The companies that win in the intelligence era will be the companies that start the intelligence era. The winners of the web era were companies that started on the web; only 2 of the top 50 websites are owned by companies that started before the web. The winners of the e-commerce era are Amazon and startups like Dollar Shave Club; not Macy’s and Nordstrom. The winners of the AI era will be AI-first companies. Very few companies from the SaaS era can adapt to a world in which proprietary data is required to develop a competitive advantage. This is because AI-first companies focus on getting data rights, generating data assets and building measurably efficacious predictive models. This has implications for how companies think about negotiating customer contracts, the importance of product management with respect to data management, metrics and how venture capital investors think about their job.
The first reason why SaaS companies are unable to adapt to the intelligence era is that they can’t use their customers’ data to feed machine learning models.
At the start of the SaaS era, vendors were asking customers to shift databases from their own sites to the vendors’ sites — to the cloud. Customers were hesitant to do that because they were scared about what would happen to the data once it left their site. Would it be stolen? Would it be used for something without their knowledge? Customers were convinced to move their data to the cloud despite these fears because the total cost of running software on the cloud compared to on their own site was much lower. However, they only did so with explicit agreements from the SaaS vendors that they would never use the data for their own purposes.
The problem for SaaS companies is that machine learning models need that data to learn. To move into the intelligence era, these SaaS startups must re-negotiate data rights with their existing customer base, or go on a data acquisition spree.
Startups in the intelligence era approach customers who are comfortable sharing their data. This is important because intelligence-era applications require lots of data manipulation — capturing, cleaning, labeling, querying and analyzing the data — in order to provide actionable predictions and automation to customers. Many enterprise customers are wary of sharing deeper access to their data with large SaaS vendors, who may become potential competitors. Startups present less of a competitive threat and are better positioned to successfully negotiate the rights to use this data .
The second reason why SaaS companies don’t adapt well to the intelligence era is that they are more focused on product strategy than data strategy. They are focused on adding features rather than adding learning loops.
Again, machine learning models need data from which to initially learn. The data required to enable continuous learning changes over time, going beyond the initial data, depending on the current state of the model. That means that companies need to constantly acquire new data that’s potentially predictive of the next decision that their customers want to make.
Startups in the intelligence era have people in their organization that are solely responsible for data acquisition operations. Here are just a few examples of operations within Zetta’s portfolio.
- Team of 40 human labelers at a Series A company — some low skill and some with domain expertise — that label images for supervised learning models.
- Data acquisition engineers responsible for acquiring data to train translation models. Hardware team that innovates on core camera and mobile chip operating systems to collect the best possible images to train a product recognition model.
- Infrastructure team that runs bots testing the top 1,000 apps in the app store, to improve their testing product for paying customers.
- Integration engineering team that constantly builds integrations into legacy systems to collect data on building operations.
- Conversion team that continuously ingests, cleans and converts government and utilities data to feed energy use prediction models.
- All of the companies for which we work have a dedicated data acquisition and management function that gets as many resources as the product management function.
The third reason why SaaS companies don’t adapt well to the intelligence era is that they are measuring the wrong things.
Magic numbers, SaaS quick ratios, cohort retention and such are very good ways to measure sales. However, just selling more does not necessarily generate a sustainable competitive advantage. Absolute focus on such SaaS metrics will lead to a lack of focus on generating a sustainable competitive advantage through data and compounding that advantage with a self-learning system.
Startups in the intelligence era measure how their models help their customers make decisions. They measure predictive accuracy, precision/recall and other functions of decision efficacy. They design workflows around predictions and measure how the execution of those workflow gets more efficient as the predictions improve. In some cases, these startups may measure functions related to labor augmentation or labor replacement. Here are some examples of what startups in Zetta’s portfolio measure.
- Mean Absolute Error and precision/recall of classifiers
- Productivity increases that result from accepting suggestions in a workflow interface
- Data coverage, i.e. how many customer data requests through the Clearbit API can we fill? 
Side note: what this means for venture capital
A side note for venture capital investors: SaaS doesn’t seem to be suitable focus. In a sense, this is an observation rather than a prediction because SaaS company formation has fallen by 44% in the last 3 years . But why? What does such a seemingly provocative statement mean?
Our job as venture capital investors is to find and support companies that are creating new technologies — new levers for humans— and build a competitive advantage along the way. Usually, this means finding and supporting companies creating new markets. Finding companies in such a realm principally requires developing models for understanding what is a sufficiently new technology, how that technology generates a competitive advantage and how new markets will develop to be large enough for the leading companies in those markets to be companies of consequence. These models are our intellectual property and, indeed, our competitive advantage in the investment industry.
Investing in SaaS doesn’t seem to be a suitable focus because cloud technology is not new, companies cannot build a competitive advantage merely building workflow software and markets are known to the point of being heavily metricized. SaaS investing is somewhat of a science at this point. Market-based evidence of this: there are a number of debt providers that issue MRR-based loans to SaaS companies. Once something can be priced as debt, it’s probably not going to yield enough for equity investors.
- The AI-first startup playbook
- AI adoption is limited by incurred risk, not potential benefit
- AI-First Companies
- Data rights are the new IP rights
- The Intelligent Enterprise Stack
- Beating Behemoths
- Don't sell your data
- Framework to grasp industrial analytics opportunities
- Beyond systems of record
- Positioning a machine learning company
- The intelligence era and the virtuous loop
- Vertical beats horizontal in machine learning
- Zetta Bytes AMA: Questions to ask about pricing
- Zetta Bytes AMA: Hiring a CTO
- AI Entrepreneurs: 3 things to check before you pursue a customer
- There are more data scientists than you think
- Stages of funding in the intelligence era
- Could data costs kill your AI startup?
- Measuring AI startups by the right yardstick
- Finding the Goldilocks zone for applied AI
- Data is not the new oil
- Zetta Bytes: Privacy Preserving Machine Learning
- 10 innovations of the next decade
- GDPR panic may spur data and AI innovation
- Computing like a human
- New opportunities for hardware acceleration in data analytics
- Hardware acceleration in data analytics