How Nomadiq tracks fares 24/7 (without spamming airlines)

Every Smart Booking on Nomadiq is built on a fairly simple premise: somewhere in the next few weeks, the fare for your trip will be lower than it is today. Our job is to figure out when, by how much, and whether the dip is worth waiting for.

This post walks through what that actually looks like behind the scenes.

Step 1 — collecting fare history

We sample fares for popular routes multiple times a day. We don't scrape — we use partner APIs that pass through to the GDS layer the airlines themselves publish to. Each snapshot records the lowest available fare, the cabin class, the day-of-week, the days-before-departure, and a handful of other features.

A typical popular route — say Mumbai to Delhi — has hundreds of thousands of price points in its history. Less-flown routes have less data, which matters later.

Step 2 — predicting the future from the past

Once we have a history series, the prediction model takes:

• The current price
• The shape of the last 6 months of fares
• Day of week, days before departure, seasonality
• Macro signals — fuel, capacity changes, festival calendars

…and outputs two numbers: a predicted minimum price for your travel date, and a confidence score between 0 and 100.

We're not trying to beat the market. We're trying to give an honest answer to the question "is the current price as good as it's going to get?" If the model says yes with high confidence, we book.

The confidence score matters as much as the predicted price. A low-confidence "predicted ₹4,500" is less useful than a high-confidence "predicted ₹5,200" — because the second one is something we'll actually act on.

Step 3 — the decision to book

Every active Smart Booking is checked on a schedule. The booking trigger fires when:

1. Live fare ≤ user's target price, OR
2. Live fare is within 2% of the model's predicted floor AND confidence ≥ 70, AND
3. The seat is still in inventory at the displayed price (we re-verify before charging the card).

If we book and the price subsequently drops further — say a flash sale hits the day after — we refund the difference. The user's prepaid amount stays the same; the refund goes back to the original payment method.

Why this is hard to do well

Three reasons:

1. Inventory drift. Displayed fares and actual bookable fares diverge by the minute. We re-query at the moment of booking to make sure the price we're about to charge is real.
2. Cold-start routes. For tier-3 routes with thin history, the model doesn't have much to work with. We fall back to a simpler "wait for X% drop from current" heuristic until enough data builds up.
3. The user's window. A Mumbai–Dubai prediction for September isn't useful if the user is travelling in June. The model has to reason about your date range, not the average history.

What we don't do

We don't sell ad space. We don't take referral fees from airlines. We don't surface "sponsored" fares above cheaper ones. The convenience fee on the booking is the only thing we make on the transaction — which is why our incentive is genuinely lined up with finding you the lowest fare.

If you want to see it work, start a Smart Booking for any route you're considering. The flow walks you through preferences, shows the live prediction, and gives you a target before any money moves.