General Learnings (and applied to Upstart)

This is a super long post, but I think may be worthwhile to some, so please bear with it. It’s what I think is the good stuff from a post that got deleted - no reason to go into it but suffice it to say I don’t want the “good” to be thrown away with the “bad.”

We owe it to ourselves to individually and collectively examine and grow from mistakes, and the mistakes we see on the board. That’s, as Saul puts it, getting us closer and closer to teaching ourselves to fish!

I think this board - including me for some of them - made a lot of mistakes with Upstart, and now it’s time for me and perhaps a few others to grow from this. If you disagree with some or all of my conclusions, I welcome that, but please do it respectfully. .

Finally, while this is not a post with a ton of analysis, this a post about potential errors in our analysis methodology. I apply my learning in the end to present my position on Upstart.

Learning #1: Don’t completely ignore guidance

Upstart had a blowout Q2 and the company cautioned it would not be repeated to the same level. There were multiple reasons why it would not be repeated including a very reasonable one: they had had a big leap over time in their automatic conversion rates, and that does not happen often.

But many members of the board were so excited that they decided to ignore the guidance, we knew better, and the price was bid up to what the board (and others outside the board, of course) collectively thought it should be rather than what a reasonable beat of guidance would be.

Yes, I know companies sandbag and we all expect them to do better. But the board’s expectations were so far beyond a reasonable beat of the guidance, that the guidance was just completely ignored. Moving forward, I don’t think we should do that, unless we see a clear and consistent pattern in extreme sandbagging - which means enough quarters as a public company to show that pattern, which Upstart did not have.

Learning #2: Don’t do second-derivative analysis, aka, “Reading tea leaves”

The board makes it clear how we analyze a stock: look at the earnings numbers. That’s the most important thing.

But we also want to look at other sources, such as new product offerings, new partnerships, big customer wins or losses, competitors wins’ etc., including items that have been shown to correlate to higher or lower revenues. These are indicative of good or bad progress. I call this “first derivative” analysis (ok, I am making these terms up.)

But what we should be cautious not to do is to try to read too much - or anything, for that matter - in the tea leaves outside of that. I call that “second derivative research.” In the case of Upstart, there were a lot of tea leaves being read - calculating the number of searches with Upstart-related keywords, reading the number of reviews, and trying to extrapolate from that, etc. None of those criteria had ever been correlated to higher revenue numbers.

We didn’t do that for any other stock, it helped build the buying frenzy, and it proved to be wrong. It’s too hard to play this game because of the profound lack of visibility, and I think we should refrain from doing that in the future.

Learning #3: If there is buying frenzy because of reading tea leaves and ignoring guidance, or for any other reason other than numbers, consider trimming

Wow, all the signs were there that Upstart would disappoint very high expectations. And the buying could only be described as a frenzy. I even told my friends not to buy until after earnings because I thought the stock would likely crash. So why didn’t I trim? Well, a teensy bit of me bought into the hype. And I didn’t want to be out during the run up. But I noted that Bert trimmed. It is something I will consider at times like this in the future. The key I will be looking for is frenzy. Be fearful when others are greedy…

Learning #4: If you ignore guidance and you read tea leaves, or have otherwise high expectations due to “what if” considerations, try not to blame the company for missing your numbers

We’re still talking about how Upstart disappointed! But they had a total blowout quarter! They had 17.6 QOQ growth and they guided next quarter on the top end to 16.2% QOQ which I would be thrilled with, even with no beat.

I know this a controversial point and that half the board thinks Upstart truly disappointed and the other half thinks otherwise. I submit that if people didn’t ignore guidance and didn’t have too-high expectations, then Upstart would not have disappointed. Upstart did not promise you more and not deliver. Assumptions were made based on the past that might have seemed reasonable at the time but that proved not to be correct.

Now, I know that sometimes people ignore guidance and it all works out. Happy for you when it does. I’m putting a reasonable beat on the guidance and until proven over many quarters that the company is a super-sandbagger, I’m going to keep it at that.

One extra point specific to Upstart: Hey, they are in the prediction business. So I would hope they could predict their earnings better than we can. Even more reason to listen to them - even if they beat by a lot in the occasional quarter.

Learning #5: Pick your criteria and stick with them (with an obvious caveat)

Ok, this one is complex, but I think it’s worth making.

When some board members entered into last quarter with expectations that were too high and were disappointed, culprits were sought. In the case of Upstart, one culprit was loan size. But loan size had never been a criteria for analyzing this stock before. Where did it come from? And why was it so important all of a sudden, when it had never been discussed as an important metric before?

(Compare this to RPOs and the decision of some members of the board to get out of Snowflake last year because of it. RPOs had been discussed as a metric before. Some of us were ok with it, others not, but the metric had been important before the earnings came out.)

Well, I don’t think loan size is important, at least not for the time being. If you follow the Upstart story, the C-levels made it clear they were going to descend lower and lower into the FICO pond, and a quick google analysis at that time that I did took maybe 30 minutes and found 1) articles that correlate loan sizes to FICO scores as in, the lower the score, the lower the loan size and 2) articles that show how loan sizes are all over the map from quarter to quarter on a macro level.

So here is the obvious caveat: Sometimes there is an important new data point that comes out. But people, if it’s not new, and you never thought it was important before, you owe it to yourself and to the board to really ponder whether it’s valuable before you react publicly to it. Is this a real problem? Could there be another explanation?

In Saul’s world there is no acceptable explanation for some things such as decreasing revenues. I’m totally on board for that! But when a company has a fantastic quarter (which, those of you who were with us at the time, Saul pointed out forcefully was the case), we should take care to be thoughtful before adding brand new criteria for evaluation, perhaps to support our own disappointment.

Learning #6: Always consider the possibility of confirmation bias

Confirmation bias is a known thing and we all have to watch for it. It is the tendency to search for, interpret, and favor information in a way that confirms one’s beliefs.

Hey, we all do it from time to time, some more than others, and it’s so hard to take ourselves up the higher level and ask: Could there be a possibility I’m not looking at everything here? Could there be a possibility I’m ignoring a good competitor? (I still worry about Zest AI) Could there be a possibility that now that I’ve turned bearish on a stock, I’m now out looking only for the bad?

Of course once we turn bearish we want to write our reasoning and send it out there. And we should. But I want to suggest that we take a lesson from the true elders of the board. I notice that when Saul turns bearish on a stock, he posts his reasons, he sells, and he moves on. If someone posts something positive about that stock, he might retort with his negative viewpoint, but he has otherwise moved on from the stock. All good.

What you don’t see Saul doing is spending time researching a stock he sold and continuing to provide negative commentary. He has moved on to the stocks he now favors, and that’s where his focus is.

If you find yourself frequently researching a stock you sold and and posting mostly negative commentary on it, and ignoring the positive aspects that you once considered, you may be having a touch of confirmation bias, and you might wish to seriously reconsider your post. It’s not a common behavior to continue to extensively research a stock you sold, and therefore you may not be doing your research to your usual high standards due to confirmation bias or other reasons.

Learning #6: Keep the emotion in check

This is just to be complete. Monkey said it so much better recently, Monkey says everything better! But I’m consolidating my learnings and this one always is worth saying. It’s emotion that will take you down. Selling when all the emotion is around fear will take you down.

Consolidating the learnings and applying them to Upstart:

Learning #1: Guidance has not changed. The team has not changed. The story - diving deeper into the sub-prime space - has not changed. The story - the AI will learn from the defaults - has not changed. The story - Upstart will slowly penetrate auto loans - has not changed.

To me lower loan size, higher default rates can be directly tied to the subprime space, Others have shown that in fact the loan packages are lower in the sub-prime space now, the yield is higher, and the default numbers are actually better than what was predicted.) Check.

Learning #2: No tea leaves. Someone posted a comment on the high number of Upstart reviews recently, but I am not going to extrapolate this into anything. I’ll wait for the numbers. Check.

Learning #3: Consider a trim at buying frenzy. Well, we don’t have to worry about any buying frenzy now, haha! What we are seeing is just the opposite! If I weren’t so big in on Upstart already, I would seriously be considering buying more right now. Remember: be fearful when others are greedy, and greedy when others are fearful. Check.

Learning #4: I fortunately did not have too-high earnings expectations and therefore Upstart the company did not disappoint. Obviously Upstart the stock has. But not the company. I am continuing to maintain a reasonable set of expectations - if Upstart comes in at the high end of guidance, I will be happy. From my perspective, they don’t even have to beat because their guidance is so great, although of course I hope they will. I’m looking for 255-265M in revenue, hopefully 265M, a 16.2 QOQ beat. Check.

Learning #5: My criteria were: 1) all that stuff we look at: revenue growth, transaction volume (because that is the equivalent of customers, I would not want to see a serious dip here although I recognize you won’t see the consistency you might with a good SaaS player), free cash flow growth, good gross margins, overall numbers. 2) Future guidance. 3) Continued evidence that they are executing on the plan. (Would love to see the number of bank partners added this quarter, although I’m not sure I’d sell if it wasn’t high if the other numbers hold up.)

I am not worried about loan sizes at this point and probably never will be. I mean, that’s a mix that I trust the team, which has not suddenly gone from brilliant to dumb, to play with. Heck, they have even gone into payday loans. Those are tiny. If the mix produces a healthy rise in revenue and looks sustainable, I’m in.

Ditto for rising delinquencies for now. If the delinquencies are below or at what they predicted and what outside agencies are saying is within a good range, I’m in. (Of course, if someone posts a comparison with similar lenders in the subprime space and shows Upstart having bigger problems than them, that would likely change my thinking.).

But otherwise, Check.

Learning #6: I am not going to be emotional about this stock just because it’s crashing. Check.

One final point: Some of you are out because of the unknowns. Of course I respect that decision. And I respect that the macro environment may play a role in loans, and we have no way of understanding what will happen. But hey, loans have always been around and rising rates may result in lower loan sizes but guess what? The overall revenues may stay the same because the rates will go up.

SaaS may be safer but not invincible either. You saw what happened to DataDog during Covid. And my guess is with our SaaS stocks, Mr. Market will be completely unforgiving if they miss expectations. Upstart is already wallowing in the stock-price gutter.

I see no reason to change my mind, certainly not now. My criteria point to staying all-in on this stock, trust this tremendously brilliant team that has been working on this problem for 8+ years, wait for the quarterly earnings report, and make a decision to sell or hold at that point.


Nice post MizzMonika. Thanks for posting it.


Lending to “poor people” — okay I’m not sure your post is even intended to be productive, but I’ll take the bait.

UPST doesn’t lend to poor people, they lend to qualified borrowers who don’t meet traditional criteria. For example, I have a tenant who is in her early 50s and has been renting for 20 years— renting from me for 5, and paying $2000 a month. She pays early every month. She also has significant cash savings, but is unable to get approved for a loan because her income is variable from month to month— primarily tips and cash from her own business.

I think UPST would be able to identify that a person who hasn’t missed a $2000 payment for 20 years can afford a $1500 one— but traditional lenders don’t.

Also… “no one has ever gotten rich lending to poor people.”

Sorry, I disagree, in fact lending to ‘poor people’ has probably made more people rich than almost any other single source in the world since the dawn of man. Think about it.

I think you’re missing the point with the AI. A lot of the power in AI and ML comes from history and “experience” for the software. New competitive AI would have a decade less data to draw from— all the while UPST is expanding their data points by increasing exposure through new partners and sectors.

AI doesn’t beat the market, but superior technology has always helped its users win (through the history of mankind)— and the same is true for the companies we invest in. If AI/ML is the modern day long-bow— then it matters very much.

When the repeater is developed, we’ll invest in that too.

And yes— this is probably OT.



Damn! You would think all those payday lenders would have long ago crashed and burned, but somehow they just keep expanding . . .

And now this from the Google machine: “With more than 110,000 loans originated, Upstart is rated #1 out of 58 lenders on Credit Karma, a tribute to the quality of experience we provide to our customers. Upstart borrowers average 28 years old, have average income of about $85,000 and are highly educated.”

An Upstart borrower may be a person with a low FICO score (or no FICO score) for any number of reasons. That status may or may not be indicative of creditworthiness.

And that’s what I believe to be the essence of Upstart’s business model. Fair Isaac came up with the 5 credit factors of the FICO score in 1989. How is it that FICO became a measure of creditworthiness? It satisfied “common sense”. It seemed reasonable, but it was not scientifically validated. It provided what seemed to be an objective, unbiased measurement. And it was calculated with data that was readily available. Upstart correctly saw this as an opportunity ripe for disruption.


I agree with a lot of what you posted, but I would take issue with your misgivings about “2nd derivative” analysis. Jon Wayne brought a lot of this type of analysis to the board and I, for one, appreciate it and would hope that he continues to expand our collective tool chest of analytical techniques.

As I see it, the fly in the ointment was not his analysis, it was an organized fraud attack that grossly skewed the numbers. No one on the board ever thought to suggest that possibly the analysis might be influenced by large scale fraud attempts. So maybe that’s a lesson to be learned, but it most certainly doesn’t invalidate the techniques that Jon developed. It stands as a warning to seek some validation of the analysis.

I fault myself for having failed to suggest that we should look for some validation. In my pre-retirement work life at one point I worked as an emissary for a data conversion organization within The Boeing Company. At one point we were trying to help an engineering organization convert digital product definition data from an older computer system to a new one. We never succeeded in this effort because we could not develop an automated way to validate that the new data exactly represented the same geometry as the old data. When you’re considering airplane design, safety of flight and all those sorts of things are pretty important. Failing validation we were left with no alternative but to keep two CAD systems in place. For all I know, they’re still both running. Maybe there’s a third one by now.


No one has gotten rich lending to poor people ever in the past 2000 years.

But UPST doesn’t lend to anybody, it merely assesses the risk for the lender, reducing (not eliminating) the risk to the lender.