Three success criteria for ‘Assistant-as-app’ companies

Nir Eyal recently wrote a great post speculating that ‘Assistant-as-app’ companies might be the next big tech trend. I think he’s right – it’s a trend that has legs, but it’s also a trend that has been going for a couple of years already but hasn’t yet been given a good label. Magic and Operator are the companies in this space that have made the biggest splash recently but companies like and Big Health from our portfolio and like Native and Vida Health that Nir mentions have been pursuing variations on this theme for a while.

In another post Nir proposes the following definition:

I’ve proposed “assistant-as-app” to mean: an interface designed to enable users to accomplish complex tasks through a natural dialogue with an assistant.

He emphasises ‘natural dialogue’ because the first success criteria is that users don’t have to learn a complicated interface. Few people can be bothered to do that for anything, let alone when there’s a simple option available which will probably get you to a result faster on the first couple of times through. Complicated interfaces are in effect asking people to make an investment of learning time against a highly uncertain outcome – not an attractive proposition.

However, the key point is that there is no learning curve, so rather than ‘natural dialogue’ I would use the more inclusive term ‘easy to use interface’.

The second success criteria is that the service delivers something more than the fully human equivalent. That doesn’t mean it’s better on all dimensions, but that it is demonstrably better on at least one important dimension. For example Big Health is a therapy service for insomniacs that offers 24-7 access to an AI therapist called The Prof. Human therapists typically see their clients for an hour per week, whist The Prof checks in multiple times per day and if you wake up in the middle of the night he’s there (in his dressing gown) to help you get back to sleep.

It seems to me that the main ways that ‘Assistant-as-app’ services can be better than humans are:

  • 24/7 presence and immediate response
  • Price – high levels of automation bring some services to a price point that works for consumers (although note the third success criteria below)
  • Access more information about the user and use better analytics to deliver the service – e.g. data from wearables, activity diaries, transaction history
  • Access more components to build a customised solution – human services are limited to what the human operator knows, Assistant-as-apps can access all the inventory on the web

If you can think of more ways ‘Assistant-as-app’ services can be better I’d love to hear them.

The third success criteria for ‘Assistant-as-apps’ is that the economics work. To get to a price point that’s attractive to consumers typically requires some heavy duty automation on the back end, often leveraging AI. Most companies in this space start out delivering the service manually and initially lose money on every transaction. Predicting the extent to which those manual activities can be automated and can be difficult at the outset, but it’s critical to the business model and should be addressed early on. It’s easy to build an amazing service that will never make money and I suspect we will see some high profile companies make this mistake.