Digital Innovation Top Stories
5 mins read

“Make something people need:” How BBC Good Food is getting ahead in voice search

alexa with speech bubble and meal

BBC Good Food are one of the most prominent food media brands in the UK, with a market-leading magazine, live events and a book business, as well as the well-known food and drink website, which attracts in excess of 90 million page views a month.

The brand was acquired by Immediate Media in September last year, and has since had a fresh focus on how new technologies can enhance the brand, including the growth of voice.

Hannah Williams, the Head of Digital Content at BBC Good Food, explored their initial foray into the world of voice search at the PPA Festival on 9th May, from identifying the opportunity to formally launching the skill in August of last year, and subsequent learnings.

The starting point

BBC Good Food has the advantage of a strong starting position to experiment with voice. With 80% of their current audience coming from Google, and the site ranking above the fold for 96% of core recipe terms in the UK, it’s a brand that most of the British public will be familiar with.

Even if the expectation that 50% of content searches are expected to come via voice by 2020 is a little bit of a stretch, monthly installations of Alexa apps have rocketed in the past year, and it’s estimated that 2 in 5 adults performs a voice search at least once a day.

It’s an opportunity that the BBC Good Food team couldn’t afford to miss. “If you look at Amazon Alexa app downloads over the last few years, you can tell the direction of travel is only going one way,” Williams explained. “And if you’re so reliant on Google for access to your audience like we are, we couldn’t simply ignore it.”

Before jumping straight into creating a product, Williams and her team explored what people were saying around their contact channels, what they could gather from web analytics, and what their competitors were doing in that space.

“Like all good digital products, the first thing I did was ask the audience,” she said. “We also looked into where people were using these devices.” 41.4% of smart speaker owners keep their device in the kitchen, which is second only to the living room at 45.5%, and highlights the opportunity for voice in the kitchen.

The proposition

They worked with tech agency ‘Hi Mum! Said Dad’ to come up with a two-step user proposition: to allow their users to access BBC Good Food’s 12,000 recipe database via voice, including filtering by popular preferences like cuisine and cooking time, and then provide a step-by-step method completely hands-free.

It’s a complicated brief for developers to figure out all the different user journeys that they might take without visual prompts, and one that Williams and her team faced a steep learning curve with.

“We started with a massive piece of paper, and we tried to map out all of the possible entry points, decision trees, certain pain points that might come up. And we quickly learned that that was pretty impossible. Rationalizing that number of possibilities down is endless.

“So instead, we took what analytics we thought we could translate from web, along with our editorial and product hunches, and came up with a quick and dirty prototype: what better way to learn than to give it to people to play with!”

Williams put together a prototype, and gave it to their cooks in the kitchen to feed back on. From their experiences, there were three key point of learning that the BBC Good Food team took away.

Firstly, the questions had to work hard. With 12,000 recipes to navigate, the voice skill couldn’t be constantly asking the user a barrage of questions to get to the right recipe.

“The solution was to enable complex conversational commands with filtering criteria,” Williams explained. “So for example, ‘Alexa, ask BBC Good Food for a quick recipe for chicken and tomatoes’, is one sentence, but four distinct filters are programmed into Alexa.”

Secondly, the content had to be appropriately structured. “BBC Good Food has a 30 year heritage as a magazine brand. That means that some of the content up there wasn’t originally written with the website in mind, never mind voice,” Williams said, highlighting that some structuring of the legacy content was necessary.

“We had to shorten sentences, we had to eradicate abbreviations, we had to replicate ingredient quantities as well as other ingredients in the method. Now that’s an editorial overhead. But optimising for voice in this way also optimised for web search, which is time well spent.”

Finally, they had to look at limiting the opportunity for error when it comes to automated speech recognition (ASR); programming Alexa so she can best understand what the user is on about.

“When we experimented with the prototype, a common area for error was at recipe selection. We were giving users the choice of three recipes to choose from, and expecting them to be able to repeat the recipe title back to us verbatim. And when you’re dealing with recipes like chicken and chorizo jambalaya, there’s a lot of opportunity for error if you need that to be said verbatim back.”

The solution was to number the options as ‘1, 2 and 3’, so that the user has only to repeat the number back, like many options menus in traditional call centres.

The journey since launch

Since launching the skill on Alexa in August 2018, there have been subsequent iterations based on learnings. In February, BBC Good Food became a strategic partner for Amazon UK and Ireland, which means that they are the default for any recipe requests made on Alexa devices.

The skill also won a 2019 Webby for the Best User Experience in the Apps, Mobile and Voice category.  

Williams highlighted how the initiative has given BBC Good Food the opportunity to gather more data about engaged users. “If you’re saving your recipe or favoriting it in Alexa, in order to do that, you have to register with My Good Food, so driving our wider strategy and moving people away from anonymous into registered and engaged.”

“It’s also an additional personal touch point. So we sell brand partnerships across a website, magazine, app social. And this is an additional platform for our advertising team to sell.”

There are also a wide range of benefits from being early to a platform, despite the risks and lack of reliable monetisation at present.

“We’re learning where our users are. We’re in constant dialogue with a platform to prime ourselves for opportune study, and we have exclusive real estate to learn where our competitors don’t.”

Williams also shared some tips for those looking to get into voice. “Keep it simple,” she explained, not for developers, but for users. “The less you can ask users to do, the better. If you can get yourself in a situation where you’re not having to program the skill to represent the many, many ways the UK population pronounce the word ‘quinoa’, you’ll be in a much better position than we were.”

“Secondly, test and learn. We’re not at five stars yet. We’re constantly having to iterate and learn. You can’t iterate and learn from a product if you don’t build one.

“And lastly, make something people need. There’s a lot of hype around voice the moment. This isn’t about shoehorning your content into the voice landscape, but about creating something that truly enriches and adds value to the lives of your users.”