HomePod, Alexa, and the voice-UI moment.

Apple announced HomePod earlier this year and the speaker ships in December. Amazon now sells five different Echo devices. Google Home is in real homes. Sonos has joined the party with Alexa-enabled speakers. Every brand brief we've had this quarter has included the words 'and a voice app'. The voice-first hype cycle has fully arrived.

We've spent the year building Alexa skills and prototyping for Google Assistant. Here's the honest field report — what works, what doesn't, what the platforms actually look like under the marketing, and what we tell clients when they ask 'do we need a voice app?'.

The smart-speaker market right now

What clients keep asking for

• 01An Alexa skill that lets users do what the website does, by voice.
• 02A voice-driven brand 'experience' that gets press attention.
• 03Some kind of Siri integration that we have to gently tell them isn't really a thing yet — HomePod is closed to developers at launch.
• 04A voice version of their customer-service chatbot.
• 05Voice ordering for e-commerce.
• 06'Skills the kids will use' — usually trivia games or audio stories.

What we keep telling them

• 01Voice is excellent for narrow, deliberate transactions — playing music, setting timers, asking for facts, simple commerce.
• 02Voice is bad at discovery. Nobody finds a brand by saying 'Alexa, what's a good Italian restaurant chain to order from'.
• 03If your product has visual states the user needs to compare, voice loses to a screen every time.
• 04Building a skill for press, not users, is expensive. The reviews never go well.
• 05Skill discoverability is broken — even the Amazon Skill Store is hard to browse, and almost no one browses it.
• 06Retention is dire. Industry data suggests around 3% of Alexa skills are used regularly after install.

"

Voice is an interface, not a category. The brands that succeed in voice in five years will be the ones whose products have always been ambient — music, weather, simple commerce. The brands that try to 'have a voice strategy' will quietly retire them in 2020.

Where voice actually earns its keep right now

• 01Smart-home control. Lights, thermostats, locks. The defining use case of the category.
• 02Hands-busy tasks. Cooking, driving, cleaning, exercising. The user is occupied and a screen would be inconvenient.
• 03Simple information. Weather, sports scores, currency conversion. The answer is short; the question is well-defined.
• 04Reorder / repeat purchases. Amazon's own data shows the highest voice-purchase rates are reorders, not new product discovery.
• 05Audio entertainment. Music, podcasts, audiobooks. The platform is the product.

What does not work yet

• 01Anything requiring nuance — booking a complicated appointment, comparing five options, undoing a mistake.
• 02Multi-turn conversations beyond two or three exchanges. Users give up.
• 03Voice as a primary commerce channel. The repeat ordering works; new discovery doesn't.
• 04Skills that need authentication. Account linking in voice is awful in 2017.
• 05Anything where misunderstanding the user has expensive consequences — medical, financial, legal.

Where to start, if you must

Pick a single, narrow transaction your customers already do every day. Build that one thing to a polished standard. Don't build a 'voice version of the website'.

Be honest about your audience. If your buyer is over 55 and lives in a flat in central London, very few of them own an Echo. If your buyer is 30-45 with kids and a house in the suburbs, smart speakers are increasingly real. Voice-strategy decisions need to be downstream of actual customer ownership data, not the CES news cycle.

Our prediction for the next three years

• 01Voice UI becomes ambient and useful, not a destination. Two-thirds of UK households will have at least one smart speaker by 2020.
• 02Apple's HomePod will sell well to existing Apple households and never crack the broader market.
• 03Amazon dominates US homes; Google catches up by being baked into Android phones; Apple stays in third.
• 04Most brand-driven voice 'experiences' will be quietly retired by 2020. The successful voice integrations will be invisible — your washing machine talks back, your car answers questions, your TV listens.
• 05Voice commerce remains a small fraction of overall e-commerce. The percentage of total revenue from voice will be measured in low single digits for the foreseeable future.

Voice UI becomes ambient and useful, not a destination. Three years from now it'll be a layer of every product, not a category of its own. Build accordingly — and don't ship the voice version of your website.