Closed-Loop TV and Streaming Attribution: What Can Actually Be Measured - Enterprise Digital Marketing

Television and streaming advertising produce no click and no lead record, so the closed-loop methods that work for search and lead channels do not apply. Vendor attribution for TV relies on household matching, which inherits the same inflation problem as direct-mail matchback. This paper lays out the three routes that produce defensible numbers for an awareness channel, using a measured case from a multi-location group: branded search impressions up 41 percent against a generic control that fell 9 percent, priced at the group's own revenue per inbound call.

August 21, 2026 · 8 min read · By Nick Martinelli

The phrase "closed-loop TV attribution" describes a product category, not a measurement. A closed loop means a specific transaction in the operator's own ledger is tied to a specific exposure to the advertising. For search advertising that is achievable, because a click carries an identifier that can be written onto the lead and carried to the ticket. For television, whether broadcast, cable, or streaming, there is no click. The viewer sees a spot and, some days later, does something: types the brand name into a search engine, opens the maps app, or calls a number they already had. Every one of those actions registers in the operator's systems as organic, branded, or repeat activity. None of them carries the spot with it.

This paper is for operators who are funding, or being pitched, a streaming or linear TV flight and want to know what can honestly be measured. The answer is more than nothing and less than what the vendor deck shows. We work through it with a measured case from the locations we manage.

What the vendors measure, and why it inflates

Streaming TV attribution products generally work one of two ways. The first is household matching: the ad server records which households (by IP address or device graph) were served the spot, and a later visit, call, or transaction from a matched household is credited to the campaign. The second is automatic content recognition on the television itself, which logs exposure and matches it to the same downstream events.

Both produce a number that looks like a closed loop. Both share a structural problem that readers of our direct-mail matchback work will recognize: the credit rule is "exposed, then transacted," with no requirement that the exposure caused the transaction. In the mail case, a vendor credited 53 percent of a location's entire revenue to a mail program, while the location's own front-desk source tags attributed 7.8 percent. The mechanism of the inflation was that the mailing list was built from the operator's own customer file, so the households most likely to transact anyway were the households most likely to be "matched."

TV household matching runs the same way when the targeting is built from a customer list or a lookalike of one, and it runs a softer version of the same way when the targeting is geographic, because the households served are the households that live near the location and were already its most probable customers. A streaming vendor reporting that matched households transacted at several times the rate of unmatched households is reporting, mostly, that their targeting found the right neighborhoods.

The defensible questions to ask any TV attribution report are the same five we use for mail: what fraction of credited transactions were existing customers, what the credit window is, what the control group is and how it was chosen, whether the lift is stated against the control or against zero, and whether the credited revenue can be reconciled to the operator's ledger. A vendor that cannot answer the third question has not measured lift.

The three routes that produce a number

Because a TV exposure leaves no identifier, the loop cannot be closed at the individual level. It can be closed at the aggregate level, three ways, each with a stated confidence.

Route one: branded search lift against a generic control. A TV flight that works changes one thing immediately and measurably: more people search for the brand by name. Google Search Console counts branded organic impressions whether or not the operator is bidding on the term, daily, for sixteen months back. Measured before and after the flight, against generic organic impressions for the same site as a control, the branded series isolates the awareness effect from seasonality and site-wide changes. This is the route worked below.

Route two: call-log lift with dynamic number insertion. Inbound call volume is the operator's own record and is not subject to vendor matching. A flight with an on-off schedule (two weeks on, two off, repeated) produces a call-volume signature that can be read directly from the call log, with a lag of one to three weeks. If the site runs dynamic number insertion, branded-session calls become separable from the rest, and the branded-search route and the call route can be joined.

Route three: the front-desk source tag. The desk-tag loop asks every new customer how they heard of the location and writes the answer onto the ticket. It undercounts awareness channels badly, because customers do not remember spots, and the measured tag rate at the best-instrumented location in the book was 9.8 percent of first-time callers. It is still the only route that produces a ticket-level record, and a rising "saw your ad" tag count during a flight is corroborating evidence even if it is not the measurement.

None of these is a per-transaction closed loop. Route one produces a modeled revenue figure with stated assumptions. Route two produces a measured call figure and a modeled revenue figure. Route three produces a traced but heavily undercounted figure. A channel report should show all three and label each by type.

The measured case: an awareness flight at a four-location group

We have not yet run a linear or streaming TV flight for a location in the book with enough pre-flight Search Console history to measure cleanly. We have run the measurement on a continuous social-video awareness flight, which belongs to the same class for attribution purposes: no click carried to the site, paid for on impressions, intended to produce branded demand later. The method transfers directly to TV and streaming, and the numbers below are what a well-instrumented flight of that kind looks like.

Setup. Four locations under one distinctive brand name with no geographic or dictionary double meaning, so a substring on query text classifies cleanly (157 unique branded queries). The flight started at roughly $100 per day across the locations in October 2025 and has run continuously. Search Console history began in April 2025, giving six months of baseline.

Design. Pre period, 25 full weeks. Post period, 40 full weeks after a three-week ramp. Control series: generic organic impressions for the same site.

Welch t on the branded series is 6.95. The branded series stepped up in the first full month, peaked above 1,000 per week, and settled onto a plateau near 600 to 640 per week that has held for nine months. Same-month comparisons against the prior year are positive in every overlapping month.

Confound check. Paid branded impressions in the search account fell over the same period, which in principle could push clicks from paid to organic. The magnitude rules it out: paid branded was 55 to 285 per month before the flight against 1,184 to 2,494 organic branded, and the organic increase is roughly 800 per month.

Pricing. Monthly, branded clicks rose from 157 to 285, an increment of 127 per month after generic clicks are netted. The group's own call log and ledger put blended revenue per inbound call at $273 ($3.24 million of completed revenue across 11,867 inbound calls) and a first-time-caller floor at $107. At a 25 percent click-to-call rate, the incremental branded clicks are worth $3,400 per month on the floor and $8,700 on the blended basis, against roughly $3,000 per month of flight spend. The full scenario table and the reasoning behind each input are in the companion paper.

Reading a TV flight with this method

Three things change when the flight is television rather than continuous social video, and all three make the measurement easier.

TV flights have edges. A continuous flight has one edge, the start, and the lift it produces is permanently confounded with whatever else launched that month. A TV flight bought in two- or four-week blocks has many edges, and each one is a test. The signature to look for in the weekly branded series is a rise beginning one to two weeks after a block starts and a decay beginning one to three weeks after it ends. Two or three clean on-off cycles are worth more than a year of always-on data.

TV flights are geographic. A multi-location operator buying streaming by designated market area can run the flight in some markets and hold others out. The held-out markets' branded series is a better control than generic impressions, because it absorbs brand-specific events (a press mention, a competitor closing) that generic impressions do not. The method in our revenue-lift paper uses the same geographic-control logic on transactions.

TV spend is large enough to see. The social flight above was $3,000 per month across four locations and produced a lift that was statistically unambiguous. A streaming flight at five to ten times that spend, if it works at all, should produce a branded signal that is visible without a t-test. If it does not, the operator has learned something the vendor report would never have shown.

What to report

An operator funding a streaming or TV flight should expect four lines in the channel report, and should reject a report that collapses them into one:

1. Branded organic impressions and clicks, weekly, with the flight schedule overlaid and the generic or held-out-market control on the same chart.
2. Inbound call volume from the operator's own call log, weekly, same overlay.
3. The vendor's matched-household figure, labeled as vendor-reported, with the five audit questions answered alongside it.
4. A modeled revenue figure from the branded-search route, with the call rate stated as a range and the revenue-per-call basis stated explicitly.

Line four is the number that belongs in the budget conversation. Lines one and two are the evidence for it. Line three is context.

Methodology notes

• Search Console Search Analytics API, web search type, query by date, 479 days ending late July 2026, 227,033 query-date rows, single-substring branded classification.
• Pre period 2025 weeks 15 through 39; post period 2025 week 43 through 2026 week 30; weeks 40 through 42 excluded as ramp. Welch's t-test on weekly aggregates.
• Call log: 11,867 inbound calls over six months from the group's call-tracking export. Completed revenue from the point-of-sale per-ticket export for the three locations with ledger access in the same window.
• Flight spend from the operator's confirmed daily budget. No platform-reported conversions and no vendor-reported attribution are used anywhere in the measured case.
• All figures anonymized across the locations we manage.