The business opportunity of Artificial Intelligence in all kinds of businesses is well documented. However, what kinds of businesses can actually make money from AI is still a topic of discussion. While large technology companies that have led the AI development such as Google, Facebook, Amazon, Microsoft have been profitably embedding AI in their products for quite a while now. This post focuses on businesses selling AI based products/ services to derive value in non-technology companies. 

Traditionally, the thought process has been that AI is a business similar to software companies and enterprise AI products focused on industry domains will make the most money. Let us take a step back and understand the economics of an AI business along with the entire value chain. SaaS businesses typically earn very high gross margins (in the range of 70-80%) because one-time product development costs are spread across multiple new customers. This allows such companies to invest more in product development, sales and marketing to expand at a rapid scale. Salesforce and Workday have been the most famous success stories of such a business model.

"Someone who sells shovels during a gold rush will make far more money than someone who goes out and mines for gold."

Now, many young companies are trying to build AI based software to solve industry specific problems. The results have not been too great because of the following reasons:

1. AI software is open source: Most AI algorithms and libraries, such as Tensor flow, PyTorch etc. are open source. There is hardly any proprietary software that can make money for AI businesses. Value lies entirely in the data and insights that can be drawn out of specific datasets.

2. AI models are less repeatable: With the current state of Artificial Intelligence technology, predictive models are often data specific and mostly not repeatable for different customers. Each customer deployment requires a big variable cost of setting up data pipelines, preprocessing data from different legacy sources and additional computational costs of retraining the models.

3. Data keeps changing requiring constant updating of AI models: While a stable software can be used for many years, AI models are as good as the input data used to train them. Obviously, data keeps on changing that requires constant adjustments in AI models adds to the variable costs. Imagine an AI software for retailers that was developed using pre-Covid purchasing behavior patterns. Its predictions would be worthless unless retrained with post-Covid world data.

In summary, there are no economies of scale in an AI business that can allow heavy upfront investment in developing software to be paid later on by high margins. In fact, some argue that there diseconomies of scale in AI. As your data size increases, costs to manage and train models increase exponentially. To achieve marginal accuracy improvements, marginal costs can be prohibitive (for an accuracy improvement from 98% to 99% might require 5X data). Hence, a more linear business model is better suited for an AI business. 

When we think about the above points, AI business looks very different from a SaaS business but very similar to a project type business where each customer deployment requires resources to customize AI models, set up data pipelines, clean up data and continuously maintain those models with new data. This is the domain of IT services and outsourcing firms where they can take up separate projects to deploy AI for various clients and provide resources needed to manage those models. 

Artificial Intelligence has shown remarkable progress in the last decade but for all its advances, the current AI technology is heavily dependent on finding pattern in old data to predict the future. That is why AI has been very successful in self-driving cars that require identification of images, which do not change with time. But, poor in predicting a football World Cup winner. This fundamental limitation inhibits AI businesses to develop a 'one size fits all' model that resembles a scalable software business. The real value in AI does not lie in software; it lies either in owning the data or in processing the data. 

Reference: https://a16z.com/2020/02/16/the-new-business-of-ai-and-how-its-different-from-traditional-software/

Disclaimer - Views expressed here are those of the author and should not be considered as views of the employer.