Enterprise AI: Building High-Trust Systems
Original post on Substack on June 4th, 2024: https://jamesin.substack.com/p/enterprise-ai-building-high-trust
On Building High-Trust Systems
Every AI startup is in the ring fighting for enterprise contracts. We’re at the stage of the cycle where AI adoption is inevitable. There is an understanding of where the foundational efficiencies and productivity gains can be found. And there are known entry points for startups to sell into.
Enterprise contracts are juicy. What better way for an early-stage startup to scale than to sell into a company that is already enormous? However, one thing I’ve seen is that, especially in AI, there is a real act of transaction required to make startups and enterprises speak the same language.
Large companies have enormous responsibilities. And far more prominent in an enterprise vocabulary is: compliance, risk, and liability. They are trying to avoid any fractions, not move fast, and break things.
With that in mind, we’re entering a new phase of enterprise AI adoption in which trust is the most important factor.
I’ve been thinking of it as — The Trust Primacy.
Trust is a bit of a spooky word. It’s powerful but sometimes undefined and ambiguous. It can be a feeling or a guarantee.
What I’ve been thinking about then is building out an idea of what trust means when it comes to enterprise adoption. I think there are a number of different elements that are important in different contexts. Breaking them down:
—
Trust in Integration.
Startups handling private company data and providing foundational LLM capabilities — such as search and summarization. A huge amount of startups fall into this category; it’s become table stakes to get a contract over the finish line.
Trust in Brand.
Reputation and familiarity matter. Take Brett Taylor, for example. He famously built Google Maps in a weekend, became Facebook’s CTO, founded Quip, sold Quip to Salesforce for $750M, and eventually became Salesforce’s CEO. Now, he launched Sierra, a conversational AI platform for businesses. Given his track record, enterprises are way more likely to choose his product over anyone else’s.
Trust in Systems & Data.
System and data validation is difficult. While generating, parsing, and summarizing data has become easier, it’s becoming increasingly crucial to ensure its accuracy and integrity. On the investment side, this is an area I’m particularly interested in. I see a lot of opportunity in ensuring the reliability of output, whether that be in financial reporting, regulatory compliance, or operational decision-making.
—
It’s tempting to measure the race toward becoming the next billion-dollar AI company as one driven by technical benchmarks. But I’m not sure that’s exactly how it will play out for this next phase of enterprise adoption.
In a landscape that is overflowing with amazing ideas and technical breakthroughs, I think it’ll be the companies that can provide a reassuring offer to enterprises that will be able to meld together the potential of disruption with the needs for incumbency that will ultimately win contracts and be able to scale.
If that’s right, then the benchmarks will be less about the models themselves and more about how the models are formed, who is leading the companies, and the reliability of the systems that are driving the decisions being made.
