Two ad tech announcements this week describe deals that adjust themselves while they run. Before anyone raises a budget on a self-optimizing deal ID, the buyer should get a written change log from the vendor, and I think that request belongs in the insertion order.
What is changing in the deal
Adnami’s page for Express Supply Agents promises “Instant multi-publisher deals, continuously optimised for attention and other KPIs,” and says deals “don’t have to be static” before it describes live optimization. We covered the product in Adnami Adds Express Supply Agents, a Self-Service Tool for Attention-Optimised Deals. Pinterest’s announcement of Programmatic Deals says its inventory sits inside Magnite Orchestration, a layer that connects agents so that buyer and seller agents can discover, evaluate, package, activate and optimize inventory. We covered that in Pinterest Makes Programmatic Deals Available Through Magnite and Four DSPs.
A deal that optimizes itself has a different shape from the fixed placement lists most media plans assume. The site mix on day twenty can differ from the mix on day one, and nobody on the buyer’s side decided the change.
The case for leaving it alone
The strongest objection is that the DSP already reports delivery by site, and that removing manual list management is the reason to buy an agent. Adnami’s page also says buyers can exclude individual sites directly in the DSP, so a control exists. Magnite’s Sean Buckley, President, Revenue and Market Strategy, framed the goal in Pinterest’s announcement this way:
“By combining those insights with our demand relationships, activation capabilities, and omnichannel publisher footprint, we can help buyers connect discovery with measurable results across more media environments.”
Measurable results are what a buyer wants, and delivery reports measure them. I agree with that much.
Why a delivery report is the wrong record
A delivery report shows where impressions ran and what they cost. It does not show when the agent moved spend off a placement, which target it was working toward at the time, or whether the buyer changed the target in between. Those facts decide whether a result can be repeated. If a finance team asks why attention costs rose in week three, or a test needs a control group, the answer depends on knowing which version of the deal produced each impression.
The vendor pages I read describe the optimization and the DSP charges. Adnami’s page says it does not invoice for the tool because the DSP charges once a deal is activated. Neither the Adnami page nor the Pinterest announcement describes a record of agent changes. That gap can be closed in the contract.
What to ask for in writing
Four items cover it. First, a timestamped log of every placement the agent added or removed, with the rule that triggered it. Second, a statement of which optimization target is live, and a requirement that the buyer approves any change to it. Third, an export of the exclusion list the agent honors, so it matches the list in the DSP. Fourth, a way to freeze the deal at its current mix for the length of a test.
How to use the log in a test
The cleanest design I can suggest is two deals on the same DSP seat with the same target and the same starting site list. Freeze one and let the other optimize. At the end of the flight, the log tells you which placements the optimizing deal dropped, and the delivery reports tell you what the freeze cost or saved. That comparison is a better basis for a budget decision than a vendor’s case study, and it works for any agent that packages inventory, whatever company sells it.
Buyers who skip the log still get results. They just cannot say why one flight beat the next, and the next flight starts from a guess.
This follows from the argument in Agentic Buying Needs a Buyer Agent No Seller Owns: when agents act on both sides of a deal, the buyer needs its own record of what happened. LG’s seller agent news in LG Ad Solutions Opens Home Screen Inventory to Magnite’s Seller Agent shows how quickly that supply is arriving.
Run the first test with the log in place, and raise the budget after the log has explained one swing in cost per outcome.
Source: Pinterest Newsroom; Adnami
