Since the start of 2024, we have spoken with nearly 250 leaders in the private equity industry with the aim of understanding customer needs in the finance AI solution space. We soon realized that buyers were finding it challenging to navigate the market landscape at all, given the overwhelming number of providers and different ways to build solutions across the stack.

With AI, scoping a solution is particularly hard, because the market itself is still emerging and re-shaping itself month by month. We asked ourselves, “what solution set are PE investors considering when making the decision of whether to adopt AI?” We also investigated whether decision-makers were actually weighing whether to invest in proprietary model development against adopting point solutions, or other parts of the stack.

Ultimately, these challenging questions are the driving force behind this article series. Our goal is to help you, as a business leader, begin to understand some of the dynamics connecting providers and buyers in this newly shaped landscape and make more informed decisions.

What we learned

Last week, we introduced the concept of the AI stack. The “stack” refers to the set of software and hardware components needed to run the AI applications coming to market. What is somewhat unique about the AI space, is that buyers show a much higher interest in adopting new tech in layers further down the stack, such as LLMs or data pipelines. This stands in contrast to other complex technology markets such as cloud computing, enterprise software or networking, where all but the largest enterprises typically start by using out-of-the-box or narrowly customizable applications.

Adding to the already complex landscape of vertical AI solutions (sourcing, due diligence, portfolio management, etc.), firms are considering where in the stack to invest in order to support different functions, which has sometimes made adopting new technology paralyzingly difficult. For example, we spoke with multiple private equity firms who were considering building out their own AI-powered data pipelines and algorithms to help source new deals from data sets they had exposure to. In our last article, we shared examples of large funds who are actually using RLHF or fine-tuning to develop their own proprietary LLMs. According to Forbes, "KKR and Blackstone are leveraging AI to analyze market trends, evaluate potential investments, and enhance decision-making processes".

This tells us that clearer lines are indeed beginning to emerge between the two major markets. It’s also worth emphasizing that many firms, especially as we move up-market and model training becomes more accessible, are eyeing both ends of the spectrum.

However, even this does not capture the full picture, as some firms will choose to work with consultants, while those consultants must also source their LLMs and data infrastructure from existing providers. Investors who do look further down the stack to build custom solutions will have many more decisions to make, from which models to use, to the integration between existing systems. A few sample components:

1. Point solutions (applications)
2. Off-the-shelf models

For example, from CB Insights:

[GPT-4] is among the largest and most costly AI models available. Using it to summarize an email is like getting a Lamborghini to deliver a pizza.” – Tom Dotan, WSJ. Models are getting smaller, making them faster to custom train and less costly to call via API, according to CB Insights. Performance gaps are also closing rapidly. For example, Arcee.ai’s models perform better than ChatGPT for specific use cases.

1. Future offerings could include models such as BloombergGPT, which seems to perform well in tests, but has yet to be meaningfully commercialized.
2. Hybrid: Companies like Alpaca and AlpOps combine elements of both LLMs and vertical applications, offering comprehensive platforms that encompass various functionalities like data analysis, deal sourcing, and portfolio management.

This piecemeal approach is much more costly, and firms need to seriously weigh whether this is the best way to make a first foray into AI.

Where do we go from here?

There is an increasingly popular thesis that custom LLMs will get “squeezed” in the middle, with the top and bottom layers of the stack (applications and compute, respectively) delivering most of the value as models become commoditized. This alone isn’t necessarily a reason to avoid working with LLM developers, but companies hoping to develop a significant moat with large language models alone may end up finding that most of a value ends up coming from a smaller handful of user applications that deliver most of the value, or some better access to data or compute than a smaller competitor.

Of the hundreds of senior PE leaders we surveyed, primarily focused on the middle market, at least two thirds of professionals we spoke with told us they were considering purchasing or already using at least one AI application, whether a custom instance of ChatGPT Pro, or a dedicated due diligence system like the one we are building at Keye.

Because applications are where we see the most widespread customer adoption and the highest proliferation of startups, this is where the remainder of our deep dive will focus.

The Run-Down

We have previously made the case that most finance firms can and should adopt AI in some form over the next 2-3 years. Because the space is so early, it is highly possible that at this early stage, your firm might not end up partnering with the provider that will eventually power its AI usage for years to come. But as we wrote about in our second article, sometimes making no decision at all is more costly than making a wrong decision. Simply getting their feet wet, many readers have told us, has brought them immeasurable insights that they are using to learn, iterate and start to build a long-term moat.

In service of that goal, we will wrap up with a deeper rundown of each function and some of the players mentioned in our market map at the top of the article:

Deal Sourcing: AI is revolutionizing deal sourcing by using predictive analytics to identify investment opportunities based on market trends and historical data. Machine learning models can scan and analyze vast amounts of data to recommend deals that align with strategic investment criteria, thus enhancing the efficiency and scope of sourcing potential investments.

Companies like Cyndx and EQT Group's "Motherbrain" are leading the way. Motherbrain, for example, integrates machine learning and big data to grow the digital footprints investors can analyze. This tool has successfully sourced multiple high-value deals and is considered a critical asset within EQT's investment workflow.

Due Diligence: In due diligence, AI is accelerating the analysis of financial statements, legal documents, and other pertinent data to quickly assess the health and potential risks associated with an investment. AI-driven tools are highlighting anomalies, predicting future performance based on historical data, and providing comprehensive risk assessments, thereby speeding up the due diligence process and reducing human error.

Keye, for example, uses AI to enrich the due diligence process, focusing on quickly extracting insights and identifying risks from large sets of documents. The ultimate goal is to free up time so that investors can consider deals more carefully, and compete with larger firms on a wider set of deals.

Co-Pilots & Automators: AI co-pilots and automation tools can streamline repetitive tasks such as data entry, reporting, and even some aspects of decision-making in the investment process. By automating these tasks, AI allows PE firms to focus on strategy and decision-making, improving operational efficiency and reducing costs.

Many in this category take a broader approach, such as Blueflame AI. The platform integrates AI and Large Language Models to streamline workflows, manage unstructured data, and assist in crucial tasks like compliance and investor relations.

Knowledge & Research: AI can dramatically improve the efficiency and depth of market research and knowledge management. By aggregating and analyzing data from diverse sources, AI-driven tools provide actionable insights and deep analytics, enabling investors to stay ahead of market trends and base their decisions on comprehensive intelligence.

Another horizontal solution, Hebbia's technology is distinguished by its ability to execute comprehensive searches over large datasets and provide clear, actionable insights, positioning it as a powerful assistant for knowledge workers.

Financial Planning, Auditing & Accounting: AI enhances financial planning by providing more accurate forecasts and modeling scenarios. AI algorithms can analyze market conditions, portfolio performance, and economic indicators to offer real-time financial insights and predictive analytics, aiding in more informed strategic planning and resource allocation.

Other functions - data cleaning, portfolio management and operations: Though often not specific to PE, supporting functions that make data analysis easier are an important part of the application stack to consider. While our analysis focuses primarily on investor tools, we are keeping a close eye on the way in which AI will be deployed to transform PortCo operations and boost ROI.

How to pick the right partner

When considering the adoption of AI-driven solutions in investing functions, firms should strategically prioritize areas where AI can deliver the most significant impact and ROI. Here’s a high-level approach that we have seen work well:

1. Identify pain points and opportunities: Start by assessing which functions are most time-consuming, costly, or prone to errors.
2. Evaluate AI readiness: Assess your current technological infrastructure and the data quality at your firm. You might need to invest in data management and IT infrastructure improvements before they can fully leverage AI capabilities, but you would be surprised at the way in which some applications can help you right out of the gate.
3. Consider compliance and integration challenges: Solutions that offer seamless integration with your existing workflow will cause minimal disruption.
4. Define clear objectives and metrics for success: Before implementing AI solutions, work with your provider to define what success looks like.
5. Have a partnership mentality: Providers should show expertise in the PE sector and a track record with similar clients. A team that is willing to understand your core needs will beat an off-the-shelf solution any time.
6. Start with pilot projects: Implement AI solutions in phases, starting with pilot projects or design partnerships. This approach allows you to manage risks and tweak strategies based on initial outcomes. Successful pilots can then be scaled across other areas of the firm.

The AI landscape is vast and ever-changing, but we hope this deeper dive into the space gives you an initial direction to navigate the market. If we can further help you down that path, please reach out to us at Keye and our founders will respond to you directly: founders@keye.co.