At Emerj, for over three years, we’ve been tracking the development of artificial intelligence startups and solution providers across industries, speaking to founders and team members and communicating with the enterprise leaders and buyers who depend on this ecosystem of AI solutions.
In the last month, we’ve put a strong focus on determining the impact of the coronavirus on the artificial intelligence ecosystem and getting a sense from vendors and large companies as to which of them are likely to survive the economic downturn. The results don’t bode well for many artificial intelligence companies.
This article will highlight two of the biggest challenges that AI startups face and how they might address them: That B2B AI startups are already struggling with cash, and that AI deployment is an uphill battle to begin with.
B2B AI Startups Are Already Struggling With Cash
Very few artificial intelligence startups are generating revenue, let alone turning a profit. Most of them are still living off of venture capital money and struggle to scale profitably. In a given industry (say, insurance) there might be 80 reputable artificial intelligence vendor companies, and only five or 10% of them may be able to survive the pandemic. For the most part, AI startups are still piloting their technologies, figuring out what works and what doesn’t, and determining a potential fit for their product in the market.
In this experimental phase, AI startups have to explore:
- If their value propositions can hold true. Many of their value propositions will need to be adjusted to appeal to the new, pressing priorities of the industries they serve (efficiencies and cost-cutting will be common new priorities, but each sector’s new priorities will be different).
- If they can access the data they would need to deliver results for their potential clients. Many startups will need to find ways to use less data sources to deliver on their AI value propositions. Long setup times and integration requirements will be less acceptable in a time of greater economic pressure.
- How different the technology ecosystems of their potential clients actually are. Finding the common data sources and business processes of clients will allow startups to find ways to shortcut their path to deployment. Completely bespoke integrations and build-outs for each client is often part of what it takes to find product-market fit with an AI solution, but the sooner companies and find what their clients have in common and make setup fast – the better. The pressure is on.
As such, most AI startups don’t have a source of stable, recurring revenue. AI integration costs have also proven harder than startups have anticipated across almost every industry. It can take months for a vendor to close a deal with a large enterprise and just as long to receive approval to begin work. Then, vendors need to work with data scientists, IT staff, and subject-matter experts at the enterprise to begin integrating their product. In addition to the iteration required when building and testing AI products, this can also take months.
Even after all this time, the vendor’s product may still yield negligible results for the client, and the vendor may not win any more business from the enterprise. Due to the length of the sales cycle and bespoke nature of the projects, AI vendors are usually working off of venture capital exclusively. If they don’t have a lot of it and can’t win deals for the next few months, they’re unlikely to survive.
Uphill Battle for Deployment
Deploying AI in the enterprise is challenging for reasons that go beyond cost and technical expertise. Many enterprises don’t understand this, and this makes it challenging for vendors to win deals. During a time when enterprises are going to be looking for quick ways to stop bleeding money, which we discuss below, AI startups are going to struggle to find a fit in the market. At Emerj, we break down the factors necessary for Ai adoption into three major categories we call the Critical Capabilities for AI deployment. These broadly include but are not limited to aspects of data infrastructure needs, cross-functional AI teams, and executive understanding of AI concepts.
Data Infrastructure Needs
Enterprises that want to leverage AI need to know what data they have access to, can get access to, and can’t get access to. They can figure this out by performing a data audit, during which data scientists, IT personnel, and subject-matter experts collaborate to determine which of the enterprise’s data could be used to build or integrate an AI product. In many cases, a data audit will reveal that data is accessible, but it needs to be cleaned and harmonized, which adds additional time to integration.
The fact that most enterprise data infrastructures are unprepared for artificial intelligence projects is of concern for vendors that want to sell AI for much of the reasons already described. Overhauling data infrastructure in significant ways is unlikely to be on an enterprise’s priority while they’re dealing with the near-term fallout from the coronavirus pandemic.
Cross-Functional Teams
Enterprises need to make sure the vendors they work with speak to multiple stakeholders in the company. They need in-house data scientists and IT specialists who have an understanding of their data ecosystems. They need business leadership that can be on the same page about the goals of a project, and they need subject-matter experts, all of whom need to collaborate if an AI project is going to have any chance of success.
For example, if a company wants to deploy a fraud application, it would need subject-matter experts who understand fraud detection processes and workflows to help set realistic goals for the project and iterate it towards success as data scientists begin to train an AI model.
This is harder than ever in an environment that is now almost entirely remote. Companies are already scrambling to figure out digital workflows for core processes. They are unlikely to prioritize collaboration for AI’s sake unless an AI project is going to solve immediate problems, discussed later in the article. Remote work poses a serious threat to new deployments actually gaining any traction whatsoever.
Executive Understanding of AI
One of the biggest barriers to entry for AI startups looking to win deals during and immediately following the pandemic is that enterprise leaders still don’t understand how AI works at a conceptual level.
They don’t need to understand code, but they need to know about the importance of data in training AI algorithms; they need to know how long AI can take to adopt at large companies; and they need to know that some of their functional leaders will need to spend their time collaborating with data scientists instead of doing the job they were originally hired for and have been doing for years. Most enterprise leaders think AI is like regular IT, but this couldn’t be further from the truth.
What this means is that AI startups that get on the phone with leaders at large companies need to find a way to convince them that their projects will be relatively simple and cost-effective. If they can’t do this and follow through with it, they’re unlikely to close deals during this time when business leaders are looking to cut costs significantly to stay afloat.
What Surviving AI Startups Will Have in Common
There are two factors on which the potential success of an AI startup through the coronavirus pandemic weighs most heavily.
Stopping the Bleeding
The first is whether or not the startup can stop the bleeding. There are some AI use-cases, such as payment and insurance fraud detection, that are not only relatively easy to deploy but prevent further loss. Companies are really clamping down on the defensive.
They want to handle immediate problems and bring themselves back to a relatively safe place. As such, they’re looking for ways to reduce risk. AI has a proven track record of doing so in the fraud detection and an anti-money laundering domains.
Payment and insurance fraud skyrocket in hard economic times. One response to the crisis may be to deploy AI applications that detect these types of fraud.
Cybersecurity applications are also likely to become more relevant as new threats emerge and are harder to detect by remote cybersecurity teams that are used to working together in person.
For example, a cybersecurity team member might mistakenly give someone a password who calls in claiming to be someone at the company, perhaps because it is harder to verify with that person when they aren’t in the same building.
Remote work is very difficult to implement when there has been one way of doing things for years and much of it has relied on in-person communication. As a result, remote work may open up more opportunities for fraudsters and hackers to infiltrate enterprise systems.
Enterprises may look to AI as a way of solving these risk-related problems, and vendors that provide them could win lucrative deals despite the economic downturn.
Ease of Deployment and Evidence of ROI
It will be harder than ever for AI vendors that don’t already list named clients on their websites and digital assets to win business with large companies. This is somewhat of a vicious cycle because these vendors will need case studies and other forms of evidence that past clients have benefited from their products if they are to win new clients during the pandemic.
Potential clients will need a measurable way to gauge the ROI of a vendor’s solution, especially because they can’t afford to take a chance on new startups when they’re losing millions by the day as productivity plummets. As a result, established companies with a history of client success are more likely to survive the pandemic than small and brand new startups.
In addition, enterprises are not going to want to work with an AI vendor if its product is not easy to deploy. They don’t have the time nor the staff to experiment and iterate with AI. Right now, they need essential personnel to be doing their regular jobs.
AI startups will need to start focusing on essential data sources and integration points that they need in order to deliver value to their clients. They will have to drop many other facets of their product’s usual integration and as many other data sources as they possibly can. If they can make their product’s integration as smooth as possible, they have a better chance of winning deals during and after the pandemic.
Addressing these two enterprise concerns should be the number one priority among the C-suite at AI startups. They will determine the winners and losers in the months ahead, however long this may last.
