We are used to trending technologies being included in pitch decks and technical roadmaps. Sometimes these are architectural trends like micro services, sometimes technologies like blockchain and machine learning and of course artificial intelligence.
It is very likely that we will start seeing features such as "advanced conversational technology" and "in-house trained models" increasingly being mentioned in the context of the platforms we are evaluating. These types of technology will in many cases be based on large language models, such as those underlying ChatGPT and Google Bard.
How should an investor think about this ? What is the value of using "artificial intelligence" in a platform ?
From a technical due diligence perspective, we are interested in identifying the risks associated with the platform, but also in identifying the intellectual property that the platform represents. Beyond the product, we like to understand to what extent the software development life cycle relies on AI powered tools.
Let's start with the intellectual property, or IP. Integrating a pre-built AI model into your platform using APIs is of course not the same as developing the model. And building a base Large Language Model or LLM is not quite the same as fine tuning an existing LLM. These cases lie on a spectrum of skill and knowledge. Our first task in the due diligence is to understand where the target business is on this spectrum. We will ask questions such as:
Describe your ML Ops processes
How do you obtain data to train your models ?
Do you rely on any third party service providers for any platform features ?
The answers to these questions will help us understand where your business is positioned.
Although integrating an API into your platform (such as the ones available from OpenAI and others) requires skill and can add remarkable features to your platform, it is again not the same as developing large language models. Similarly - if the chat bot you use is a customised integration with a third party API, that is a far cry from owning the IP that enables the chat bot's conversational abilities.
On to the risks. There are a number of areas to consider, for example:
Ethical use of AI - for example, how do you ensure that your offering does not expose your customers to bias, and how do you ensure their privacy ?
What is your ability to obtain data to train your models, and how expensive is this training process ?
Are you able to attract and retain talent in this domain ?
The extent of the risk could vary. If you provide transaction fraud detection or anti money laundering services for fintechs like lenders or payment gateways, you would be entirely remiss not to have a roadmap towards adopting AI - rule based and expert systems won't cut it any more.
In terms of e-commerce and other B2C platforms, using AI for recommendation engines and customer interactions will become table stakes - it will not represent any kind of a competitive advantage anymore, rather it will be needed just to stay relevant.
A significant risk of AI to business is likely to be that of not adopting any machine learning methods and algorithms.
The third aspect is tooling - the capabilities of AI assistants such as Github's Copilot brings potential productivity enhancements to the team. If you are using it - what gains have you seen ? If you are not - why aren't you ? The same questions apply for other aspects of operations - security, monitoring and more.
In summary - we consider the use of AI in the same context that we consider the use of other technologies: what are the associated risks, what value does it add, and who owns the intellectual property. Additionally, AI can (and perhaps should) be used to increase the development team's capacity and productivity.
The one important difference between AI and other "trends" is that we think AI is a fundamental technology - it will be used as a de facto part of the tool set in the majority of software platforms.
Coda: this post was written without AI assistance. The header image, however, was NOT created without AI assistance. It is based on a prompt generated by an LLM and fed to a diffusion model.
Rosewood has performed technical assessments on a number of businesses that use machine learning and artificial intelligence in different ways: businesses whose core products are chat bots, other that provide anti money laundering and fraud detection, some that train proprietary models in the fields of engineering and medicine, and more. Our experience allows us to evaluate the impact of these technologies on business' platforms and products. Contact us for assistance with your technical due diligence needs.