Ttd bidding and optimization

The Trade Desk bidding and optimization

How The Trade Desk turns budget into won impressions against a KPI: the platform AI scores each impression for its value to your campaign, you steer that AI with a goal, a seed, and bid factors, and the system optimizes price, allocation, and pacing across the flight. This skill explains the publicly documented model so an agent can reason about a TTD campaign and give sound advice. The exact menu of bid controls, the optimization settings, and the numbers live in the partner platform behind a login, so this skill states the concept and flags where the operator has to confirm specifics in the product.

This skill assumes you know CPM, CPA, CPC, ROAS, win rate, and pacing. For those definitions and the KPI math, see the programmatic-foundations skill. For deciding what report proves a goal was met, see the reporting-by-campaign-goal skill. For where a campaign sits in the account, see the ttd-campaign-structure skill (sibling). For audiences and data that feed the bid, see the ttd-targeting-and-audiences skill (sibling).

When to use this skill

• "How does bidding work on The Trade Desk?" / "What is a TTD bid strategy?"
• "What is Koa?" / "What does Kokai's AI actually do?" / "How does Koa optimize my campaign?"
• "How do I optimize a TTD campaign?" / "Performance mode vs doing it by hand."
• "What are bid factors / seeds / KPI prediction / predictive clearing?"
• "How does TTD forecasting work?" / "Will this change help before I make it?"
• "How does pacing work on The Trade Desk?"

Boundaries with sibling skills:

• Audience definition, first- and third-party data, the data marketplace, contextual, and the seed's data side: ttd-targeting-and-audiences.
• Campaign and ad group structure, where bidding settings live in the hierarchy: ttd-campaign-structure.
• Inventory, deals, and supply paths such as OpenPath: ttd-inventory-and-deals.
• Identity (Unified ID 2.0, EUID) that underpins addressability: ttd-identity-and-uid2.
• Programming changes through the API: ttd-api-and-automation.
• Reading results and attribution: ttd-measurement-and-reporting.

Quick reference

The user wants	Concept that answers it	Where the specifics live
To understand how TTD bids	AI scores each impression by value to the campaign	Partner platform
To pick a goal for the AI	Set the KPI the optimization steers toward	Partner platform
To tell the AI what they value	Seed plus bid factors on chosen dimensions	Partner platform
To avoid overpaying	Predictive clearing against first-price clearing prices	Automatic, tune in platform
To hand execution to the AI	Performance mode runs Koa features within your guardrails	Partner platform
To test a change before committing	Forecasting shows projected impact first	Partner platform
To control delivery over the flight	Budget, flight dates, and pacing settings	Partner platform

The recurring pattern: the AI proposes value per impression and price, you constrain it with a goal, a seed, bid factors, and budget. The names and toggles for those constraints are in the product, so confirm exact labels there rather than guessing.

Core process

Use this to reason about or advise on a TTD campaign's buying and optimization.

1. Anchor on the objective and KPI first. The optimization is only meaningful relative to the goal it serves, so name the KPI (reach, CPC, CPA, ROAS, completed views) before touching bids. The reporting-by-campaign-goal skill maps objective to KPI set.
2. Confirm there is signal. AI-assisted optimization toward a conversion KPI needs conversion volume to learn from, the same constraint as any DSP. With thin signal, lean on a higher-funnel KPI or more manual control until data accrues.
3. Define what "valuable" means to the AI. The platform AI calculates the value of each impression to your specific campaign; you shape that by creating a seed (your ideal customer) and adding bid factors on the dimensions that matter to the brand. Hand the audience and data side of the seed to ttd-targeting-and-audiences.
4. Let predictive clearing set the price. The AI analyzes historical clearing prices across first-price auction environments so you win at efficient prices and cut wasted spend. Treat this as price discipline, not a separate KPI.
5. Decide how much to automate. Performance mode lets the AI handle execution (Koa Optimizations, audience expansion, predictive clearing, identity features) within the goals, KPI, seed, and targeting you set. You can refine inputs or override at any point. Choose more automation when you have signal and a clear KPI, less when you need tight, absolute control.
6. Forecast before you change. The forecasting engine projects the impact of an edit before you commit it, so model a change first instead of learning from spend. Confirm the exact forecasting surface in the platform.
7. Set budget and pacing to match the strategy. Even, goal-aware delivery gives the optimization room to learn and spread across the flight. The exact pacing options and their labels are in the platform; confirm them there.
8. Hold changes stable long enough to read. Frequent resets restart learning, so let an edit run before judging it, then verify against the KPI with ttd-measurement-and-reporting.

Decision rules and thresholds

The Trade Desk does not publish fixed numeric learning thresholds or bid-cap formulas on its public pages, and the live controls sit in the partner platform. Apply these public, concept-level rules and confirm the exact settings in the product.

• Match automation to signal. Conversion-goal optimization needs conversion volume. With little signal, optimize toward a higher-funnel KPI (reach, viewable, completed views) or keep more manual control until data builds. This mirrors the general DSP rule in programmatic-foundations.
• A goal is a target the AI steers toward, not a guaranteed hard ceiling. When you need an absolute price cap that can never be exceeded, you need an explicit control, so verify the cap setting in the platform rather than assuming the KPI target enforces it.
• Use a seed plus bid factors to express value, not blunt exclusions. Bid factors let the AI keep weighing every impression while leaning toward what you value; hard exclusions throw away opportunities the AI could have priced correctly.
• Let predictive clearing manage price in first-price environments. Overriding it with a rigid manual bid can forfeit the efficiency it finds across historical clearing prices.
• Change one lever at a time and forecast first. Stacking edits hides which one moved the KPI, and the forecasting engine exists precisely so you can predict impact before spending.
• Give every change a stabilization window before judging it. Reading a KPI too early, mid learning, leads to false negatives and premature reversals.

When a recommendation depends on an exact toggle, target field, bid-factor dimension, or numeric threshold, say so plainly: that specific control is set in the partner platform and the operator should confirm the current label and range there.

How the AI fits together

A plain-language model of the publicly described pieces, so an agent can explain them.

• Koa is The Trade Desk's AI, introduced in 2018 to help set up and optimize campaigns against business objectives. Kokai, launched in 2023, distributes Koa's AI across the buying workflow rather than confining it to a single step.
• Impression valuation. The AI considers each impression opportunity individually and calculates its value to your specific campaign, prioritizing the most relevant opportunities in a fraction of a second, at very high throughput (the public pages cite analysis of up to 15 million ad opportunities each second). "Value" is the sum of many AI calculations, publicly described as including KPI prediction, relevance, bid factors, and inventory quality.
• Seeds and bid factors are how you teach the AI. A seed defines your ideal customer; bid factors tell the AI which dimensions matter to the brand so it leans toward them when scoring. These are operator inputs set in the platform.
• Predictive clearing is price optimization. The AI analyzes historical clearing prices in first-price auctions so bids clear at efficient levels and waste less spend.
• Performance mode is the "let the AI execute" posture: you set goals, KPI, seed, and targeting, and the AI runs performance features (Koa Optimizations, audience expansion, predictive clearing, identity features) that adapt to live signals, with override available.
• Forecasting is the "look before you leap" surface: the forecasting engine projects the impact of edits and optimizations so you can understand them before committing.

Koa Audiences and Koa Optimizations are the publicly named optimization features. Their exact configuration and any newer feature names live in the partner platform; confirm current naming there.

Templates and examples

Advice framing, conceptual (confirm exact controls in the platform):

"Set your KPI to CPA, build a seed from your converter list, and add bid factors on the few dimensions you care about (for example device or content category) rather than hard exclusions, so Koa keeps pricing every impression. Let predictive clearing handle the bid in the first-price auctions. Before you change the goal, run the forecast to see the projected impact. The exact toggles and fields are in the platform, so confirm the labels there."

Triage framing when a campaign underdelivers against KPI:

"First confirm the objective and KPI are set correctly. Then check there is enough conversion signal for a conversion goal; if not, optimize to a higher-funnel KPI for now. Confirm budget and pacing are not starving delivery. Make one change, forecast it, and let it stabilize before reading the result. The specific settings to inspect are in the partner platform."

Common pitfalls

• Treating a KPI target as a hard price ceiling. It is a goal the AI steers toward; for an absolute cap, confirm an explicit control in the platform.
• Pointing conversion-goal optimization at a campaign with almost no conversions, then blaming the AI. With thin signal, start higher in the funnel.
• Smothering the AI with hard exclusions instead of bid factors, which removes impressions the AI could have priced well.
• Stacking several edits at once, so no single change can be credited or blamed.
• Judging a change before it stabilizes and reverting on noise.
• Inventing bid-factor names, menu labels, or thresholds. When the detail is platform-specific, say it is set in the partner platform and have the operator verify it.

Sources

• The Trade Desk, Our Demand Side Platform (DSP) overview: https://www.thetradedesk.com/our-demand-side-platform (as of June 2026)
• The Trade Desk, AI for Advertising (Koa, bid factors, seeds, predictive clearing, Koa Audiences and Koa Optimizations): https://www.thetradedesk.com/our-demand-side-platform/ai-artificial-intelligence (as of June 2026)
• The Trade Desk press release, "The Trade Desk Launches Kokai" (Koa history, per-impression scoring, KPI prediction, predictive clearing), June 6, 2023: https://www.thetradedesk.com/press-room/the-trade-desk-launches-kokai-a-new-media-buying-platform-that-brings-the-full-power-of-ai-to-digital-marketing (as of June 2026)
• The Trade Desk, Performance mode: AI-powered trading (goals, KPI, seed, automated price and allocation, override): https://www.thetradedesk.com/resources/outcome-driven-optimization-powered-by-ai (as of June 2026)

Note on sourcing: The Trade Desk's detailed product documentation and API reference sit behind a partner login and are not cited here. The exact bidding controls, optimization settings, and numeric thresholds require access to the partner platform.