As Senior Product Designer at Readdle, I worked on Spark — an email client used by nearly 16 million people across iOS, iPadOS, Android, macOS, and Windows. My scope covered three intertwined initiatives: the +AI Composer feature that shipped as part of the Spark 3.0 launch, an early exploration of Meeting Notes with AI (leveraging Spark's calendar integration), and the design system rebuild that unified all five platforms under a single token-driven language.
Readdle is a Ukrainian productivity company behind Spark, PDF Expert, Calendars, and Scanner Pro. Spark itself competes in a dense category — Gmail, Superhuman, Shortwave, Apple Mail, Outlook — where the product has to earn attention every day. The Spark 3.0 launch in December 2023 was the biggest product moment in Spark's ten-year history: a full visual redesign, subscription business model, and OpenAI-powered AI features baked into the core email workflow, not bolted on as a separate assistant.
Core product foundations
⌘F
Smart Search — natural-language search across every connected account, not just keyword matching.
⌘K
Quick Actions — a command palette that surfaces snooze, delegate, pin, and send later without leaving the keyboard.
Done
One tap archives the entire thread, not just the latest message — the core loop behind Spark's zero-inbox philosophy.
Before any AI work started, Spark already had a distinct point of view on inbox velocity. Smart Search (⌘F) and Quick Actions (⌘K) exist so a power user almost never needs to reach for the mouse. "Done" archives a full conversation in one tap — not a single message — which is the small, repeatable action that keeps a Smart Inbox actually empty rather than just visually tidy.Layered on top: Gatekeeper screens unknown senders before they reach the inbox, Pin / Snooze / Set Aside give three distinct ways to defer something without losing it, and a genuinely unified inbox lets a user run several accounts through one interface instead of switching apps. AI features had to sit inside this existing behavioral model, not replace it — that constraint shaped every decision that followed.
Challenges & key areas for improvement
Adding AI to a mature email product is harder than adding it to a new one. Every design decision runs into the same wall: users have muscle memory built over years of Gmail, Apple Mail, and Spark Classic. Introducing AI in a way that helps without disrupting that muscle memory was the central design tension of the entire Spark 3.0 project.
Three specific problems shaped the work. First, most AI writing assistants live in a sidebar or a separate panel — a mode you enter, use, and exit. That framing communicates "this is a special AI tool." We wanted the opposite: AI as an ambient capability inside the normal compose flow, not a destination you visit.
Second, generated content often sounds generic. Even when the AI is technically correct, the tone gives it away — polite, hedged, over-formal. For an email client, that's a trust failure. Nobody wants to send messages that read as "AI-written" to people who know them.
Third, Spark ships on five platforms — iOS, iPadOS, Android, macOS, and Windows — with product velocity that had to accelerate, not slow down, during the redesign. Any inconsistency in the AI interaction pattern between platforms would erode confidence in the whole feature.
Shared drafts
Product discovery: earning the right to add AI
40+
user interviews across Free, Premium, and Team subscribers before a single AI screen was designed.
2
core personas identified during discovery, each with a fundamentally different reason to open Spark every morning.
Before any AI work started, the team ran a discovery phase built around one question: what job is a person actually hiring Spark to do? We didn't want to add AI in a bolted-on, gimmicky way — a marketing checkbox rather than something that earns its place in a daily workflow used by nearly 16 million people. That distinction, bolted-on versus native, became the filter every AI feature had to pass before it got a design review.Through structured interviews, usability sessions, and behavioral data from the existing user base, two clear personas emerged.
Persona 1 — the everyday all-in-one user
This person doesn't run their business out of Spark. They use it occasionally for work correspondence, but what they're really hiring an email client to do is unify their life — a personal calendar and a work calendar under one roof, one inbox that doesn't demand a separate app for scheduling. Their ROI isn't measured in dollars; it's measured in fewer context switches per day. For this persona, an efficient all-in-one tool matters more than any single AI feature.
Persona 2 — the B2B / SMB operator
Runs or works inside a small-to-medium business — sales, support, agency, ops — where email is the primary revenue and client-communication channel. This is where Spark for Teams (Pro and Enterprise plans) earns its subscription price. ROI for this persona is concrete: hours saved per rep per week, faster first-response time, a consistent voice across a whole team's outbound. These two personas didn't just inform which features got built — they determined how the same AI capability had to behave differently depending on who was using it.
Approach & solutions
Adding AI to a mature email product is harder than adding it to a new one. Every design decision runs into the same wall: users have muscle memory built over years of Gmail, Apple Mail, and Spark Classic. Introducing AI in a way that helps without disrupting that muscle memory was the central design tension of the entire Spark 3.0 project.
Three specific problems shaped the work. First, most AI writing assistants live in a sidebar or a separate panel — a mode you enter, use, and exit. That framing communicates "this is a special AI tool." We wanted the opposite: AI as an ambient capability inside the normal compose flow, not a destination you visit.
Second, generated content often sounds generic. Even when the AI is technically correct, the tone gives it away — polite, hedged, over-formal. For an email client, that's a trust failure. Nobody wants to send messages that read as "AI-written" to people who know them.
Third, Spark ships on five platforms — iOS, iPadOS, Android, macOS, and Windows — with product velocity that had to accelerate, not slow down, during the redesign. Any inconsistency in the AI interaction pattern between platforms would erode confidence in the whole feature.
Email summarization
Long threads compress into a two-sentence summary at the top of the conversation. The model reads the full thread history, not just the latest message, so the summary updates as new replies land — the difference between opening a 40-message thread and skipping it entirely.
AI Assistant — your private secretary
Beyond one-off compose actions, the AI Assistant sits across the whole inbox as a standing capability: surfacing what needs a reply today, drafting responses to routine requests, and answering questions about your own mail — "what did the client say about the deadline?" — without scrolling or searching. Internally the brief was never "chatbot." It was the assistant a busy person wishes they had, minus the headcount.
AI writing assist — compose & reply
The +AI Composer works in both directions: drafting a new email from a short prompt, and generating context-aware replies inside an existing thread. Reply suggestions read the whole thread rather than the last line, so a three-message negotiation gets a reply that actually accounts for what was already agreed.
My Writing Style
The retention unlock. My Writing Style analyzes a user's own sent mail and adapts every AI-generated draft to their tone, sentence length, and level of formality, instead of one generic "AI voice." It's the feature that turned the AI Composer from a novelty into something people trusted enough to send with minimal editing.
Comments
AI email templates
AI email templates matter less for any single message and more for a team's known email flow. Outbound sales is the clearest case: the first email in a sequence is courtship — courteous, discovery-oriented, listening for the client's specific need. Everything downstream — sending the contract, chasing a signature, confirming a kickoff date — follows a predictable pattern with almost no room for creative judgment. Those are exactly the emails a sales or ops team can delegate to AI with confidence, because the variable isn't the wording, it's the data.
Spark for Teams: collaboration built for Pro & Enterprise
Pro / Enterprise
the two Spark for Teams plans built specifically for small and mid-sized businesses running email as a shared, revenue-facing channel.
4
core collaboration surfaces — Team Comments, Shared Drafts, Shared Templates, and Shared Emails — that turn a personal inbox into a coordinated team channel.
For the B2B persona, the AI Composer and My Writing Style are only half the story. The other half is Spark for Teams — the collaboration layer that makes the case to a finance stakeholder easy: less time duplicating communication across tools, faster response times, and a consistent voice across every rep.
Team Comments
A private side-conversation on a specific email, with @-mentions for context, replacing the instinct to paste a client email into Slack and fragment the thread across five separate reply-alls.
Shared Drafts
Real-time, Google-Docs-style co-editing on outbound email — a manager and a rep building the same reply together, which matters most for exactly the high-stakes messages where a second set of eyes lowers risk before send.
Shared Templates +AI
Where the outbound-pattern logic becomes a team asset rather than one person's personal shortcut. Templates are created once, shared across the team, and every rep sends a consistent, on-brand message instead of reinventing the wording each time.
Shared Emails (internally "Posts")
Replaces the habit of screenshotting or forwarding a client email into Slack or Microsoft Teams. A rep pastes a link, and the whole team gets full context on the original thread without a noisy cc chain.
AI summary
AI summary
Results & Impact
The Spark 3.0 launch was the biggest release in the product's history. The redesigned inbox and inline +AI Composer landed in December 2023 and immediately drove the strongest week-over-week active user growth Spark had seen in three years. 68% of Premium users tried the AI Composer within 60 days — and more importantly, 71% of those returned to use it again within the same week.
My Writing Style, released in February 2024, was the feature that unlocked long-term retention. In the same 60-day window post-release, drafts generated in a user's own voice were sent (rather than deleted or rewritten) at 2.3× the rate of the earlier "neutral tone" drafts. The trust gap closed almost overnight.
Meeting Notes with AI shipped as an early preview to Team subscribers and is now being explored as a broader Spark feature. Its most cited benefit in user interviews: replying to "how did the meeting go?" emails without having to type from scratch, based on notes the AI captured directly from the calendar view.
The design system rebuild delivered the least visible but most compounding value. Cross-platform bug reports tagged as "design inconsistency" dropped 62% year-over-year in 2024, and per-feature design-to-ship time on new features (the whole point of a design system) dropped from ~8 weeks average to ~5 weeks average across all five platforms.
AI summary
Conclusion
Working on Spark taught me that adding AI to a mature product is a design problem before it's a model problem. The technical capabilities of GPT are impressive, but they're not what determined whether users trusted or rejected the AI Composer. What determined it was where the button lived, what happened after they tapped it, and whether the output sounded like them.
The three initiatives — +AI Composer, Meeting Notes exploration, and the design system rebuild — reinforced the same lesson from different angles. AI belongs inside the flow, not beside it. Trust is built through voice, not features. And a design system is only durable if it's built from tokens, not components.
Spark now serves nearly 16 million users across five platforms, competing with products that spend an order of magnitude more on marketing. What kept us competitive wasn't any single feature — it was the discipline of shipping AI that felt native to email, not native to AI.
If you'd like to talk about product design at scale, AI-native interfaces, or cross-platform design systems, I'd love to hear from you.