The Twelve-Cent Briefing
On paper, watching 30 web pages every day for a week costs about twelve cents in AI model calls, because the pipeline sets aside trivial changes before any model runs. The bill follows the changes that matter rather than the pages watched.
Watching 30 web pages every day for a week costs about twelve cents in AI model calls, because the pipeline sets aside every trivial change before a model is called.
The figure is arithmetic at published model prices, not a bill we have paid. WatchBase, the monitoring engine we are building, runs this pipeline; its narration step, the one that costs money, has not run against a live model. What the arithmetic shows is where the cost of the thinking sits.
The Worked Example
Thirty pages across five hypothetical competitors in a professional services market: pricing pages, team pages, service descriptions, news sections and careers pages. One check per page per day for seven days makes 210 checks. The aim is to surface the changes that matter and set aside the rest.
Most Checks Find Nothing Worth Reporting
Illustrative: a hypothetical week of 210 checks (30 pages, daily, for seven days). The shares are the worked example's, not a measured run.
In the example, 151 of the 210 checks find no change. Another 38 find changes such as new whitespace, a formatting tweak or a reordered FAQ, which score below the significance threshold and are set aside. Fifteen changes are substantive enough to narrate, and six need action. Only those 21, one check in ten, ever reach a model.
What the Briefing Says
Three examples of what a briefing could say about a week like this. They show the shape of the output; they are not findings from a run.
The Bill
Four of the five steps cost nothing in model fees.
210 page checks
$0: a browser fetches each page (Playwright, open source, runs locally).
210 comparisons
$0: comparing old and new text is plain code, with no model.
210 significance scores
$0: a small classifier scores each change on your own server.
21 narrations
$0.005 each = $0.105: only substantive changes reach the model.
1 weekly briefing
$0.01: one call compiles the week into a summary.
The model runs 21 times. At half a cent per narration (an estimate at published model prices) that is 10.5 cents, and compiling the briefing adds about one more: $0.115 a week, about fifty cents a month.
What It Changes
Competitive intelligence used to need a budget line. Enterprise tools are priced for the analyst hours and manual monitoring they replaced.
Priced for analyst hours: annual contracts, onboarding, and someone on your side to interpret the output. Published prices are rare, and the tier that watches competitors is seldom the entry tier.
About $0.50 a month in model calls on the arithmetic above (not a measured bill), plus hosting. The briefing arrives in Slack or email.
The model runs only after the cheaper steps confirm that something worth narrating has happened, so the bill follows the number of significant changes rather than the number of pages watched.
Start with the list: the thirty pages whose changes would alter a decision, such as competitors’ pricing, their careers pages and your regulator. The risk is a threshold set too high, which sets aside a change that mattered, so for the first weeks someone reviews what the filter discarded.
Which pages would you want watched every day?
Book Free Assessment