Here’s a selection of my work over the last few years:
2019: Confidential Client – Voice-first Music Assistant
This project’s brief was to design a first-party voice assistant, aligned to the company’s evolving brand by offering a unique streaming music proposition to help differentiate the client’s unique smart speaker hardware in a crowded market.
Significantly, the team already had a fully-working implementation of the smart assistant, complete with a personalised playlist creation algorithm, text-to-intent recognition engine and a phrasebook of responses. My role was to create a framework of user-centred design rigour around this existing experience.
This involved, in part:
- Developing new ways to harvest utterances – the assistant would only be as good as the ‘utterances’ (things people said) it understood. The recognition engine was relatively deterministic – it needed a good source of sample data to provide accurate intent matching. So, we developed new techniques to harvest these utterances from users around the world. We released a test app that would allow users converse with the engine, gave them a series of vague tasks (being careful not to bias their choice of words!) and captured how they phrased their questions to the assistant.
We later refined this technique by establishing a screener to ensure we were effectively targeting utterance harvesting from users in our core demographics.
- Diary studies to give longitudinal feedback – Voice experiences are notoriously difficult to test in the lab. Artificial sound conditions, embarrassment, the difficulty of conveying a task to be performed without biasing, plus all the other limitations that come with lab-based user research.
So, we developed and ran a diary study over three months to investigate the long-term adoption of our voice assistant. We were able to measure the effectiveness of the language we were using to describe the proposition, the reaction of users to the assistant’s ability to improve music recommendations over time, how much ‘personality’ we should convey in our tone of voice, etc…
- Personality distillation to create a voice of the brand – As the client developed and strengthened their marketing tone of voice, based on a newly-refreshed brand, we simultaneously created a personality guide based on the character traits of the company. This described how the assistant should react in given scenarios, what the assistant’s relationship to the user was… even how much the assistant should say vs. just getting on with the task it was given!
To facilitate the creation of the personality guide, we developed a process and workshopping technique to ensure the wider business teams were able to input on the creation of this new type of design documentation.
- Creating a phrasebook of responses – given the personality guide, I wrote hundreds of responses the assistant would say. The client’s response phrase construction engine was sophisticated and would allow for compound, randomised phrases to be created. This gave the assistant a much more life-like feel – humans rarely say the same thing in the same way twice, so why should our assistant!
In addition, we developed a copywriting process that allowed us to quickly generate conversational snippets to test the engine’s phrase structuring algorithm, and sense-check the responses, while then allowing us enough time to iterate and finesse the copywriting with SSML.
- Defining a rigorous engagement measurement & reporting framework – this gave us the insight into how our pre-release users were using the system: what they were asking, how often, phrase repeats, success rates, etc… Significantly, we wanted one or two simple measures of success to track on an ongoing basis (an idea adapted from Lean Analytics thinking), but finding that leading metric that indicated user satisfaction wasn’t obvious. We eventually settled on a compound metric, trigged from a defined pattern of user behaviour we considered to indicate successful behaviour. This required very close work with the data science, user research and engineering teams to expose the right data – and in sufficient quantities – to be effectively analysed to look for this trigger metric.
We also developed a post-release utterance triage plan to ensure the system was quickly and robustly trained on new utterances as the product is released to larger and larger numbers of users – with the aim of rapidly reducing the number of unsuccessful (unmatched) requests.
As of writing (Jan 2019), the product is unreleased, so I can’t say too much more. Suffice it to say, this will be a landmark new voice-enabled music experience unlike anything seen in the market before.
2018: Just Eat – Alexa Skill and Google Assistant Action
Working Just Eat’s product research and development team, we released the UK’s first e-commerce Alexa skill– ready on day 1 of Amazon’s release of the Amazon Echo device in the UK – and, later, the first Google Assistant Action to be based on Google’s new Transaction API and featured in the Google I/O keynote 2018.
For this project, we developed a while series of design firsts:
- Adapting wizard of ozresearch techniques to quickly validate concepts with zero-code.
- Created new voice UX documentation patterns, closely following atomic designprinciples.
- Worked with product copywriting, marketing and brand teams to create a voice style guide.
- Created a workflow to enable phrase response changes, version controlled and checked in alongside project code.
- Creating back-end tooling to facilitate easy localisation.
2017: Just Eat – Xbox One App
The goal of this project was to port Just Eat’s successful online ordering experience to a native Xbox One app. Analytics showed that large numbers of users were using the browser on their console to buy through Just Eat’s website. Not only was this a sub-optimal user experience, it wasn’t driving retention or higher engagement.
We conducted a series of needs-analysis research sessions with existing users to see how we could create a better experience. A key finding was that when users are deeply immersed in a gaming sessions (usually for several hours at a time) they didn’t want to switch focus away from the console.Â
So, along with adapting to the console controller as input mechanism, we established a project goal to let people order as quickly as possible and return to their gaming. By supporting XBox side-mode, storing payment details, and clever recommendations algorithms, we created a popular, profitable 1-click ordering experience that met this need, and through analytics, lab-based user research and in-home ethnographic insight, were able to quickly iterate the design to make it an effective experience.
For more information, please take a look at my short portfolio here: