Scoring
Why purchase-intent scores fail in high-consideration categories
Journal · Switch Routehub · reading time about ten minutes
Intent scores copied from fast-moving e-commerce assume that delay is death. In appliances, insurance, education, and some healthcare adjacencies, delay is the work. Households compare, ask an uncle, wait for a bonus cycle, visit a shop, return to the PDP three weeks later. A score that treats that stall as drop-off will spray discounts at people who were going to buy anyway and will suppress people who were still deciding.
Where the score goes wrong
Most off-the-shelf intent models overweight recency of PDP views and add-to-cart. In a Bangkok electronics extract, add-to-cart was often a bookmark. Removing an item was not rejection; it was list hygiene. The model called those users “cooling.” Sales associates in store still closed many of them after a weekend visit — a channel the score could not see because identity across web and store was optional.
Cross-Channel Intent Mapping exists for this gap. It does not pretend identity is solved. It requires an honest unknown bucket. Pretending every store visit joins to a login is how intent scores become fiction.
Stall as a named state
In the Atlas language, stall is a route plate, not a failure. Inclusion rules might read: multiple spec-sheet views spanning more than ten days, no support-ticket anger, no competitor-price scrape pattern if you have it. Exclusion: single-session cart abandon after a shipping-fee shock — that is a different name, with a different offer.
The third axis here is rarely frequency. Frequency of PDP views in high-consideration goods can mean anxiety. We have used “days since last human consult” (chat, call, or store) as a third axis more than once. It is operationally ugly and commercially clearer than a 0–100 intent gauge.
Discounts make the score worse
If the only action attached to “high intent” is a coupon, you train customers to wait for the coupon, which then inflates recency and teaches the model that discounters are intentful. Brand leads like this loop because conversion week looks busy. Margin does not.
If your category considers for months, do not buy a score trained on snacks. Recalibrate or refuse. Recency–Frequency Recalibration covers the RF part; intent is the cousin that fails in the same calendar and the same identity fog.