As a service dedicated to supporting its content creators, Twitch needed to create an alerts experience that made it easier for creators to entertain an audience, build a community, and make a living.

In order to ensure the new experience could effectively attract adoption, my team and I focused on providing differentiating value to creators, building strong, familiar foundations, and reducing creator work.

Twitch Alerts would also be the first alerts system operated by a live-streaming service provider, so we sought to connect creators and viewers in new ways by using alerts in direct combination with the live viewing experience on web and mobile.

Our goal was to develop and grow new alerts software that would drive increases in channel engagement and non-recurring revenue for established creators.

At first glance, the alerts space didn’t appear ripe with opportunity, even for a core service provider like Twitch.

Established creators were accustomed to relying on 3Ps for most of their software and hardware needs — in fact, the vast majority of Twitch creators already had alerts and were satisfied with them. The software they used was mature, fully-featured, and offered much of their functionality for free. By contrast, Twitch’s prior entry in the streamer-tools space had lacked the power and customizability of peer offerings and had been largely disregarded by successful streamers.

It was clear that driving adoption of a new alerts system would be challenging, so we went to our customers for a deeper understanding. I wrote a UX research plan, and my PM partner and I conducted 20 interview sessions — 10 with medium-sized creators and 10 with viewers. Our objective was to better pain points relative to their existing alerts experience, where they wanted more, and what specifically they were looking forward to from a Twitch-authored alerts experience.

In light of our research results, we now had several hypotheses as to how Twitch alerts might provide differentiating value to creators. However, given the amount of time and effort required to build a full alerts system, we needed further confirmation that alert design could amplify the emotional value of supportive moments in ways that shifted viewer behavior. We used focused live experiments to test our research hypotheses.

Our first experiment was called “Celebrations.” I worked closely with a senior engineer to programmatically design a set of animated fireworks, which used channel emotes as particles and were drawn on-screen using a canvas layer. In order for this display to effectively add emotional value, we felt it was important for the firework animations to feel authentic — down to the movement of the particles, the drag, decay, and hang-time — so they could invoke the remembered awe of real fireworks.

For the experiment itself, we worked directly with a small cohort of ~60 channels over a duration of six weeks. When a viewer in these channels purchased a large Community Sub Gift, a series of emote fireworks would appear. The size of Sub Gift that invoked a Celebration was relative and determined on a channel-by-channel basis.

At the end of six weeks, we found that Sub Gifting spend in participating channels showed a directional increase of ~6%.

In our second experiment, we used Community Gifting to test whether adding the ability to express sentiment with support would change monetary support behavior. As a proxy for alerts, I designed animated themes for Community Gifting user-notices in chat, which could be selected by users via a low-friction interface at point-of-purchase.

This experiment was rolled out to 20% of monetizing channels and generated a directional increase in Community Gift revenue of ~3%. Learn more about this experiment here.

A full alerts system has a lot of moving parts. My team and I took on the challenge of designing a structure for those parts that made it all feel simple. This meant integrating the power and functionality that creators wanted into a small set of surfaces that were simple, familiar, easy to navigate, and reliably conveyed data relationships.

In modern alerts systems, it was expected that each type of live event could be supported by multiple custom alerts. Each alert could be configured to appear only when a certain condition was met, which ensured recognition and celebration provided by the alert was contextually appropriate. For example, “Subscription” events are differentiated by measures of loyalty and commitment, so their alerts would proportionally recognize and celebrate different levels of tenure (brand-new, 1+ years, 2+ years, etc.) and subscription tier (Tier 1, 2, or 3).

We spent time early in the process exploring ways to represent and customize events, alerts, and conditions.

We landed on a familiar three-column layout that used accordion UI to represent all events and their alerts in the left column. In order to make workflows efficient and easily learnable, we strove to minimize “rabbit-holed” content and maintain task-to-surface consistency throughout the app. We emphasized a large, responsive preview area so that the effect of any change would be immediately visualized, and provided contextual information throughout to help ensure small decisions remained true to the creator’s vision.

We knew it would be a tall ask to get creators to migrate their existing alerts to a new system. Our approach was to focus on building an alerts system that was “familiar, but better.”

While it was essential that we provide creators with differentiating value versus their current alerts, we also had to deliver the foundational features and capabilities that creators felt they needed. This approach would both effectively serve customers and also build trust with creators that we were listening and understood their needs, contrary to their current expectations. We made sure to provide them with alerts features they highly valued, and those they considered essential for any alerts system. In our initial launch, over thirty customizable settings were available for every alert, as well as support for custom HTML and CSS.

In terms of how these alerts visuals and sound were integrated into their live entertainment, creators expected alerts to integrate into the their video stream, and to be able to control if and where alerts appeared in their various OBS scenes. We chose a solution that would both meet the expected standard and be familiar to the large established creators for whom we were building, in order to help de-risk their decision to adopt.

A key part of the alerts product strategy was to build differentiating features that provided material value to creators. In a saturated market, only differentiating value would motivate creator customers to migrate from their existing, satisfactory solutions.

For starters, Twitch Alerts gave communities more to celebrate by adding net-new event types such as Hype Trains, charity contributions using Twitch's new Charity tool, Multi-Month Subscriptions, Subscription Goals started and achieved, and custom Channel Points redemptions.

To help creators make large contributions and big moments more of a spectacle, we integrated the Celebrations test as a Twitch Alerts feature. The updated feature included three styles of emote effects — (Fireworks, Emote Rain, and Rock ‘n Roll Flamethrowers — at multiple levels of intensity. Each style was carefully designed to evoke the awe and excitement of its real-life counterpart, while the display was programmatically driven so each celebration is unique.

Arguably the most innovative feature in Twitch Alerts was viewer-customizable alerts, also known as Interactive Resubs.

This concept had excited the team from the beginning. If the alerts and the viewer experience are connected, creators can use alerts to seamlessly create engaging interactive experiences for viewers.

Unfortunately, the reality wasn’t as simple as, “the alert appears and they can click on it.” Alerts in live channels are stochastic and time-bound, making them a poor starting point for a new interactive behavior, especially when it would only apply to some alerts on some channels.

But what about something like Themed Sub Gifting -- viewer interaction before the alert event, which actually customizes the alert itself? We solicited feedback using a simple prototype and creators reacted very enthusiastically to the idea. They highly valued a new way to engage their community and help their subscribers feel special.

We were wary of the friction inherent to adding new viewer-facing complexity and choice, so we focused this feature on resubscription events. Resubs were the lowest-friction support event on the viewer side, and targeting these viewers guaranteed us experienced users with strong channel affinity.

At launch, Interactive Resubs allowed creators to configure a separate set of visual and sound assets, from which viewers could choose the assets used by the alert that celebrated their resubscription event.

While our research (and our years of experience with the Twitch service) made us confident that creators will customize their alerts, we didn’t want our defaults to need customizing in order to be usable. Instead, they should be able to do their job out-of-the-box.

My team collaborated with Twitch’s brand designers and our product-facing motion designer to create sets of fit-to-purpose animations that embodied the energy of the live on-stream events they would represent.

At launch, every new Alert Box came preloaded with 14 default alerts across 9 different events, each with custom high-quality animations designed specifically for their alerts. The alerts asset library included another 23 high-quality animations and 6 sound effects, all free for creators to use.

Established creators typically have scores of alerts across all supported events. In our qualitative research, several of the creators we interviewed expressed a desire to change their alerts or update them to match their current branding... but they hadn’t found the time. This was because existing alerts software required creators to edit each alert individually, making any large-scale changes very time-consuming.

Twitch Alerts addressed this issue by introducing bulk editing. Creators can select any number of existing variants and edit any common attribute — text color, layout, text-to-speech voice, etc. — across all selected variants with a single action. During a bulk edit, creators can preview any selected alert to see how the changes are impacting that specific variant.

While we were very proud of the systems we had created, we hadn’t forgotten that established creators expected Twitch’s creator tools to be underpowered, below bar, and not useful to them compared to third-party options.

We anticipated a strong narrative at go-to-market that would resonate with creators, but we also needed a way to keep that story fresh and reach those who had missed our initial launch. To this end, we designed a rich landing page within Twitch’s creator dashboard that could  keep creators updated on new features and improvements, and continue resetting creators’ prior expectations and facilitating adoption throughout the service’s life cycle. This page also served as the starting point for alerts management and customization by adopting creators.