--- title: "Year in Review: Design Decisions That Shipped" description: "Before-and-after comparisons, metric impacts, lessons learned, and honest reflections on what we would change — the annual design retrospective that holds us accountable." canonical: https://nowah.xyz/blog/year-in-review-design-decisions-shipped lastModified: "2026-08-07T08:03:52.787Z" --- # Year in Review: Design Decisions That Shipped Before-and-after comparisons, metric impacts, lessons learned, and honest reflections on what we would change — the annual design retrospective that holds us accountable. Shipping is the only design review that matters. Ideas in a design file are hypotheses. Shipped changes are experiments with results. This is our annual retrospective: the design decisions we made, what happened when they reached users, and what we learned. We are not presenting this as a victory lap. Some decisions worked better than expected. Some did not. The value is in the honest accounting, because the industry publishes enough "we shipped it and it was great" case studies and not nearly enough "here is what we got wrong and what we would change." ## Decision 1: three options instead of ten ![Illustration for this section](https://pics.nowah.xyz/website-media/design-073-img-1.webp) Before: our initial flight search results showed eight to ten options, matching the density that users expected from traditional booking platforms. The thinking was that more options meant more value. Users would feel they had thoroughly explored the market. After: we reduced the default to three curated options with a clear reason for each inclusion. Users could always ask for more. Impact: card view-through (the percentage of users who viewed all presented options) jumped dramatically. When shown ten cards, fewer than half of users scrolled through all of them. When shown three cards, nearly all users viewed every option. More importantly, the time from search to selection decreased substantially. Users made faster, more confident decisions. What we learned: more options create the feeling of thoroughness but the reality of paralysis. [Three options](/blog/why-three-options-not-three-hundred) with reasoning build more confidence than ten options without reasoning. The AI's curation is the value, not the quantity of raw results. ## Decision 2: streaming text instead of block delivery Before: the AI's response appeared all at once after generation completed. The user sent a message, saw a loading spinner, waited several seconds, and then received the full response as a block of text. After: text streams token by token at 30 milliseconds per token. The user sees the response being generated in real time. Impact: perceived [response time](/blog/ten-second-rule-ai-response-time) dropped by a large share even though actual response time was identical. Users reported the experience feeling "faster" and "more responsive." Session duration increased because users engaged with the streaming content rather than abandoning during wait times. What we learned: perceived performance matters as much as actual performance. Streaming is not just a visual trick. It fundamentally changes the interaction dynamic from "waiting for a result" to "watching the AI think." The trust implications are significant: users who watch the AI work trust the output more than users who receive it as a black box. ## Decision 3: the five-tab navigation with chat at center ![Supporting diagram](https://pics.nowah.xyz/website-media/design-073-img-2.webp) Before: early prototypes used a three-tab navigation with chat as one of three equal tabs. After: we moved to five tabs with the chat tab at the center position, visually differentiated with a slightly larger icon and accent color. Impact: new user engagement with the [chat interface](/blog/adapting-chat-interface-mobile-desktop) increased. The center position and visual distinction made the primary interaction point unmistakable. Users who might have explored other tabs first (and potentially churned before experiencing the core value) went directly to the conversation. What we learned: navigation design is not neutral. The position and visual weight of tabs creates a hierarchy that guides user behavior. Placing the primary experience at the center is a small change with a large behavioral impact. ## The decision we regret We shipped the onboarding [preference collection](/blog/preference-collection-as-onboarding) as optional, with a "Skip" button prominently placed on each card. The reasoning was that forcing preferences felt heavy-handed and might cause abandonment. The data told a different story. Users who completed onboarding reported significantly higher satisfaction with AI recommendations compared to users who skipped. The AI was measurably better for users who took thirty seconds to share their preferences. But many users skipped because the option was there and the value of completing onboarding was not communicated clearly enough. If we could redo it, we would make onboarding feel essential rather than optional. Not by removing the skip option, but by better communicating the direct connection between the preferences and the quality of subsequent AI responses. "Tell me these 5 things and I will find you better flights" is more compelling than "Set up your preferences." ## Metrics dashboard Our design metrics track six key performance indicators. Time to first actionable result: the seconds between the user's first message and the AI presenting bookable options. Target: under ten seconds. The streaming architecture helps here because the user starts seeing useful information before the full result set is ready. Booking completion: the percentage of users who start a booking review and complete the booking. This measures how well the review and payment flow works. The industry average for traditional platforms sits between 2 and 5 percent at the top of funnel. Our target is significantly higher from the review step, because users who reach review in a conversational flow have already been qualified by the AI. Voice adoption: the percentage of sessions that include at least one voice input. Higher adoption indicates the voice input design is discoverable and usable. Our target: over 30 percent. Onboarding completion: the percentage of new users who complete all preference cards. Higher is better for subsequent AI quality. These metrics are not vanity numbers. Each one connects to a specific design decision that we can iterate on. If time to first result increases, we examine the streaming and search flow. If booking completion drops, we examine the review and payment design. The metrics make design decisions accountable. ## Design decision tracking We recommend every design team implement a decision log. For each shipped design change, record: the hypothesis, the before state, the after state, the metric you are watching, and the result after a defined measurement period. This log creates institutional memory. When a new team member asks "why does it work this way?" the log provides the answer: "we tested the alternative, here is what happened." It also prevents repeated experiments. If a design direction was tried and produced negative results, the log prevents the team from unknowingly trying it again. Shipping is the experiment. Metrics are the results. The retrospective is the analysis. Together they form a design practice that improves with every cycle. --- Nowah is an AI travel agent that searches and books real flights and hotels through conversation — no filters, no thirty open tabs. [Plan your next trip](https://app.nowah.xyz).