--- title: "Why We Track Time-to-Book, Not Just Conversion Rate" description: Conversion rate is a lagging indicator. Time-to-book measures experience quality — from 45 sessions over 3 months to one conversation in minutes. canonical: https://nowah.xyz/blog/time-to-book-metric lastModified: "2026-08-07T07:57:39.369Z" --- # Why We Track Time-to-Book, Not Just Conversion Rate Conversion rate is a lagging indicator. Time-to-book measures experience quality — from 45 sessions over 3 months to one conversation in minutes. Every product team in travel obsesses over conversion rate. What percentage of visitors complete a booking? It is the number that gets reported in board decks, that gets benchmarked against competitors, that drives A/B test decisions. And it is the wrong primary metric for an AI travel product. Conversion rate tells you what happened. It does not tell you why. It does not tell you how the user felt. It does not tell you whether the experience was good or just good enough. Worst of all, it is a lagging indicator that moves slowly and hides the signal that matters most: how much work the user had to do to reach a decision. We track a different metric as our north star: time-to-book. How long does it take from the moment a user starts looking for a trip to the moment they complete a booking? Not in pageviews. Not in sessions. In actual elapsed time. This metric has changed how we think about every [product decision](/blog/dark-theme-product-decision) we make. ## The current baseline is absurd ![Illustration for this section](https://pics.nowah.xyz/website-media/product-049-img-1.webp) Let me lay out the numbers that motivated this shift. Google and Phocuswright research shows that the average traveler visits roughly many websites and engages in 45 or more sessions over the course of two to three months before making a booking. Two to three months. For a purchase that, once the decision is made, takes about five minutes to execute. Think about what those numbers represent. The actual transaction, entering payment details and clicking confirm, is trivially fast. What takes months is the research, comparison, deliberation, second-guessing, and eventual decision-making process. The product is not helping with any of that. It is just presenting options and leaving the user to figure it out. OTA [conversion rates](/blog/low-conversion-rates-ai-fix) have been stuck in the low single digits for over a decade. This means 95-98% of people who show up on a travel site looking to buy something leave without buying. Some of them will come back later. Many will buy on a different platform. Some will give up and not travel at all. Traditional product teams look at these numbers and optimize the funnel. Better search results. Smarter sorting. Urgency signals. Retargeting ads. Price drop alerts. All of these tactics are designed to push a user from "browsing" to "buying." Some of them work, incrementally. None of them address the fundamental problem: the product makes the user do all the work. ## What time-to-book actually measures Conversion rate measures outcomes. Time-to-book measures experience quality. When a user books a flight in 8 minutes, that tells you something specific. The user knew what they wanted (or the AI figured it out quickly), the right options were presented without a lengthy search-and-compare cycle, and the user felt confident enough to commit without extensive deliberation. When a user books a flight after 45 sessions over 11 weeks, that tells you something very different. The user was uncertain. The product did not help them make a decision. They left and came back repeatedly because each session produced more options but not more clarity. Eventually, fatigue won and they booked something that felt acceptable. Both users "converted." The conversion rate treats them identically. Time-to-book reveals that one had a dramatically better experience than the other. There is a nuance here. Very fast time-to-book is not always better. If a user books in 30 seconds because they felt pressured by a countdown timer and "only 1 seat left!" urgency messaging, that is fast but not good. Time-to-book needs to be paired with satisfaction and rebooking metrics to tell the full story. But as a leading indicator of experience quality, it is far more useful than conversion rate alone. ## How AI collapses the booking timeline The reason time-to-book is our north star is that it directly measures the value AI adds to the booking experience. Traditional booking is slow because the user bears the entire cognitive burden. They have to know which dates to search. They have to scan results and identify the best options. They have to compare across multiple dimensions (price, time, stops, airline quality, layover airports). They have to do this for flights and hotels and maybe car rentals and activities, each as a separate search-compare-decide cycle. No wonder it takes months. An AI agent compresses this process by handling the parts the user should not have to do. Preference application. The AI already knows you prefer direct flights, aisle seats, and morning departures from your past bookings and stated preferences. It does not present options that violate these preferences, eliminating entire categories of results you would have spent time reviewing and rejecting. Intelligent curation. Instead of 200 results that you sort and filter, the AI presents three options that represent meaningfully different tradeoffs. Best price. Best schedule. Best overall. You are comparing three options, not scrolling through pages. Context accumulation. The AI carries context from flights to hotels to activities. When you have booked a flight arriving at 9 PM, the hotel search already accounts for late check-in availability. You do not start from scratch for each booking component. Proactive information. Instead of you researching [visa requirements](/blog/ai-agents-visa-requirements-documents), weather, and safety information separately, the AI surfaces relevant information within the conversation. "You will need a tourist visa for India. The average temperature in Delhi in March is around 30C. Would you like me to search for flights?" Decision support. When you are stuck between two options, the AI can help. "The 6 AM departure is $120 cheaper, but given your stated preference for avoiding early mornings, the 10 AM option at $540 might be worth the premium. That is about $15 per hour of sleep." Each of these capabilities shaves time off the booking process. Together, they can reduce a multi-week, multi-session journey to a single conversation that takes minutes. ## Building a metrics framework around time-to-book We do not just track one number. Time-to-book decomposes into several components that each tell us something actionable. **Time to first search.** How long from conversation start until the AI executes a flight or hotel search? This measures how quickly the AI understands user intent and whether it asks the right clarifying questions without over-asking. **Time from search to selection.** How long does the user take to choose from presented options? If this is long, our curation is not good enough. If users frequently ask for more options or different options, the initial three were not well targeted. **Time from selection to payment.** How long between choosing an option and completing payment? Friction in the booking review, payment, and confirmation flow shows up here. **Session count.** Does the user complete the booking in one session, or do they come back? Multi-session bookings are not inherently bad (some trips require consultation with travel partners), but unexplained multi-session patterns suggest the product is not giving users enough confidence to commit. **Reopen rate.** How often does a user reopen a conversation and ask about the same trip without booking? This is the "still deciding" signal. It means the AI presented options but did not help enough with the actual decision. Each of these sub-metrics points to a specific product area to improve. Slow time-to-first-search means our intent parsing or clarification flow needs work. Long search-to-selection means our curation algorithm needs tuning. High reopen rate means we need better decision support. ## The benchmark: AI must beat self-service Here is the standard we hold ourselves to: if the AI cannot help a user book faster than they would on their own using Google Flights and Booking.com, the product is not working. This is a high bar. Self-service tools have gotten good at the mechanical parts. Google Flights is fast. Booking.com's search is efficient. If you know exactly what you want (LAX to NRT, April 5-12, economy, nonstop), you can find and book that in maybe 10 minutes on existing platforms. The AI advantage is not in the case where the user already knows exactly what they want. It is in the far more common case where they do not. "I want to go somewhere warm in March for about a week, nothing too expensive." On Google Flights, this query cannot even be entered. You need a destination to start. So you research destinations first, then search flights. That is where the weeks of browsing begin. In a conversation, the AI handles this immediately. "Based on your preferences and budget, I would suggest Cancun, the Algarve in Portugal, or Bali. Cancun has the cheapest flights from your area right now. Want me to search all three?" The user goes from vague intent to specific options in one exchange. That is the kind of time compression that justifies an AI-first product. We are not yet at the point where every single booking is faster than self-service. Complex multi-[city itineraries](/blog/launching-multi-city-itineraries-complex-planning) with specific requirements can still take several back-and-forth exchanges. But for the typical booking, we are seeing conversations conclude in under 15 minutes. Compare that to the industry baseline of 45 sessions over months. ## Why conversion rate optimizations can be counterproductive Here is the thing that really turned me against conversion rate as a primary metric: many of the tactics that increase conversion rate make the user experience worse. Urgency messaging. "Only 2 rooms left at this price!" This increases conversion rate because it creates fear of missing out that pushes uncertain users to commit. But it also creates buyer's remorse. Users who book under urgency pressure are more likely to cancel, less likely to rebook, and less likely to recommend the platform. Dark patterns. Making the "no thanks" button tiny. Pre-checking the travel insurance box. Hiding the total price until the last step. These all increase immediate conversion rate while degrading trust and long-term retention. [Decision fatigue](/blog/decision-fatigue-travel-science) exploitation. Showing users so many options that they eventually give up comparing and just pick something. This technically converts, but the user did not feel good about it. If you optimize for conversion rate, you end up in an arms race of manipulation tactics. If you optimize for time-to-book (while monitoring satisfaction), you end up building a product that is genuinely useful. The user books quickly because the product helped them make a good decision, not because it tricked them into a hasty one. ## Segmenting time-to-book Not all bookings should take the same amount of time. A [weekend getaway](/blog/weekend-getaway-data-fastest-growing) should be faster than a three-week honeymoon. A solo business trip should be faster than a multi-family vacation. We segment time-to-book to account for legitimate complexity differences. **Quick trips.** Domestic, short duration, solo or couple. Target: under 10 minutes. These are straightforward bookings where the AI should be able to search, present, and complete with minimal back-and-forth. **Standard trips.** International, 5-10 days, familiar destinations. Target: under 20 minutes. The AI handles flights and hotels, possibly with some preference discussion and option comparison. **Complex trips.** Multi-city, multi-week, group travel, unfamiliar destinations. Target: under 45 minutes. These legitimately require more conversation, more research, and more decision points. But 45 minutes is still dramatically faster than the 3-month industry baseline. **Aspirational trips.** Users browsing without firm intent. No time target. These are users who are dreaming, not deciding. Time-to-book is not the right metric here. Engagement and return rate matter more. By segmenting, we avoid the mistake of treating all bookings equally. A 30-minute booking for a simple domestic flight is a problem. A 30-minute booking for a complex international itinerary with three cities is excellent. ## What happens when time-to-book drops We have been tracking time-to-book since launch, and the trend is encouraging. As the AI gets better at applying user preferences from memory, the average time-to-book has been decreasing steadily for repeat users. First-time users take longer because the AI is still learning their preferences. Every clarifying question adds time. "Do you prefer window or aisle?" "What is your typical hotel budget?" "Do you have airline loyalty memberships?" These questions are necessary but they add minutes. By the third or fourth booking, the AI knows these answers. It stops asking. The conversation gets shorter. Preferences are applied automatically. The user sees three options that are already filtered through their personal preferences, and they choose faster because the options are more relevant. This is the compounding value of agentic memory applied to the time-to-book metric. Each booking makes the next one faster. And that creates a real retention loop: the platform gets better the more you use it, which makes you want to use it more, which makes it better. ## The metric that matters for AI travel products I am not saying conversion rate is useless. It is still a valuable output metric. But as a primary metric for driving product decisions, it fails AI-first products because it does not capture the value that AI uniquely provides. AI's value in travel booking is compression. Compressing research time. Compressing comparison time. Compressing decision time. Compressing transaction time. Time-to-book measures all of that directly. If you are building an AI travel product and your primary dashboard metric is conversion rate, you are optimizing for the wrong thing. You will end up building urgency signals and dark patterns instead of better AI curation and smarter decision support. Track time-to-book. Segment it by trip complexity. Decompose it into its component phases. Watch it improve as the AI gets smarter and memory accumulates. That is the metric that tells you whether your AI is actually making travel booking better, or just adding a chatbot to the same broken experience. The goal is not to convert more browsers into buyers through psychological tricks. The goal is to make the booking process so efficient and helpful that the gap between "I want to go somewhere" and "I have booked my trip" collapses from months to minutes. Time-to-book is how you measure progress toward that goal. --- 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).