How are companies using AI today? 10 takeaways from Index Ventures' Applied AI event in NYC

Last week, the Index Ventures team hosted a power lunch in NYC with 50 active operators focused on applied AI, to help companies navigate a fast-moving field and share best practices and pitfalls when it comes to using LLMs to either improve revenues, reduce costs, or improve user experience.

There were many fascinating topics discussed, but here are the 10 points that stood out to me:

1. Every company seems to be actively looking at ways to leverage AI as broadly as possible - most of the top NYC-based tech companies ( Datadog , Squarespace , Duolingo , Ramp , Justworks , Maven Clinic , Yext , Dataiku , etc.) were represented at the event by folks (mostly CEOs, CPOs, CTOs or VP Operations) working on live AI projects. These tech execs are pragmatists though: they are not looking at AI to save the world but to improve the efficiency of their business, grow revenues or offer a better customer experience.
2. The past few months have been spent narrowing down use cases from 10-15 potential ones to the 2-3 major ones moving the needle on 1 well defined metric (revenue per visit for a consumer company for example).
3. Use cases with most traction right now include: 1. Copilot (although still only typically adopted by a large minority of developers, but who can be up to 2x more productive than the rest of the team), 2. Marketing automation (press releases, emand ail marketing, content creation, etc.), 3. Customer support (estimates that up to 50% of conversations that required a human can be fully automated), and 4. Internal knowledge management.
4. OpenAI is ahead of the pack and part of most conversations but this is a narrow lead and we heard of instances of companies going for their cloud provider’s embedded AI offering (GCP for instance) for the sake of simplicity and cost (bundled into a discounted enterprise contract) - having the highest-performing model is not critical as long as the offered one is good enough, which seems to indicate a rapid commoditisation of LLMs for the most common use cases.
5. In general, incumbents with existing customer relationships, established distribution channels and large proprietary datasets are well positioned to leverage AI to become even stronger: early signs point to AI being currently an enabling technology rather than a disruptive one.
6. The companies which are moving the fastest (eg. Notion) have buy-in from the most senior execs (the cofounders in this case) who are taking part in hackathons and using LLMs themselves.
7. The first step is to offer AI features as add-ons to the core product. The second is to embed AI into every part of the core product where it makes sense. The third may be to rebuild the entire product around AI - but we are a few months/years away from it.
8. This has an implication for team structures: what generally begins as a commando-style “AI team” is already starting to get embedded into every tribe - first with one dedicated AI person, but eventually just becoming a part of every engineer and designer’s tool kit.
9. New entrants are finding success with specific vertical use cases and workflow built on unique datasets. This inspired my partner Paris Heymann to write a blog post about the potential for Vertical AI: https://www.indexventures.com/perspectives/the-rise-of-vertical-ai/. If you are building a Vertical AI company, please get in touch!
10. Consequently, data quality is critical: several participants were focused on data labelling and fine-tuning models to improve their performance for their specific use case.

Thank you to all the participants and to Martin Brodbeck from Priceline , Angela McNeal , Alexis L. from Datadog , George Sivulka from Hebbia , John Littzi from Coalition, Inc. , Vinay Hiremath from Loom , Jonathan Rosenbluth from Cohere , Linus L. from Notion and Benjamin Sesser from BrightHire for sharing your insights.