September 29, 2024
Gen AI for Finance: Expectations vs Reality

Expectations vs. Reality: Gen AI for Financial Services

We've spoken now to thousands of investment professionals at leading private equity, crossover, investment banking, and private credit firms. One common theme has become clear: the entire category of Gen AI for financial services has fallen short of expectations. Products have exquisite demos, flashy launches, and buzzy TechCrunch articles, but when you get the product in your hands, the platform breaks and doesn’t give you an output anywhere close to the quality that you expect.

The confusion is exacerbated by the fact investment firms frequently exaggerate the capabilities of their in-house tools and their AI efforts to portray themselves as tech-forward and differentiated. Each time senior leadership at investment firms hear about how their peers are leveraging AI, they often walk away feeling “behind the curve”. Then, out of defensiveness, they form an AI committee and push their teams to either buy a tool or go on the perilous route of building in-house (see our previous article about in-house software building)

However, many investment professionals' expectations for what these tools can do (i.e. reason across large sets of data and drive new insights) are far beyond what is available in the market today. Gen AI for finance is a category that is in its infancy. Most companies are in its first year or two of production, and products that call themselves “agents” are most certainly not agents (in IB/PE land anyways). The financial services usage contrasts starkly vs. leading horizontal companies in such as Glean, which have high NPS and meaningful adoption among its user base (which have a heavy focus on engineering / customer services use cases).

There is no “Gen AI for finance” company that has found product market fit today, nor is there a player that has meaningful engagement among its users.

If you’ve found a player in finance with true PMF, shout – because we certainly haven’t seen it. Tell me one company that you like, and we will find you a dozen disgruntled customers in various PE firms / investment banks who do not use that product. We know because we meticulously track over 30 of our competitors and their product progression, regularly speaking with their customers and design partners across private equity, crossover funds, and investment banking. See footnote of the investor tool landscape page below for a broader list.

This sentiment is shared across the board, and to an outsider, it may not be clear why there is no PMF in financial services, while horizontal players such as Glean are crushing it in every regard.

Reasons outlined below:

There is no one singular burning problem in financial services today
  • Financial services professionals are unbelievably well-catered to as a demographic. Everyone is trying to sell them something, from expert calls to alt data and sourcing software – as a result, all the core problems of the role have at least a B quality solution.
  • These roles are also unbelievably varied in terms of work output. If you asked me what I did in private equity, it would change by the week.
  • One week might all be portfolio company work, and the next might be blown up by a live process – the number of tasks and things that I worked on was too much to distill into a singular problem statement.
Tolerance for mistakes is low and cost of mistakes are extremely high
  • Mistakes are expensive, and someone’s reputation is always on the line, particularly when it comes to live deal execution
  • AI diligence tools have no utility in core investment committee use cases until they reach a certain degree of accuracy – usefulness is binary. A similar analog is self-driving cars – one death with self-driving cars is one death too many, even if they are categorically safer than human drivers
AI usefulness is binary in financial services
  • Joining a private equity firm and hoping to outsource due diligence is like joining a sales team and hoping to outsource calls with customers – this is the essence of the job itself
  • This is exacerbated by the fact that LLMS are probabilistic and generally not great at math (except for o-1, which has its own set of issues in production)
  • In a separate article, we'll explain our view on automating due-diligence, and why we think this may not be a solvable problem
Investment professionals are creatures of habit who require immediate time-to-value
  • I was an investment professional only a year ago, and dozens of our friends still work in financial services - if a product does not have immediate time-to-value (i.e. I can figure out how it works immediately), I won't use it unless my partner/principal asks me to
    • One example of a tool with immediate time-to-value is BAMSEC – you can just pull filing links and source them within seconds
  • Separately, if this product does not integrate immediately into my existing workflow, I’m probably not going to pick it up unless it truly has some extraordinary value that I can’t find elsewhere
    • However, there are products that provide such exceptional value that people just use them all the time despite a learning curve. These are companies such as Alphasense and Tegus
Broader category of "software for investment professionals"

Looking outside of Gen AI, however, there are several core software companies with PMF / adoption among investment professionals. There are deeply entrenched players like Bloomberg or Factset that are deeply integrated into workflows, and new age leading players such as Alphasense / Tegus that provide valuable insights and are generally have high NPS.

There are distinct buckets with respect to the evolution, scale, and growth of these different types of software. In order of descending scale and ascending growth, the categories are:

  • Legacy data providers – Bloomberg, Factset, Cap IQ, Refinitiv Eikon
  • Private data / services providers – GLG, Alphasense, Alphasights, Tegus
  • Horizontal gen AI platforms – Glean, Sana, Qatalog (Not necessarily the most relevant comps as they generally do not focus on investment firms, but on occasion may sell to them)
  • Vertical Gen AI for financial services – All the new age companies that have come out in the last few years

We provide some details on the broader categories below:

Note: This includes only front-office software companies and excludes back/middle office software such as fund admin and IR software. Also excludes sourcing / CRM software that sell primarily to venture / growth firms such as Affinity, Synaptic, Harmonic, etc. Pages above represent simplified view without details on specific Gen AI for finance providers, given highly sensitive nature of product details and customer feedback.

Including key punchlines below:

Building successful investment software takes a VERY long time.
  • The most mature market, legacy data providers, account for >$20B+ revs with high PMF, growing at LSD annually.
  • However, it took them 30- 50 years to get there!
Horizontal Gen AI is more mature than finance
  • Horizontal Gen AI players companies such as Glean are loved by users, particularly in engineering / customer support
  • Leading companies in this category are scaled ($50M+ ARR), with best-in-class net revenue retention (150%+), high engagement (with active users using products multiple times per day), and superior logo retention vs. vertical players in finance (90%+)
Gen AI for financial services is nascent, it is the most rapidly growing category
  • Total market spend is <$20M today, but this market had $0 of revenue 2021
  • Low-PMF category and no leader in terms of adoption – though there are leaders in terms of GTM / marketing

Conclusion

Today, Gen AI finance tools are more similar than they are different. They succeed in the same ways – flashy demo, high-brow marketing, text summarization- but also fail in the same ways – complex reasoning, quantitative work, and generation of net-new insights. The limitations to the underlying models make it difficult for companies to build highly value-added products.

While the adoption curve is slow, when these gen AI tools become truly useful, we believe they have the potential to transform the way financial services firms do business. When they do find PMF, the payoff will be immense. The players that are truly entrenched into workflows, such as Bloomberg or Factset, are billions of dollars in revenue and some of the stickiest products around. If Microsoft Excel doubled prices for investment firms only, you would see exactly no churn. We don’t have the engagement metrics for Microsoft Excel for investment professionals, but we can infer based off our own days on private equity / investment banking days that it’s probably got a higher DAU / MAU ratio than any other business software ever.

Given these inherent challenges (variable workflows, bespoke nature, high-fidelity reasoning / quantitative workstreams), we believe any attempt to automate an entire workflow end-to-end is doomed for failure with today’s technology.

This is why we focus on i) pre-final outputs (think outside-in analysis, pre-meeting reads) and ii) simple workflows that barely require AI (think recreating pages, data extraction, etc.). For simple use cases that are not life or death, an LLM can put together a decent 80/20 cut on written text that does not include math, and for other defined use cases, regular software can solve much of the problem without the use of AI. This will certainly change as models improve to be capable of reasoning, and we have these use cases teed up, but not currently in production.

In conclusion, while the landscape of AI tools for investment professionals may fall short of expectations today, we believe there is immense potential in the category. As AI models continue to evolve, we remain committed to partnering with leading providers and helping our clients to develop solutions that truly enhance their workflows in financial services. The journey to create the investment software of the future is undoubtedly challenging, but we believe the payoff – in building the operating system of the modern investment firm - will be well worth the effort.

If you have any questions, or have any technically challenging problems at your investment firm, please reach out to us directly and we’d be happy to share our learnings or have an intro discussion with your team. For more insights like this, please subscribe below.

You can reach us at info@prosights.co

Subscribe to our insights!

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.