Voice is going to change the internet

by | Dec 9, 2019 | Experiments, News & Blogs, Emerging technologies

Reading time: 9 minutes

One of the topics where prominent attention is given at Novum is the Voice technology of (among others) the Google Assistant. We asked our colleague Jeroen Vonk to:

How Project Silver makes it clear that online services must be completely different

Just talk to the computer. It has recently become possible with, for example, the Dutch version of Google Assistant. It was presented at the end of 2018. “For me that was the time to explore the speech possibilities for the Social Insurance Bank. Wouldn't it be great if you could just ask the computer when the child benefit will be paid, or how you can apply for AOW? "

Now, a year after we started the first experiments, we know more: to make a success of speech technology, the web must be set up completely differently. It only works if you can give clear, human answers to questions. The coming years will therefore be dominated by 'Voice first', or rather: 'human first'.

How did we come to this outcome? In retrospect it seems simple. We started thinking from the older target group of the Social Insurance Bank: recipients of AOW. And how we can be of service through a voice assistant.

The elderly do not lead

Younger people, so people under 65, can already keep up with the most new technology. They are more comfortable with computers and government services. They also understand better how 'voice' works. So if you are going to do something for people who receive child benefits, that is not a real challenge for anyone. Doing something for the elderly is more challenging, especially because that target group is very often skipped.

'Voice' can be a godsend for elderly people who are put off by keyboards or icons on touch screens. Just ask, without entering complicated codes or key combinations, or having to click on submenus and tabs. Everyone wants that, but the elderly certainly do.

In the preparation of our experiment I discovered that we hardly have any scientific knowledge about this group. Studies in America that were done with the elderly turned out to be about people between 50 and 55 years old. They are not older people. Older people are the people who are retired and who no longer participate fully in the labor process. People who miss a connection to the new world.

Everyone has such a neighbor

When I told other companies about this study, they said, 'People over 65? But that is very difficult, isn't it? They don't want that at all, do they? ' If I then asked further, it turned out that they also thought that the elderly could no longer do that at all. But people of 65 have probably spent half their lives in an increasingly modern office. They have learned to use computers, experienced the rise of the smartphone. So: how do you get it that it doesn't work?

Then it turns out that Everyone always has a neighbor, a character who meets their image to not have to do it. Of course I also know people who do not want anything to do with digitization. For example, they have been in the store, the cash register is being viewed digitally and they no longer need that, they just want to read a newspaper. Fine.

But the world doesn't look like this anymore. In the meantime, partly thanks to our findings, Google has also chosen to seriously include people over 55 in investigations.

No practical examples

The technology around speech applications, in short: voice, is still so fresh that there are hardly any practical examples to be found. That is why we first started investigating usable tools ourselves. In the end we have a very simple one Google action, say a Voice app made with Dialogflow, a free tool from Google. With this Voice app you could ask what a certain scheme of the SVB was, how you qualify for it and most importantly: on what date the money is deposited. With this demonstration we were able to inspire the internal stakeholders at the SVB to further our research.

The app soon turned out not to be suitable for continuing. He worked with a fairly simple tree structure: a question leads to a limited number of possible answers, which in turn lead to other questions and answers. but that is not enough for the problems we wanted to solve.

During our search for practical examples I came across the name Maarten Lens-Fitzgerald more often. He is a voice specialist and always busy with new media. He already had intensive contacts with Google and during our conversations Anbo and Achmea also joined.

We decided to collaborate and formulated a specific research question. The emphasis had to be less on help: People had to start using it daily, for their convenience, so also simply ask questions about the weather or the train connection to Amersfoort.

Our dream was clear: how great would it be if organizations offered services that meet the needs of the elderly. Then the more care-oriented applications would automatically land. Project Silver was born with this.

Project Silver

After an initial scoping session, in which all parties involved were introduced and their wishes formulated statements and goals, we got to work. We used all modern techniques and work forms for this, such as Design thinking, Google Design Sprint and Lean Startup. Main feature: from idea to practice as quickly as possible.

First we had to solve the question of the target group: how well did we know it? As I indicated before, there has been little research into people over 65. So we had nothing but assumptions.

Under the leadership of the ANBO, we have therefore investigated what technology older people are already using, whether they are already using their voices and whether they wanted to participate in an experiment with a Google Home. In the end we received 3450 surveys. The results were surprising: three-quarters (74%) of the people over 60 used a smartphone. A little less, more than two-thirds, also uses smart technology in the home, such as a smart meter or a smart TV.

Intelligent thermostats and audio systems such as Chromecast and SONOS were less known. One fifth of the respondents already regularly used voice control, for example in the car.

In the survey we also asked what people had for speech technology. This resulted in a wide range of responses, such as use for route planning, looking up information, alarm clocks and reminders, operating doors or opening and answering emails. So very practical matters.

Unpack: not all at once

One of the questions in the survey was whether they wanted to participate in a practical test with a Google Home. 250 people applied for this, of which 14 were chosen. During a meeting with these 14 elderly people, we explained what a Google Home is, how it can be used and how it can be installed together.

That produced interesting things. We discovered that one Google Home installation works best, but that it becomes problematic if you want to do more in the same space. Now that will not happen often in practice, but this test did. Google hadn't realized that. There is a number on such a speaker, and you can't read that number without a magnifying glass. It is the number that makes the Home unique. The moment you install it with the Google Home App, your Google Home will appear. With one that is easy, but if you have more than one, you don't know what it is. Is it 36669 or 34789? We had six in the air. Then you can press a button and then sound will be heard, but which one is now playing?

The trick is to install one, unplug it, close the app, and then the next. Something like that must be included in the manual, as is already the case with Ikea smart lamps. These are insights that you can only gain in practice. But then again: there was also a lady who tore open the packaging without further ado and did not look at the instructions for use: 'I always do that', she said. So you have to take that into account.

Everything is technically possible, but not yet in terms of content

What the test showed is that these people can handle new technology well, but that they miss a lot of possibilities when they use speech. The technology did not answer all questions. Many information sources are not yet accessible via voice. If you now ask a question about a municipality or about the SVB, you will not receive the desired answer from the SVB. You get an answer at best that you know for sure it can't be right.

Technically everything is possible. We can create a voice application and use the SVB website as a source for the answers. The only problem is that the average website is not written to be read. You can have it read aloud, for example, that is already built in with us, but during such a test you find out that the texts are unsuitable for such a practical application.

We then formulated answers ourselves, but people found those answers, which were already very short, far too long. So we had to make it much shorter. The rule of Amazon and Google is that it must be possible to say an answer in one breath. But a breath of who? A breath of someone 65 is shorter than my breath. So you should also keep in mind that you are out of breath when you start practicing and then record your answer.

Who else can you trust?

We have also extensively tested our Voice app with AOW questions. There we got the response: "The answer will be correct, because I trust you." When we asked why they trusted us, we got back, 'You are from the government, so I have to trust you. If I can't trust you, who else? ' There were also people who said, "I have to grab a notepad now to take notes, because it's too much information." And then we have given simple answers in our eyes. Most people also did not understand that you had to call the Voice app with “Talk to SVB”, how is someone supposed to know that you have to be at the SVB?

We then looked at whether we could optimize it via voice search, so that you can get search results via voice. Then it turned out that the answer you wanted to give had to be written out completely in advance, taking into account that someone will only hear it. So, if someone asks the question, "When will the child benefit be deposited?" then they just want to hear: "Child benefit will come on April 23rd." and not: 'The Sociale Verzekeringsbank is responsible for paying the child benefit. They will do this on April 23, July 22 and October 24. When it is in your bank depends on your bank. If you do not receive it on time, please contact us. Look at our contact page… 'and so on. Then the questioner has completely lost what has been told and does not know that it will come on April 23. The problem is that all our websites are full of answers like that.

Voice: everything has to be different

The arrival of Voice technology makes a new way of thinking about websites necessary: Voice First: first from speech. If you extend it even further, it is actually 'Humans First'.

This is due to the way Google searches for answers. Google has a very large database: the Knowledge Graph. It contains all the information that Google retrieves from existing sources. They choose that based on what is important to users, and that is determined by how users click on it, what they read, and what they continue to do. If the SVB also wants to enter that source system, Google first looks at the number of visitors. With 10,000 visitors a day, it is not interesting for the search engine. Wikipedia, for example, has 10,000 per minute. That is why Google does include information from Wikipedia, but not from the SVB.

The SVB must therefore come up with a different solution. That lies in the restructuring of the website. For example, if you ask Google how far it is from the Earth to the Moon, then Google first looks to see if they have the answer in the Knowledge Graph. If that is not the case, then they search the normal search results for the most likely answer. Google Home will then read that. To ensure that the SVB answers a question about the services of the SVB via Google Home, the SVB must restructure its own website. This must be such that the most relevant answer to a user's question, for example, when the Child benefit is transferred, is directly found by Google.

No more sending, but listening

I think that within two years the website will be optimized for speech applications, so that to the question: 'How can I apply for aow?', The answer is: 'The SVB website states: you can apply for aow in these steps. Do you want to do that now? '. If you then say yes, you will be included in the SVB application.

In fact, you no longer have to go to a website for this type of information: you go where you want to be via Voice. We are going there. What will you do with a website? We now focus too much on sending the information. First go and ask people what information they actually want.

This is a task, also for the SVB and other parts of the government.

Would you like to know more about the Silver collaboration project in which Novum is involved? Take a look at the WEBSITE of this project

Work with us on this challenge

4 + 7 =

Share this article