What are Actions on Google?

The terminology of Assistants is a bit muddy right now, particularly as it’s in its early stages, but Actions on Google are the equivalent of a mobile app, a website or an Alexa Skill. They are the assistant or chatbot a company can create specifically for them to allow users to interact with them over Google.

Below is an example of the Uber Action being demonstrated by Google’s Scott Huffman:

Basically, the user journey goes:

  • Invocation: User invokes the Action by saying a key phrase associated with your bot.
  • Interaction: The Action you have created responds, asking what the user wants. The user then interacts with the bot.
  • End: Once the user terminates the communication, they are no longer in contact with your Action. The next time they try to interact again, it will be with Google Assistant, and they will have to invoke your Action again if they want to interact with you.

This pattern is pretty much identical to Alexa Skills, with the key difference that an Alexa user must download the Skill in advance to use them. In this way, an Alexa Skill is more like an app, which the user can typically only make use of when installed, whilst a Google Action is more like a website, which users can connect to at any time as long as they know where/how to access it.

It’s also worth noting that these Actions can be used over text – using Google Allo – or over voice – using Google Assistant on Pixel and similar mobile devices, or from the Google Home. The work only needs to be done once for an Action, and then it will work over all mediums.

What can I use Actions on Google for?

Some of the functionality is still locked down, as only companies with early access can publish them to the wider market.

Fundamentally, however, you can use it to interact with a database or API to do pretty much anything you want.

A user could ask a question about product details or availability, order a product, enquire about the business and its contact details, or ask for jokes. It could even function as a Customer Services gateway by feeding in your frequently asked questions and troubleshooting database.

Over time, specialised ones could potentially develop to assist with internal communications. Imagine being able to ask if any of your meeting rooms are free!

Unlike older voice interfaces, such as ones reliant on you pressing 1 on your number pad, interactions with an Action allow for much more of a conversation; you can skip between questions or stages rather than having to follow a prescribed flow chart.

User: What’s my closest office?
Assistant: Your closest office is the Leeds office, at Beechwood Estate, Elmete Lane, Leeds, LS8 2LQ.
User: Are they open now?
Assistant: They are open today until 5.30 pm.
User: How old is the business?
Assistant: Home is 15 years old.
User: And what’s the phone number?
Assistant: The Leeds office’s phone number is 0844 576 8976.

With voice search already making up 10% of Android mobile Google searches in the US, and more than 50 devices being announced at last year’s CES with Alexa integration, it’s already an exciting market.

Adoption amongst users is also remarkably simple compared to previous technological leaps, with a simple interface when done right: just talk to it.

How secure are Actions for Google?

Right now, there’s not a lot of security built in, and it suffers from similar problems which plague the Amazon Alexa (see this piece on how the Alexa accidentally ordered dollhouses). For this reason, we’d recommend that any transactions should prompt a confirmation email, or similar safeguard, to make sure the user actually wants to go through with it.

How do I build an Action on Google?

In short:

  • Decide what content you want to make accessible.
  • Develop a database/API hook-ups to make sure you can access that content, if needed.
  • Develop your list of question/answer responses which could lead to the content.
  • Build it – using API.AI/Google Developers – and publish.The basics of building an Action on Google are not hard, particularly for a developer, but the nuances can be.

For instance, the API.AI interface makes it relatively straightforward to build up a basic conversation and specify responses, even with only minimal-to-no experience. The moment you want to start interacting with a database to ask a very specific question or make a transaction, it gets a bit trickier and you’ll likely need help from a more experienced developer.

More significantly, as with a website, you need to anticipate all the user needs, queries and then even the language users could use. Just as building a website typically requires understanding of development, user interface design and the content which has to go on it, an Action or Skill requires more than just development experience to get right.

For instance, if we want to prompt the user to say yes or no, we need think of all the ways they could use to say yes:

  • Yes
  • Yep
  • Yup
  • Yeah
  • Ok
  • Okay
  • k
  • Great
  • Fine
  • Sure
  • Why not
  • Righto
  • Okey-dokey
  • Aye
  • Affirmative
  • Very well

There are more, but you get the picture. Google will understand some of these confirmation responses from the start, but others you will need to setup. This goes for almost any interaction, as there is considerable variety between different peoples’ modes of speech.

Building an Action with only the basic responses in mind will lead to poor user experience, as they will have to ask the same question/give the same answer several times before the bot knows what they mean. Plus, you may miss out key bits of content the user will want to know – and this could be more than just the stuff they’d normally look for on a website.

After you have deployed an Action, you can look through how users have interacted with it to find out what things the Action didn’t understand or couldn’t help with – e.g. ways of forming an answer you didn’t anticipate, or questions/topics which you didn’t create a response for.

This can then be updated, so you can “teach” your Action how certain things should have been handled, and with machine learning, Google will learn over time, so less maintenance may be necessary in the long run. You can also identify content gaps, and work in additional conversational topics to your Action.

At Home, we can set up all of this, from development and audience research, to the maintenance of your assistant, whether it’s an Action for Google or an Alexa Skill. If you want to chat to us, tweet us @homeagencyuk or call Paul Lenihan on 0113 237 5371.