How Sparkfare actually decides something is a deal

Methodology

Most sites that call something a "deal" don't say what that actually means. Ours does, because the definition is simple enough to just write down — and simple enough to check against real numbers, which is the whole point.

The core rule

A route is flagged a deal when today's cheapest fare is at least a cluster-specific percentage below that exact route's own trailing 30-day average price. That's it. No external market benchmark, no machine-learning prediction, no comparison against any other route or any other traveler's price. The baseline is entirely this route's own accumulated history — nothing more exotic than that.

The exact calculation

pct_below_avg = (trailing_avg − today's_cheapest_price) / trailing_avg

• trailing_avg is a straight arithmetic mean — not a median, not weighted or decayed toward recent days — of that route's cheapest daily fare over the preceding 30 calendar days.
• Routes are keyed by origin and destination together. A JFK–Lisbon average and an ORD–Lisbon average are entirely separate baselines; we never pool prices across origins.
• Today's own price never influences the average it's compared against. We classify against history as it stood before today, and only add today's price to the history file afterward — so the bar can't quietly get easier or harder to clear based on today's own number.

A route needs history before it can be judged

We require at least 7 distinct days of accumulated price history before a route can be classified as a deal — or ruled out as "priced, not a deal." Below that, it shows as Building history, visibly dimmed on the board, rather than being scored against a statistically unreliable average. A missing day is simply skipped, never backfilled with a guessed price, since a synthetic number would quietly corrupt the average everything else depends on.

The bar moves with how volatile a route actually is

Some routes swing in price far more than others on their own, with nothing unusual happening — a smaller dip on those routes isn't a meaningfully rare event, so we don't badge it as one:

Cluster	Threshold
Long-Haul Volatility	25% below average
Shoulder-Season Cliffs	15% below average
LCC Routing Anomalies	15% below average
Visual Clickbait	No threshold — always shown on its own merits, never badged as a deal

What this deliberately doesn't do

• No external market-wide benchmark — the baseline is only ever this one route's own history.
• No prediction of future prices.
• No weighting toward recent days — every day in the 30-day window counts equally.
• No cross-route or cross-origin comparison — a route is only ever compared against itself.

Commissions don't factor into any of this either. A route isn't shown as a deal because it pays us more, and a genuine deal isn't hidden because it pays us less — see our affiliate disclosure for the full picture on how we make money.

See today's board