose.
4. For extra obscure corporations, the variation within the outcomes is way larger. One of many examples that I’ve within the appendix is eurokai which obtained scores between 5 and 9 out of 10.
5. Generally, the fashions are usually not capable of appropriately add up the scores per class. So that you get 6 occasions a rating of 1 and the overall rating is 5.
6. Regardless of successful the Math-Olympics, calculating a ten yr CAGR from two EPS values appears to be actually tough for all of the fashions.
7. The fashions use very completely different sources for market knowledge with a really huge number of high quality
8. The “deep analysis” modes don’t at all times produce higher outcomes. Generally it will get worse. For my goal, the deep analysis modes take too lengthy in any case.
9. Solely Mistral was capable of embed inventory worth charts
10. Usually, the outcomes are getting higher should you ask the mannequin to only do it once more. Which is kind of annoying
11. It’s also possible to get completely different outcomes should you immediate the identical factor in two completely different languages (translation supplied by ChatGPT as an example)
12. It makes most sense to begin with corporations you already know nicely in an effort to choose the place every LLM has strengths and weaknesses.
13. For those who make the standards too difficult, the outcomes typically worsen
14. I attempted to ask the LLMs to outline standards themselves, however they typically didn’t work all that nicely

On the finish of the day, counting on one mannequin, particularly for smaller corporations is kind of hazardous, it is advisable have a look at at the very least at two of them. It’s at all times value to aks 

General, I’ve to say that my favourite LLM continues to be ChatGPT, though briefly, ChatGPT5 was actually unhealthy and sloppy. That is adopted by Gemini and Perplexity, which may be very quick.

Claude is enjoyable to make use of however not very correct. LeChat from Mistral continues to be Okay for a free device. I might by no means use Grok for apparent causes.

Abstract:

General, I’m fairly impressed how these LLMs work and enhance. I had been concerned in taking a look at Chatbot startups 6-7 years in the past and the development by LLMs is just breathtaking. Even within the weeks that I’ve been iterating on this duties, the fashions constantly improved, which I discover exceptional.

My “dwell knowledge Question & interpretation” job is clearly not the best use case, nevertheless it saves effort and time on my facet in comparison with guide work.

As I simply use this as a place to begin for additional analysis, I can dwell with the randomness within the outcomes.

To be continued….. 

Bonus track: Styx – Mr. Roboto

Appendix: Examples

Instance 1: EVS Broadcast

ChatGPT 5 with the “assume tougher” immediate to keep away from the sloppy fast and soiled end result. It took the longes with round 5 minutes.

It got here up with an 8 out of 10 rating. It truly bought the ten yr EPS development incorrect however determined to make use of the significantly better 5 yr price which is one thing I might do as nicely.

I just like the detailed evaluation of the Capital allocation chapter. The market phase was so so…

Perplexity got here to eight out of 10. Funnily sufficient, my weblog was referenced as a supply in an earlier question, however Perplexity appears to make use of completely different sources at completely different intervals of time for a similar immediate. It bought nevertheless the EPS development incorrect. EPS per share has elevated by ~6% per yr over 10 years and rather more over the previous 5 years.

Perplexity was actually quick.

Google Gemini normal:

Gemini 2.5. got here up with 8/10 factors. Available in the market attractiveness part it made one mistake: it thought-about the EMEA area as one nation and due to this fact didn’t give a degree which is clearly incorrect. Aside from that the outcomes have been Okay.

Claude additionally bought 10 yr EPS development incorrect, however nonetheless gave it a 1. In any other case the outcomes have been fairly respectable , too. I preferred the output format loads.

Mistral bought to a 6/10. It made a couple of errors, comparable to the ten yr EPS development price, capital allocation coverage and the share worth efficiency. Curiously, it’s the solely Chatbot that is ready to embed share worth charts.

Instance 2: Eurokai

Eurkai is clearly a little bit bit tougher than EVS, as there’s much less frequent info, much less analyst protection and so forth. Additionally it is a extra difficult firm (firm construction, Pref shares et.). Eurokai’s 10 yr EPS development is round 6,8%.

ChatGPT scored 6/10.

It bought EPS development incorrect. With the cyclical sensitivity, it gave a 0 regardless of acknowledging the very low beta. Capital allocation coverage might certainly be thought-about a 0. Market attractiveness is one thing I might not assist as on a glance by foundation, greater than 50% of exercise is in a single nation.

Perplexity got here up with 9/10 factors. It bought EPS development incorrect and was a little bit optimistic on the capital allocation standards. It additionally dodn’t know that Tom Eckelmann is now CEO.However total the outcomes are very respectable.

Gemini 2.5. flash normal this time returned 6/10 factors, and had clearly some issued retrieving more moderen info. I simply opened a brand new chat after wanting into EVS Broadcast and the efficiency appears to have immediately degraded and contained a whole lot of errors

Gemini within the Deep Analysis mode bought 7/10. Curiously it tried to calculate the ten yr EPS development from historic EPS numbers, however ended up calculating solely a 9 yr CAGR. Then again it appropriately discovered, that from a market attractiveness, Eurokai is relying an excessive amount of on Germany (40%).

Claude normal got here out with 5/10. It made a number of errors, particularly with ROE and EBIT margin.

Then again, Claude dug out some great things on compensation and the bios of the Eckelmanns.

Claude in  Analysis mode additionally got here out with 5/10, however did an error in including up the components. It ought to have been a 6/10.That is one thing one sees in different fashions as nicely.

It got here up with some unusual stuff, like a non-existing share dilution.

Le Chat from Mistral scored 7/10. It additionally bought the ten yr EPS development incorrect, but additionally the dominating shareholder proportion. Curiously it was the one mannequin that embeds inventory charts.

10-Faktoren-Analyse: Eurokai KGaA