Choose Your Own Adventure: Which AI Business Model Is Right For You?

This post was originally published in Forbes.

Kespry, which uses aerial drones to assess storm damage on houses for insurance, is an example of one successful AI business model

Artificial intelligence is revolutionizing every aspect of work and life, as nonstop news headlines make clear. Yet one aspect of the AI disruption remains relatively unexamined: the right business model for AI companies.

It’s a commonly held belief that successful AI startups, like cloud / SaaS companies before them, share a common business model. Like AI, cloud software’s business model initially puzzled customers and investors. Cloud pioneer Salesforce, for instance, eschewed the big, upfront license payments common with traditional software vendors, and also shunned recurring maintenance fees and costly on-site installations. Now, the Salesforce model is the norm for cloud/SaaS companies—and investors can recite common SaaS company metrics like magic number, CAC and LTV in their sleep.

But it’s difficult to graft this cloud business model onto nascent AI startups. This is mainly because cloud and AI technologies differ in fundamental ways: AI is fueled by data, vast amounts of raw computing power and mind-bending algorithms. It’s inherently more complex for customers to digest than cloud startups, so the technology has to be sold in different ways, too.

Slower software deployment? More on-premise integration? Sometimes that’s what you get with AI. So how can that beat the status quo, and convince customers to buy new AI technology? Over the last two years, we’ve observed three emerging AI business models that are starting to work. Each one is different, and they each have different strengths that work better for some solutions than others.

AI Business Model #1: Bolt-on

The first type of AI solution is deployed much like a product from a SaaS company, and the business models are almost interchangeable. These AI solutions sit seamlessly on top of other systems of record, like a CRM (customer relationship management) product or an ERP (enterprise resource planning) system. AI accesses data flowing through these systems, fueling business improvements over time.

Many AI startups fit into this model: Chorus AI and Gong both tap into Salesforce, using AI to optimize a company’s sales practices. Customer support software Solvy sits on top of Zendesk or ServiceCloud and automates replies to support tickets. Sift Science uses machine learning to reduce customer fraud like payment abuse or fake content.

Because this business model resembles the SaaS model, it seems easy to evaluate. The strategy is to “wedge in”, starting as a value-added feature and maturing into a platform. These solutions are fast to deploy, like cloud software, so the sales cycle is quick, with an easily-defined ROI. By getting many customers quickly, the AI solution rapidly builds a data moat, so it gets smarter faster, too.

But this speed and ease might also bring disadvantages. Just like cloud software, it could be easy to rip and replace these AI solutions. And if an AI solution doesn’t progress beyond a nifty, nice-to-have feature, it’s even more vulnerable to budget cuts. Just because this business model is familiar doesn’t necessarily make it the optimal one.

AI Business Model #2: Enhanced process

In ts second AI business model, deploying a new AI product doesn’t change existing workflow at all; it just turbocharges the effectiveness of current workflows by integrating AI into them. These are deep-surgery integrations and require lots of implementation work, with much-improved processes as the payoff.

AI startups in this category include Ayasdi, IBM Watson, and H2O.AI. Playing across various industry verticals, these solutions help customers improve core business operations. Take IBM Watson, which first garnered international attention through its appearance on the popular TV show “Jeopardy” by answering questions posed in natural-language form. It analyzes big-data patterns in real-time, flagging insights that might be worth responding to. Watson is being used to help prepare tax returns, and even manage building elevators through complex sensors transmitting data back to computers.

This AI business model differs from the prevailing cloud one. Its disadvantages are obvious: With intensive deployments, the sales cycle is long. Low volume means each deal must be big to keep the startup selling the product afloat. But this model’s advantages are high-stakes, too. Once implemented, these solutions are incredibly sticky and lend themselves well to upselling. Just like AI’s potential, this model’s ROI may be unlimited.

AI Business Model #3: Letting the machine stand alone

In the third AI business model, the AI technology changes an entire workflow by introducing an AI-infused, better-way-to-complete-a-business-process. AI “owns” the experience end-to-end, with very little human-required assistance, giving algorithms the full control over the experience.

Example companies in this category include autonomous cars and drone companies like Kespry, whose aerial drones collect data for construction, mining or insurance purposes. After a storm, Kespry drones can assess roof damage, so there’s no need to send an insurance adjustor on top of your roof. Since the data is directly sent to the cloud and analyzed using AI-powered computer vision, the insurance company can estimate claims data almost immediately.

Because this model involves maintaining hardware, the advantages and disadvantages differ from pure cloud plays. In this model, hardware is a cost center and a commodity the startups must service and store. The AI software inside the drones (or vehicles) is the differentiating IP and the startup’s revenue generator. These AI startups sell software subscription packages to companies who rent the hardware; those subscription packages can be expanded over time to do even more.

In conclusion…

The future might bring even more viable AI business models. The AI era is upon us, and investors think in herd-mentality terms. It will take time before investors sort out the “right” formula for AI success.

What does this mean for AI startups? Founders: choose a business model that enables your business to grow effectively. Explain how your technology offers meaningful impact and delivers value to customers. Don’t be discouraged if your business doesn’t fit into a familiar, SaaS-style mold. That might mean defending a professional-services model to investors and customers who’ve previously shied away from that. Lean into it if you must. Trust your instincts and find the right business model for your company.