Three Questions Every Enterprise AI Startup Should Ask

Nearly three years ago, I wrote a column based on a thought-provoking question a founder asked me at an entrepreneur event: Is there a metric for deciding when to fund a Series A round?

Looking back, the response I offered then still seems valid: Is the founder mission-driven? Can the idea become a superior product with large demand? Can the company grow into attractive metrics at scale?

Today, however, I believe a more pressing question has emerged for both entrepreneurs and investors as AI has taken center stage: What are the key considerations of funding early-stage enterprise software companies in the AI era?

This question has become more important as the venture capital community increasingly seeks out promising ideas at earlier stages, often before a single line of code has been written and long before any revenue has been generated. Our recently launched Greylock Edge, a bespoke company building program, has generated an extraordinary response from ambitious engineers and entrepreneurs at the seed, pre-seed, and even pre-idea stage.

The potential of enterprise AI has added a welcome dimension to these conversations. In earlier times, many software founders feared seeking funding “too early.” They assumed that, absent a beta product or early revenue, they weren't ready to seek an alliance with seasoned venture investors. The AI gold rush has changed that.

When seed-level founders meet with us now, it is to test ideas, brainstorm use cases, engage customers with our help, and envision product platforms from the start. In my discussions with them, I suggest focusing on three vital questions:

1. What is the highest value customer use case you can solve?
2. Is there proprietary data that you can access or develop to create a moat?
3. Can you insert your product alongside incumbent vendors?

Develop the Highest Value Use Case

Unlike earlier eras of tech innovation, AI has seen the largest software vendors quickly respond and launch significant initiatives. In an increasingly crowded field, the next generation of AI founders have to ask themselves: What fundamental use case are we enabling that is not already covered by the incumbents?

While it might be daunting to develop a product where a Microsoft or a ServiceNow already have an installed base, large enterprises still welcome new startup products that enable high value use cases not previously possible or existing high value use cases that can be solved significantly better or much faster or cheaper. Slack did this to Outlook’s email. Palo Alto Networks entered the market when Cisco and Check Point Software had the leading firewall software. MongoDB inserted itself into a market historically dominated by Microsoft and Oracle.

A good example today is how start-up Cresta saw an opportunity to enhance sales results without replacing the existing sales infrastructure or the salesperson. Cresta’s management platform uses AI as a coach that assists sales teams and call centers. Its software provides real-time behavioral coaching to improve soft skills, generates response suggestions to product questions, and offers insights into performance and customer trends. Performance metrics of sales professionals at CarMax, Cox, Intuit and elsewhere are demonstrably higher when work is done in collaboration with the Cresta AI coach. Cresta’s growing business provides a case study of how a new AI startup can quickly win enterprise acceptance by driving superior results on high-value use cases.

Build a Data Moat

A broad consensus has emerged that access to a proprietary pool of data can give an AI software business a competitive advantage. Alex Ratner, the founder of Snorkel, recently argued in Forbes, that for some companies, using an off-the-shelf LLM is not much of an advantage if a competitor is using the same tool.

He is right. Most large enterprises are sitting on pools of data that contain unique, high-value data that has yet to be leveraged in a meaningful way. This is true across industries including health care, financial services, consumer packaged goods, retail, and manufacturing. These companies are seeking ways to create value from their proprietary data that goes beyond just reporting and basic analytics. They want to train AI to take unstructured data to make connections, automate business-relevant tasks, distill insights, build accurate models, and accurately predict future outcomes for business advantage. The specialist LLMs and other AI models that emerge from this work will create a moat against competitors.

Seed and pre-seed companies are unlikely to have proprietary data at their founding stage. Yet the pursuit of unique data should be a component of their strategy. Initially, they may be building breakthrough tools that enable a larger organization to harvest exclusive data sets unavailable to the public.

The longer-term vision for a seed or pre-seed company could potentially include discovering a way of retaining the rights to data their model trains on. They also may plan to develop insights or metadata that would be useful to their first customers and have broader industry applications.

Abnormal Security is an example of how a data moat can gradually be developed by a startup. The company did not start with its own data set, but its software was designed to ingest diverse existing signals and telemetry. That allowed them to develop a proprietary database which they have leveraged to build baselines across users, cloud email, and collaboration applications. Today they marry this data with advanced AI to precisely detect anomalous behavior and automate remediations for over 1,400 companies.

Start with an Insertion Strategy

When we talk with early AI product builders, we are realistic about the fact that few large enterprises will instantly “rip and replace” incumbent platforms, especially for newer AI vendors with incomplete and unproven offerings. Given this, we encourage early-stage entrepreneurs to formulate a clear insertion strategy in which their newly developed software can be easily plugged into or sit alongside existing incumbents.

It’s hard to overstate the importance of getting the insertion strategy right. An early-stage AI product should be deliberately designed to offer a key advance within a business-relevant customer use case. Importantly, it should be able to quickly demonstrate the new capability in a low-friction way, without creating new risk for the CTO or technology team and the forward-looking champions in the enterprise.

Once a startup establishes a beachhead and can deliver growing value, it can then earn the legitimacy to expand capabilities. If done right, this can lead to a growing platform, which might eventually replace the incumbent. The AI seed companies that build large franchises over time are those that eventually develop platform strategies (and not just narrow niche feature modules).

Rubrik, which recently celebrated its 10th anniversary, began with an ambitious vision and the realization that they had to be very deliberate about how to compete with entrenched incumbents. The company began with a narrow offering around backup and recovery exclusively for VMware. This was their insertion strategy. The reputation and trust they developed positioned them to expand to a growing set of enterprise workloads across hybrid cloud. Their growing platform (with applied AI) has evolved into a market-leading solution for ransomware and enterprise data resiliency. Today the company serves 5,500+ customers, with a business that is in the high hundreds of millions of ARR.

The Best Mindset for AI Startup Entrepreneurs

As we progress into 2024, we will see an explosion of new enterprise AI startups - across foundation models, enabling infrastructure and intelligent applications. We are still very early in imagining or being able to predict what may lie ahead. What is certain is that the appetite for enterprise AI solutions is real, and with the potential to disrupt and eventually transform all of enterprise information technology. The startup entrepreneurs who bring the right mindset – and are willing to ask the right questions – will be best positioned to win in a rapidly evolving market.

(Disclosure: Greylock is an investor in Abnormal Security, Cresta, Rubrik, Snorkel, and was an investor in Palo Alto Networks.)