What do all of you view as Upstart’s long term competitive advantage?
Some thoughts:
Upstart’s success on a company is predicated on it’s AI model’s performance relative to FICO, and to the other currently available credit lending methods. If Upstart is unable to have a superior AI model, they will likely go out of business and lose market share.
- Upstart’s competition for best credit model can be grouped in 2 phases:
- Upstart Model vs. FICO score ( current phase)
- Upstart Model vs. AI competitors ( entering this one, between the 2 )
Phase 1: How will Upstart get an advantage over FICO ?
- AI technology improvement - AI training costs and model improvements are falling dramatically
- AI training costs are falling 50x faster than Moore’s law!! - Ark Invest
" In 2017, for example, the cost to train an image recognition network like ResNet-50 on a
public cloud was ~$1,000. In 2019, the cost dropped to ~$10, as shown below. At the current
rate of improvement, the cost should fall to $1 by the end of this year.[2] The cost of
inference—running a trained neural network in production—has dropped even more precipitously.
During the past two years, for example, the cost to classify one billion images has fallen from
$10,000 to just $0.03, as shown below.[3] " - https://ark-invest.com/articles/analyst-research/ai-training…
- This means that Upstart( and the competition) can iterate much quicker at a far lower cost
than before. This should accelerate their model performance improvements, while reducing costs - AI performance should scale well with more compute power + more data
- The more data the AI gets the better it tends to do, given it has the compute to match with
that. - Andrew NG - “AI is the new electricity” https://youtu.be/21EiKfQYZXc - Upstart model performance will likely get better the more data they get.
Phase 2: how will Upstart defend it’s model superiority vs. AI competitors over time?
- Maybe a positive network effect?
- Each bank partner that Upstart adds loan origination volume to Upstart’s network which makes
Upstart’s credit models better ( more data ) which increases the value of Upstart’s network
which makes even more bank partners join. - Additionally, there is little incentive for a bank to leave Upstart’s network unless the
competition makes an exceptionally good AI model. There would be a switching cost in terms of
degraded performance to going to a competitors solution. - More importantly: This network effect is reversed for upstart’s competitions
- The better Upstart’s models become → the more bank partners leave to join upstart’s
platform → The less data the competition’s models have → The less improvements they
are able to make → The more banks go to Upstart ( better loans) → The better
Upstart’s models become. - A death spiral for the competition
- Buying out the competition
- Upstart has the capital to buy out early stage competition
- Regulators will probably not like this.
The next post will go into some risks with these moats.