How to Use The Trade Desk to Reach New Households on CTV

The short answer: nothing on The Trade Desk starts without a contract, so the first move is the contact form or your agency's trading desk, and the free move is the Trading Essentials: Kokai course on Edge Academy (Edge Academy, checked on 5 November 2026); the CTV page advertises more than 185 million streaming households, and your budget buys a measurable slice of them (CTV page). Then: convert the budget into impressions and households, hash the customer list into UID2 so existing buyers are excluded, build a seed segment for Koa to extend, pick CTV inventory and a contextual group, cap frequency, let Koa run bids and pacing under a ceiling a person set, pull reports through the API, and judge the campaign against a measurement partner's number and a holdout rather than the platform's own before-and-after.

Disclosure: I may earn a commission through some links on this site. It does not change what this page says.

The job here is the one most mid-sized brands bring to programmatic first: reach people who look like your buyers on the big screen, without paying to show ads to the buyers you already have. The Trade Desk is where an agency does that for you, or where your own trader does it once a seat exists. What the platform is, what it calls itself and who it competes with is in the Trade Desk entry.

Before you touch the platform:

• Price the reach in households, not impressions. For example, EUR 20,000 at a EUR 10 CPM is 20,000 / 10 x 1,000 = 2,000,000 impressions; at a frequency cap of 4 that is 2,000,000 / 4 = 500,000 households, and at a cap of 2 it is 1,000,000. The cap, not the budget, decides how many homes see the ad.
• Count what you can upload. Emails raw or hashed, UID2s, EUIDs for Europe and the UK, cookies, device IDs and IP addresses are all accepted; phone numbers only in APAC (data onboarding guide). A CRM export of buyers with a consent record is the suppression list.
• Fix the yardstick. Scheduled reports land daily with a one-day delay; on-demand queries refresh every 30 minutes (reporting guide). Pick the outcome, the measurement partner and a holdout region before anything runs.

What you need

• A seat, your own or an agency's. The API guide says accounts are created by an Account Manager and that setup fees are discussed with a Business Development or Client Services lead (platform API guide).
• A signed service agreement if you will upload emails or phone numbers, unless you already hold UIDs.
• Video creative in the lengths your inventory partners accept, and a product feed if you want product-level measurement.

Step 1: Get in, or get in through an agency

The contact page asks for company type and area of interest and promises a call about partnership opportunities (contact page). The glossary's own definition of a trading desk, a unit within an agency that buys programmatic ads for clients, is the other door (glossary): if your agency has one, ask for a view-only login to the reports before you ask for anything else.

Step 2: Learn the interface for free

Edge Academy is a free course library with a registration wall, and Trading Essentials: Kokai is the course for the current platform (Edge Academy). Kokai has been the platform since June 2023 (press release); the Knowledge Portal behind it needs a platform login, so the course is the only manual you can read beforehand.

Step 3: Upload the customer list

The guide names the First-Party Data Uploader, a GraphQL tool, as the most efficient and flexible option, with a Python Data SDK and Snowflake and Databricks connectors beside it; the older CRM route is legacy (data onboarding guide). Send raw emails and let the platform normalise and hash them, or hash them yourself. Personal data sits briefly in a separate secured environment and is deleted once UIDs are created. UID2 itself is the open-source identifier built from a hashed, encrypted email or phone, and an advertiser's copy carries a refresh timestamp it has to honour (UID2 advertiser guide).

Step 4: Make two segments

One is suppression: everyone who bought in the last twelve months. The other is a seed: your top customers by value. The Koa write-up says it builds similar profiles from a seed and shows a relevance score for each (Koa article); the seed is what it extends from, so a seed of everyone is a seed of nobody.

Step 5: Choose CTV inventory and a contextual group

The CTV page promises frequency control across networks, channels and devices and a partner directory for inventory (CTV page). For context, open the Contextual (X) tile: segments from third-party partners and The Trade Desk's own, targeted or blocked by topic, grouped across providers with and/or logic, saved in a Contextual Library and reported on in a Contextual Performance report (contextual update). Providers build those segments from URL and keyword data (contextual tools). Set the cap from the arithmetic above.

Step 6: Let Koa bid under a ceiling

Koa refines bidding, pacing and delivery continuously and predicts the bid rate to avoid overpaying in first-price auctions (AI page, Koa article). The human check is the budget and the goal: a person sets both, and Koa moves inside them. The conversational assistant that would diagnose pacing and spend was in closed beta on the AI page when read, so do not plan around it.

Step 7: Pull the reports by API

The Platform API runs over REST and GraphQL and can create campaigns, ad groups and bid lists, automate activation and customise reporting (platform API guide). For this job use it to read: on-demand GraphQL queries return aggregated data every 30 minutes, hourly performance feeds arrive 2 to 3 hours behind, and raw event data streams to Amazon S3 or Snowflake near real time (reporting guide). An agent that drafts the Monday summary from the GraphQL response is a fair use of it; pausing an ad group is not, because the API can do that too and a person should press it.

Step 8: Judge it against something the platform did not make

The measurement page offers a marketplace of independent measurement providers and lets you plug in your own mix model or attribution (measurement page). Use one. The results the vendor shows for Kokai are spend-weighted averages from a beta, comparing the same campaign's KPI before and after adoption (Kokai results); a before-and-after with no holdout hands the platform everything that happened after the switch, the season included, and that is the shape a vendor's own measurement takes when the vendor also sells the media. Your number comes from the holdout region you fixed before launch.

What the vendor's pages flag

• Phone numbers upload in APAC only; EUIDs in Europe and the UK only.
• Email and phone uploads need a signed service agreement unless you already hold UIDs.
• The conversational assistant is closed beta; the AI page says not all listed capabilities are available.
• Reports are a day behind unless you take the feeds; budget decisions made on this morning's dashboard are made on yesterday.

How to know it worked

Take the four weeks before launch as the baseline for sales in the holdout and the exposed regions, run the campaign, and compare the lift in exposed regions against the holdout's movement over the same weeks. That comparison needs no tool beyond a spreadsheet. Later, if you want it run continuously against your own sales data rather than the DSP's report, a causal attribution tool like Causality Engine does it. Disclosure: I founded it.

FAQ

Is there a minimum spend?

None is published. The contact form and the API guide's mention of setup fees are the only signals (platform API guide); StackAdapt states no minimum on its plans page (StackAdapt plans).

Do I need UID2 to run this?

No. Cookies, device IDs and IP addresses are accepted identifiers too (data onboarding guide). UID2 is what makes a logged-in audience addressable where cookies are gone.

Is Kokai a separate product from The Trade Desk?

No. Kokai is the name of the platform since June 2023 and Koa is the AI inside it (press release).