Neeva Combines AI and Search – Now Comes The Hard Part

The Very Hardest Thing

What’s the hardest thing you could do as a tech-driven startup? I’ve been asked that question a few times over the years, and my immediate answer is always the same:  Trying to beat Google in search. A few have tried – DuckDuckGo has built itself a sizable niche business, and there’s always Bing, thought it’s stuck at less than ten percent of Google’s market (and Microsoft isn’t exactly a startup.) But it’s damn hard to find venture money for a company whose mission is to disrupt the multi-hundred billion dollar search market – and for good reason. Google is just too damn well positioned, and if Microsoft can’t unseat them, how the hell could a small team of upstarts?

But despite what might be the harshest environment in a generation for moon-shot startups, one relatively obscure company has been unwaveringly focused on beating Google for the past four years. It’s called Neeva, and with nearly $100 million in funding, it’s a serious contender. Neeva’s founders have decades of experience running key parts of the Google empire (Google Ads and Youtube monetization). They left in 2018 to pursue a mission to free search from what the company calls “Corporate Influence” – in other words, the very business models they worked on for years.

To that end, Neeva launched in June of 2021 with a freemium subscription model – you can access a limited version of it for free, but roughly $50 a year buys you a clean, uncompromising search engine that delivers results unburdened by the data-drenched compromises inherent in surveillance capitalism.

But while I’ve quietly cheered for Neeva these past 18 months, I’ll admit I’ve not become a subscriber. I want them to succeed, but I was too invested in my web of Google entanglements – Mail, Docs, Chrome, etc. – to make the switch. Plus, I know how hard it is to build a subscription business – asking people to change long held habits is a very, very tough slog. Neeva’s founders, technology, and philosophy were noble and commendable, but the chances of it actually breaking through seemed … slim. Not only must the company build and maintain a better search engine – an incredibly hard (and expensive) feat – it also has to completely change the business model that drives search, from an ads-driven model to one based on all of us paying for something we’ve gotten for free for over three decades.

The company isn’t just trying to boil the ocean. It’s trying to boil two of them.

Enter AI

THEN AGAIN, isn’t that the point of ambitious startups – to try to do things that otherwise would be impossible? Shifts in consumer behavior like the one Neeva is looking to spark do happen, if rarely. It helps if there’s a massive disruption in an established, highly profitable market – a disruption in the form of a product so superior that consumers can’t help but change their habits – and pay for the privilege of doing so. The timing of these disruptions is incredibly hard to predict. They often build for years, then break through all at once. First come incremental improvements in underlying technologies, slow accretions of novel consumer habits, and the accumulated institutional arrogance endemic to market leaders who enjoy seemingly immutable profit models. The iPhone in 2007 is one example – a huge market was already using cell phones, but long-simmering improvements in chip technology and internet services combined with new consumer behaviors (remember the iPod?) delivered a game changer for Apple.

After speaking with Neeva co-founder Sridhar Ramaswamy this morning (check his and his co-founder’s bios), I’m starting to think we could be on the brink of a similar shift in search. Thanks to the rise of multiple large language models like Anthropic and OpenAI’s GPT, Google and Microsoft’s diamond-encrusted business model handcuffs*, and the investment energy building behind AI-driven startups overall, Neeva just might get its shot at breaking through in 2023. And if it does break through, today will mark the firing gun, because today Neeva announced its NeevaAI service – essentially integrating a ChatGPT-like feature into its core search platform.

If you’ve been reading me these past two weeks, you’ll already know I’m excited about the impact of AI on search. I think it’s possible search might be close to the kind of fundamental interface and business model shift we saw in personal computing back in the 1980s – from command line to graphical user  interfaces, or GUIs. You also know I predicted that both Google and Microsoft would integrate conversational AI into their products this year. What I missed was the idea that an independent company might beat them both to the punch. But that’s what Neeva has done with NeevaAI. If you sign up for the service (you can use it for free for three months), Neeva will now summarize what its systems believe to be the most relevant answer to your query in a chat-like experience at the very top of the results.

Today, for example, I asked it the very same search I asked Google earlier this morning: “how to remove rear wheel specialized bike through axel.” I was struggling with my bike and years of habit taught me that there was likely a video buried somewhere on YouTube that would show me what I was doing wrong. Indeed, when I first entered that search into Google, I was presented with Google’s best guess at the answer: A YouTube video of a guy taking the rear wheel off a bike. I watched for about 30 seconds, but as soon as I saw he had a different axel mechanism than I did, I had to go back and modify my search. I did this four or five times until, by combining various snippets of information gained over those half-dozen searches, I finally figured out how to take the damn thing off.

That process I went through – iterative, combinatorial deduction, I suppose you could call it – is innately human, and it’s also frustrating as hell. I knew if I had my bike mechanic in front of me, he’d have shown me what to do with one withering flick of his wrist. By combining some nifty, highly technical approaches to information science (Ramaswamy and I talked about some of that, and I want to save the details for another post), Neeva’s goal is to be that mechanic every time I search.

So did it succeed this time? Well…it certainly didn’t fail. Neeva had the very same video as Google lower in its results, but the top result was its new AI feature, which offered me a step by step set of instructions. It turns out, those instructions are entirely accurate for the job I had to do. Had I gotten that answer from Google, I might have saved myself a minute or two of wasted time. Compound that minute saved across billions of searches, and maybe, just maybe, the world might shift on its axis.

NeevaAI also addresses some of ChatGPT’s biggest drawbacks – its answers are based on an old dataset (pre 2021 web), its answers aren’t vetted by any kind of citation or apparent ranking mechanism, and it often confidently spouts nonsense.

But for the Patience of Investors 

So NeevaAI offered me a pretty impressive test result, but will its unique features be enough for Neeva to break through? It’s hard to say, but one thing is certain: The more people use Neeva, the better it will get at finding the right answer for long tail queries like mine. The question for the company (and its star-studded board and investors) then becomes this: can the company outlast the shift it hopes to capitalize upon? Will AI-driven conversational search take off within the window of its current capitalization?

So far the company has raised $90 million – which seems like a lot, unless you’ve ever seen the compute bills for crawling and refreshing an index of 3-4 billion pages, then running AI across it (and that’s just a fraction of Google’s corpus). Not to mention the ongoing staff costs – Neeva has 55 employees, and top notch Silicon Valley engineers still command meteoric salaries. I’d estimate people costs at roughly $10-12 million a year, and infrastructure costs of at least that much. That means the company is likely burning north of $20mm a year – even with a growing subscriber base (Ramaswamy confirmed  that the company is closing in on 20,000 subs, for an annual run rate of roughly $1mm).

If my rough math is correct, the company will need to raise more capital this year. Unfortunately, the fundraising environment for tech startups is in the dumper. Then again, AI is the one bright spot – OpenAI just closed a financing valuing the company at $29 billion, and the world has fallen in love with apps like DALL-E, MidJourney, and ChatGPT. I certainly hope Neeva finds success with NeevaAI, and that the Valley, which has been hurtling through middle age with a recession of imagination, will find the courage to support Neeva’s ambitions. As far as I’m concerned, it’s not a question of if Google’s search model will be upended. It’s simply a question of when.

—

*There’s an entire post to be written about why Google and Microsoft can’t do what Neeva is trying to do – mainly due to their innovators’ dilemma. For another time.