Marketers who search for a log-level data programmatic agency are usually trying to answer one question: can I see, impression by impression, what my budget bought? Log-level data is that record. It is a file with one row for every bid won, every impression served, every click, and every conversion, each carrying a timestamp, the site or app, the exchange, the deal, the price paid, and a pseudonymous identifier for the device or household. A programmatic agency that provides it is giving you the means to check its work. One that will not is asking you to rely on its summary.
This article explains what log-level data contains, how the major demand-side platforms make it available, the seven checks that justify the effort, and the contract language that secures your access. It is written for a marketing leader who will not query the files personally but needs to know what to require and what the answers should look like.
What Is Log-Level Data in Programmatic Advertising?
Standard reporting is aggregated. A dashboard shows that a campaign delivered 4.2 million impressions at a $31 average CPM with 0.9 percent invalid traffic. Each figure is an average or a total computed by someone else, using rules you did not set.
Log-level data, sometimes called event-level or impression-level data, is the raw material behind those totals. A typical impression row includes:
- When: timestamp to the second
- Where: domain or app bundle, publisher ID, and for streaming TV, the app and sometimes content genre or channel
- Path: exchange or supply-side platform, seller ID, and deal ID if bought through a private deal
- Price: bid price and clearing price (the media cost for that single impression)
- Fees: platform fee, data fee, and other costs attached to the impression, where the platform exposes them
- Who: a pseudonymous device ID, cookie ID, or household ID, plus IP-derived geography
- How: device type, operating system, creative ID, ad format, and player size
- Quality: viewability and completion events, and verification flags where integrated
Conversions and clicks arrive in separate files that can be joined to impressions on the identifier. A month of activity for a mid-sized advertiser can run from tens of millions to several hundred million rows.
How is log-level data different from a standard report?
| Standard platform report | Log-level data | |
|---|---|---|
| Granularity | Totals by day, campaign, site | One row per event |
| Who defines the metrics | Platform or agency | You |
| Frequency analysis | Average per campaign | Exact, per household, across campaigns |
| Price visibility | Average CPM | Price of each impression |
| Supply path | Usually top-level exchange | Exchange, seller, and deal per impression |
| Joins to your own data | Limited | Possible, in a privacy-compliant environment |
| Effort to use | None | Requires storage and an analyst |
Why Should a Luxury Advertiser Care About Log-Level Data?
Independent studies of the programmatic supply chain explain the concern. In 2020, ISBA and PwC attempted to trace advertiser spend through to publishers in the UK and could fully match only 12 percent of the impressions studied; 15 percent of advertiser spend, the "unknown delta," could not be attributed to any party. A 2022 follow-up, with better data access, reduced that unknown delta to about 3 percent. The Association of National Advertisers' 2023 study of US programmatic spend found the average campaign ran on roughly 44,000 websites and that about 15 percent of spend went to made-for-advertising sites.
The lesson from the ISBA work is practical: the unexplained share fell when log-level data became available. Access is what made the supply chain auditable.
For a luxury or high-consideration brand, three features make this more pressing than for a mass advertiser.
Audiences are small. If a campaign targets 120,000 qualified households, a small number of misdirected impressions is a large share of the plan, and uneven frequency is costly.
Media is expensive. At $30 to $55 CPMs for premium streaming TV, a 10 percent leak is a meaningful sum.
Environment is part of the product. A summary report that lists the top 50 sites by spend does not show the long tail, which is where unsuitable placements sit.
What a Programmatic Agency Can Actually Provide, by Platform
Access differs by demand-side platform, and an agency cannot give you what its platform does not release. This is a summary of general availability; confirm current terms with each platform, since products and fees change.
| Platform | Log-level product | What you receive | Notable limits |
|---|---|---|---|
| The Trade Desk | Raw Event Data Stream (REDS) | Impression, click, conversion, and video event files delivered to cloud storage | Some identifiers and publisher fields are redacted by region or by supply partner |
| Google Display & Video 360 | Data Transfer files | Event-level files delivered to Google Cloud Storage | User identifiers are restricted; household-level joins are done inside Ads Data Hub |
| Amazon DSP | Amazon Marketing Cloud and Amazon Marketing Stream | Event-level data queryable inside a clean room; near-real-time aggregated delivery metrics | Raw rows do not leave the clean room; outputs are aggregated above minimum thresholds |
| Other independent DSPs | Varies | Typically impression-level exports or reporting APIs on request | Field depth, cost, and retention vary; ask for a sample schema |
Two points follow from the table.
First, "log-level access" covers two models. In the export model, files land in storage you control. In the clean-room model, you run queries against event-level data inside the platform's environment and receive aggregated answers. Both support most of the checks below. The clean-room model limits joining across platforms.
Second, the walled gardens of social and some retail media do not offer equivalent data. A log-level program covers the open programmatic portion of a plan, which for most of our clients is the majority of working media.
What does a log-level data programmatic agency do with the files?
Receiving files is the easy part. An agency that is serious about log-level data will:
- Enable the feed at campaign launch, since most platforms do not backfill
- Store the data in a warehouse the client can access, such as BigQuery or Snowflake
- Maintain a documented set of standard queries and run them on a schedule
- Share findings and the changes made because of them, in plain language
- Hand over the data and the queries if the relationship ends
Seven Checks Worth Running on Log-Level Data
These are the analyses we run for clients, in rough order of value.
1. True household frequency
Count impressions per household per week across every campaign and line item. Platform averages hide the distribution. A campaign with a reported average frequency of 6 often has 10 to 15 percent of households receiving 25 or more impressions and a third receiving one. In our reviews, impressions above a reasonable weekly cap commonly account for 15 to 30 percent of spend. Redirecting them to unreached households is usually the largest single gain.
2. Placement audit, including the long tail
List every domain and app with impressions and spend, then compare against the approved inclusion list. Flag anything outside it, anything declared as one app but showing device or bundle patterns of another, and streaming impressions that ran in screensavers, mobile apps, or low-quality channels. For a luxury brand the target is 95 percent or more of spend on approved properties.
3. Supply path and duplicate auctions
Group impressions by publisher and by the exchange and seller ID that delivered them. Many publishers sell the same impression through several intermediaries. Compare clearing prices by path for the same publisher; differences of 10 to 30 percent are common. Consolidate onto the direct or cheapest authorized path, validated against the publisher's ads.txt and the exchange's sellers.json files.
4. Price paid and fee reconstruction
Sum clearing prices to get true media cost, then add platform, data, and verification fees as exposed in the log. Compare the total with what you were invoiced. The gap is the agency's margin plus any undisclosed costs. In a transparent arrangement, the gap equals the stated agency fee.
5. Private deal performance
For each deal ID, compare bid requests received, bids submitted, wins, and average clearing price against the negotiated floor. Deals that rarely win, or that clear well above floor, need renegotiation. Deals that deliver outside the agreed inventory need to be raised with the publisher.
6. Timing and geography
Check impressions by hour and by postal code. Heavy delivery between 1 a.m. and 5 a.m. local time, or concentration in areas outside the target geography, suggests poor pacing or invalid traffic that verification missed.
7. Path to conversion
Join conversion events to impression history for each converting household. This shows how many exposures preceded a conversion, over how many days, and on which channels, and it provides the exposed-household list needed for incrementality analysis. It does not prove causation by itself; pair it with a holdout group.
What Does Log-Level Data Cost, and When Is It Worth It?
Costs fall into three parts.
- Platform fee. Some platforms include event-level feeds; others charge a flat monthly fee or a small CPM surcharge. Budget in the low thousands of dollars per month where a fee applies.
- Storage and compute. For an advertiser spending $100,000 a month, cloud warehouse costs typically run a few hundred dollars a month.
- Analysis. The main cost. Initial setup takes 20 to 40 analyst hours, and a monthly review takes 6 to 12.
As a rule of thumb, a log-level program is worth running once programmatic spend passes about $50,000 a month. At that level, recovering even 10 percent of spend from excess frequency and inefficient supply paths covers the cost several times over. Below roughly $25,000 a month, a quarterly placement and frequency audit from standard reports is usually sufficient.
Privacy and Compliance
Log files contain pseudonymous identifiers and IP-derived location, which are personal data under the GDPR and under US state privacy laws such as the California Consumer Privacy Act. Requirements to build in:
- A data processing agreement covering the agency and any warehouse provider
- Access restricted to named analysts, with identifiers hashed or truncated where full values are not needed
- A retention limit, commonly 13 to 25 months
- Customer file matching performed inside a clean room, never by loading customer records beside raw logs in an open environment
- Honoring of opt-out signals passed in the bid stream
For wealth management and other regulated categories, have compliance review the data flow before the feed is switched on.
Contract Terms to Require From Your Programmatic Agency
Access that depends on goodwill tends to disappear when it is most needed. Put it in the agreement.
- Seat and data ownership. The advertiser owns, or has an irrevocable license to, all campaign data, including log-level files, audience definitions, and deal IDs.
- Feed activation. Log-level feeds are enabled from the first day of activity on every platform that offers them.
- Delivery location. Files are delivered to storage the advertiser controls or can access directly.
- Disclosed costs. Platform, data, verification, and agency fees are itemized on every invoice, with no undisclosed margin on media.
- Audit rights. The advertiser or its appointed auditor may review platform invoices and log data on reasonable notice.
- Retention and handover. Data is retained for an agreed period and transferred in full at termination.
Questions that reveal whether an agency really works with log-level data
- Which of your DSPs provide event-level data, and in what form?
- Can we see a sample schema and one of your standard frequency queries?
- What did your last log-level review change in a client's campaign?
- Who on your team writes the queries?
- Will the data sit in our warehouse or yours?
Specific answers come quickly from a team that does the work. Vague ones are informative too.
Common Mistakes
- Requesting logs after the campaign. Most feeds are not retroactive.
- Collecting without analyzing. Storage with no scheduled review produces cost and no benefit.
- Expecting logs to prove incrementality. They describe delivery. Causation needs a control group.
- Ignoring redacted fields. If 30 percent of rows lack a publisher domain, say so in the findings and press the platform or supply partner on why.
- Treating one platform's data as the whole plan. Household frequency across DSPs requires a common identifier or a clean room.
How Stillwater Media Handles Log-Level Data
For engagements at sufficient scale, we enable event-level feeds at launch, hold the data in an environment the client can access, and run the seven checks above on a fixed schedule. Findings are reported with the action taken: households capped, paths consolidated, deals renegotiated, placements removed. Platform and data fees are itemized, and the data belongs to the client.
We take on a limited number of engagements each quarter, for luxury and high-consideration brands with customer lifetime value above $5,000 and sales cycles longer than 30 days. If you want a programmatic partner whose work you can verify row by row, apply to work with Stillwater Media.
About Stillwater Media
Stillwater Media is a selective performance media agency for luxury and high-consideration brands. We plan and buy premium CTV on services including Disney+, Netflix, and Prime Video, along with programmatic video and display, digital out-of-home, streaming audio, YouTube Select, podcasts, native, and social, using private marketplace deals, affluent audience engineering, and incrementality testing. We work with brands in private aviation, luxury real estate, wealth management, private clubs, luxury automotive, premium DTC, and luxury hospitality. Signal. Strategy. Scale.
Frequently Asked Questions
What is log-level data in programmatic advertising?
Log-level data is a raw file with one row for every impression, click, and conversion a campaign generates. Each row records the time, site or app, exchange, deal, price paid, device details, and a pseudonymous identifier, which lets an advertiser calculate its own metrics instead of relying on platform summaries.
Should my programmatic agency give me log-level data?
Yes, where the demand-side platform makes it available, and the right to it should be written into the agency agreement. An agency that enables the feed at launch and delivers it to storage you can access is giving you the ability to verify placements, frequency, and fees independently.
Which DSPs provide log-level data?
The Trade Desk offers the Raw Event Data Stream, Google Display & Video 360 offers Data Transfer files, and Amazon DSP offers event-level analysis inside Amazon Marketing Cloud. Other independent DSPs generally offer impression-level exports or APIs on request, with field depth and cost that vary by platform.
How much programmatic spend justifies a log-level data program?
A scheduled log-level program generally pays for itself above about $50,000 a month in programmatic spend, where recovering 10 percent of budget from excess frequency and inefficient supply paths exceeds the cost of storage and analysis. Below about $25,000 a month, a quarterly audit from standard reports is usually enough.
Can log-level data prove that advertising caused sales?
Not by itself. Log-level data shows which households were exposed, how often, and what they did afterward, but proving causation requires comparing exposed households with a similar holdout group that did not see the ads.
