Initial thoughts on the disruptive potential of generative AI
By Eugene Lee and Julianna Vitolo
Why Enterprise AI?
You may have seen our recent posts on social media (i.e. here and here), showcasing our desire to be open about our learnings. Given the monumental nature of AI and rapid innovation, we wanted to be honest about our insights and early thoughts as they develop. Like the rest of the industry, we are students of innovation. So, we’re making it our objective to study new developments, new ideas, and early signs of traction to help shape our opinions on where tech is headed. We see no benefit in holding any early thoughts or insights about the market to ourselves. We believe the more willing we are to share our ideas with the public, the more willing they might be to share some thoughts back with us. And naturally, we might royally embarrass ourselves in the process, by having our theses publicly proved wrong – but the founders proving us wrong are the ones we want to meet.
Our discovery process has, of course, led us to dig deeply into AI. While VC blogs have recently been inundated with the topic, this is a case where consensus is aggregating around a topic for good reason; when it comes to AI, we think there is massive potential to completely reinvent traditional workflows within the enterprise. What wasn’t possible yesterday, might be possible tomorrow. This will require some level of change management — we need to change how people work to reap its benefits. A knowledge work revolution is looming, but we’re still peering over the cliff into a sea of uncertainty.
The Knowledge Worker Revolution
Some AI solutions will directly replicate, and therefore replace, traditional roles within the workplace. The first wave of this replacement will focus on the more methodical, formulaic, and often repetitive roles of entry level workers. Example professions include bankers/analysts, lawyers, customer service representatives, SDRs, recruiters, sourcers, etc. This is the lowest-hanging fruit, given the primary function of junior workers is to dig through data, perform a routinized task, and then summarize key insights for more senior level team members – a process we know LLMs are skilled at. In general, we can expect AI to transform the bulk of manual, outdated processes into programmable systems.
This promises greater efficiency at the foundational level of an organization, which should create a bottoms-up momentum driving improved performance throughout. Benefits will most likely be lower overall costs and time savings for the business. For workers, it's natural to fear eventual replacement, but the higher-level, strategy-oriented, creative and thought-provoking tasks where human minds shine will continue to be done by humans. As in previous societal shifts driven by technology (the industrial revolution for example), the immediate fear likely underestimates our adaptability. It’s possible that roles of the future will have less monotonous data entry and time-consuming workflows. They may instead revolve around more of the high-impact reasoning, idea generation, and execution-oriented tasks.
However, the concept of replacing human-held roles entirely with AI – a technology that has notoriously unclear data privacy standards, non-deterministic outcomes, and trust and security concerns – is a bit ambitious in the near-term. The pace of technological development is moving at warp speed, but implementation within the enterprise still lags some of the more heady expectations. While foundational models are immensely powerful, the productization of that power at scale remains theoretical.
How Do We Get to Mass Adoption?
Widespread adoption will likely be a gradual process and will require keeping the human in the loop at first. This is evident in the proliferation of co-pilots – AI that augments traditional knowledge worker roles rather than replacing them. With this implementation, both sides of the equation benefit: employees get comfortable completing their work with the helpful assistance of AI, gradually learning how to integrate it into their existing workflows and maximizing its benefits. The AI meanwhile ingests ever more data and learns how to automate human-centric workflows, using ML to improve them.
We’ve had many conversations with enterprise buyers across functions to glean a better understanding of where they are on the AI adoption curve. Very few AI startups have unlocked the keys to enterprise buyer budgets, whether it’s the finance, marketing, sales, product, HR, or engineering team. Surprisingly, while most companies are very interested in exploring new AI solutions, only a few are actually paying for tools that have a meaningful impact on their day-to-day activities. It’s not practical to adopt AI at any significant scale in many of these organizations just yet. While they use ChatGPT, Claude, Perplexity, and other general purpose tools to speed up their personal workflows, (i) ease of implementation, (ii) trust and data privacy, and (iii) a tangible proof of value (ROI) are all hurdles yet to be overcome.
On the implementation front, it’s difficult for buyers to champion new products internally if they can’t determine the quality of the output. The other huge unknown is the cost of that implementation. As companies experiment with volume-based pricing models, their customers are wary of higher-than-expected charges from AI vendors. With the price of inference and compute still high, any excess volume comes at a high cost to customers, and this problem is only amplified as companies scale. Enterprises also encounter the age-old issue of wondering if the startup pitching them can scale along with them.
When it comes to trust and safety, there is still a lot of uncertainty. Many internal and external workflows can only entertain a margin of error of ~0.01%, and AI tools simply cannot deliver that consistently yet. There is also substantial reputational risk associated with customer-facing products that fail even a small fraction of the time. Take the Chevrolet Watsonville chatbot anecdote as an example. The dealership was using a chatbot for customer service inquiries, but it wasn’t ready for prime time. The bot failed to protect Chevy’s bottom line, suffering from a lack of guardrails and bouts of hallucination that allowed one customer to buy a Chevy Tahoe for $1, and another to buy 2 vehicles for the price of 1 [source: VentureBeat] .This is just one example of why many companies are hesitant to deploy external AI solutions and have primarily been using internal tools that don’t carry the same risk profile. As of now, there is no way to trust that AI solutions will be reliable in the ways companies need them to be.
Lastly, the elusive nature of ROI makes it difficult for businesses to purchase new AI solutions when the return might not be immediately quantifiable. Efficiency for the sake of efficiency is great, but it doesn’t cut it for enterprise teams that are pushed to drive greater profits and optimize costs. Leaders are also trying to assess whether building new technology will generate a competitive advantage in their industry. If they estimate that it will result in a long-term advantage (or even, in some cases, a distinction that may not even be an advantage), then they’d rather build that technology in-house. If the solution doesn’t align with management’s go-forward strategy to build, or if it’s too onerous to build, then those are the products they’d prefer to purchase. For non-core technology, estimates show that more than three quarters of enterprises have demonstrated willingness to buy AI tools rather than build them.
We’re left in an interesting predicament. The potential power of the technology is clear, but the market simply isn’t ready for it yet. Above, we’ve summarized the key blockers we need to dismantle in order to open the floodgates for AI adoption. By taking the people doing these jobs today along for the ride, transparently showing them how the technology works and proving that it’s both accurate and reliable, companies can build market trust. And at that point, we’ll be off to the races. There’s no question that AI is powerful and can have a profound impact on modern organizations; the question is when we'll see it implemented on a mass scale. Breaking down these barriers in the buyer’s mind will force open the corporate coffers, and then it’s a matter of getting your product into as many hands as possible. May the best distribution strategy win.
