You can only automate what you know
Many start-ups are building artificial intelligence products for businesses to use. There are hundreds of machine learning, computer vision or deep learning businesses propping every week.
Many of them want to enhance or change the way business is done. There is a common mistake that many AI companies make - they focus on the disruption and overshoot the market. Let me expand, you can’t sell a disruption to a market which doesn’t know what to do with it.
For example, if your AI can scan people’s face in a retail store and score their intent or desire to purchase. This is great, super valuable information, but if they have no known workflow or system to use these intelligence to drive business decisions, they are not going to be able to capture much value. Sure, you will get some enterprise customer whose CEO is desperate to appear technologically advanced and is willing to create new teams, new processes to leverage this data, but mass adoption by an entire industry is going to be a far cry.
From my experience in building products and following tech products for over a decade, B2B product categories follow the following arc
Phase 1: Data
People just want data about their systems - either these are existing deployed systems or newly deployed systems. The data has to be something you collected through manual laborious methods or a data you were always deeply interested in but had no way of collecting it. The indication of the later is to look at large corporations in your target industry and look for evidence of them having invested money in building expensive mechanisms or processes to gather that data. For example, when a company launches a website it was interested in knowing how many people came to the website. Today, its all too easy but in the early 2000s it wasn’t so, but big companies were building tools and methods to capture and use this data. Hence, analytics company had a good run from 2005. Similarly, if you are a manufacturing facility you might be interested in sensors which can get you data about your machine like uptime, power usage, performance etc.
This is all too obvious, but my point is you cannot build an automation layer if the underlying data problem has been solved, adopted by majority of the industry and standard workflows/processes for using that data have taken into effect.
Phase 2: Workflows
Once firms have the data, they need tools to be able to use it in their day to day processes and workflows. Going back to our examples, if websites have analytics data, they need charts, graphs and views that they can review weekly. If factories have sensor data, they need a daily health dashboard or alerts when a system is down or below its regular performance benchmarks. You want the team to be able to log in the system what caused the said performance drop etc.
The deeper you integrate the data and associated workflows into your end-users operations and key objectives, higher your ability to stick through.
Salesforce is a great example of building tools to embed itself into every sales teams workflows.
Phase 3: Automation/Insights
All of AI, automation or insights can only be adopted once the customers have built processes and workflows around the said data. You can’t expect the customers to jump to this step, if they haven’t gotten used to phase 1 and phase 2.
So if you wanted to move from analytics to helping teams automate A/B testing, you need majority of businesses to be consuming their web traffic data, using it in their weekly meetings to make decisions on what features, messaging or designs to push forward.
If you are serving the factory with sensors,, you cannot say we will build a predictive insights on what machines are about to underperform or go down without the organization first having built comfort and trust around data and using the said data to drive decisions or fix faults. Once that is done, then predictive insights or any other AI magic will actually help the customers move its core metrics forward.
Too many AI start-ups operate in Phase 3 in domains and industries where the market hasn’t matured to Phase 1 or Phase 2. Lots of these companies will fail or aqua-hired by one of their few big enterprise clients.
Phase 4: Developer Ecosystem, Platform Play, Integrations
This is when you become HUGE, you become the gold standard. You get to finally claim, you are in the promise land of network effects.
If your system is embedded in workflows and insights from your system are driving critical business decisions, soon the businesses will want you to connect to their other critical systems.
Salesforce, once again, is an excellent example. Once sales CRM works, you want it to connect to your mail automation software, your payment software, your product delivery software etc.
Similarly, if you using a web analytics tool to drive insights you want it to connect to to your customer support tool, you want it to connect to your attribution tool, your marketing CRM, your data visualization layer etc.
The more integrations you have, the deeper you embed into the systems of the company and more valuable you are. More valuable you are, you attract more integrations and developers to build around your platform.
Platforms are usually built there only after building a successful application. You can’t build platforms on day 1 (unless you are entering a mature market, but it comes at a massive cost) and you can’t build platforms in industries and categories where the customer hasn’t gone through the first 3 phases.
The really successful products therefore are those that can start when the customer is in Phase 1 and/or Phase 2, and solve those problems really well and then evolve the product alongside the learning curve of the customers.
The only time you skip to phase 3 or phase 4 at the beginning of a product cycle is if you are entering a mature customer segment or industry. For example, if you are building a sales CRM for tech companies, its a mature market for this product category, and hence you will have to build a product which does the essential aspects of all four phases. And because you have to build such a rich product from get-go you are looking at long product development cycle before product launch and millions of dollars and engineering hours spent before you can push your version 1 of product into market.
But if you want to ship fast, run on a small lean team, and get customers fast - you want to identify an industry or a segment which is still in Phase 1 or phase 2 of customer evolution. Find industries or verticals where key business data is unaccessible or workflows around this data are terrible and you would have found yourself a great entry point. Hopefully, its a big enough vertical or industry for you to build a large business out of.
Thus, before you automate some part of any business, ask if the customers are ready for adopting and value capturing from this automation yet?
Contributions and Thank you:
These ideas have evolved from deep product and go-to-market discussions for B2B SaaS companies with my colleague at Samsara, Hadi Hajimari and my co-founder at my new company, Tanuj Thapliyal.
