Choosing a people-based DSP: identity, households, and outcomes

What is a people-based DSP?

A demand-side platform becomes "people-based" when it resolves impressions to persistent real-world identities rather than transient device signals. Standard programmatic buying identifies a bid opportunity by device ID, cookie, or IP address. Each of those signals is session-level and device-specific. A person who moves from a laptop to a phone to a connected TV generates three independent device signals. The platform cannot know they are the same person, the same household, or the same buyer.

People-based DSPs address this by anchoring device signals to a more durable identity layer, typically one of three approaches:

• Email-based hashing: The publisher or advertiser provides a hashed email address (or phone number) that matches to a persistent identifier. The user's email was collected consensually and is constant across devices. This is the approach underlying Unified ID 2.0 (UID2).
• Household graph: Deterministic or probabilistic linkage groups all devices registered to a single residential IP or connected via a shared Wi-Fi network into a household unit. This is Viant's primary architecture under its Household ID technology.
• First-party commerce and login graphs: Platforms with large authenticated user bases, like Amazon's shopping and Prime ecosystem, match advertising impressions to known purchasers and browsing histories without any external identity solution.

None of these approaches uses third-party cookies, which are irrelevant in CTV environments and increasingly absent on the open web. People-based buying matters most precisely where cookies never existed: connected TV, mobile apps, and cookieless browsers.

Identity graphs vs cookie and device graphs: what actually changes

The practical difference is durability and cross-surface coherence. A cookie expires, is cleared, or is blocked. A device ID changes when a user resets their ad ID or switches hardware. Neither travels across surfaces without additional matching logic.

An identity graph replaces the device signal with a persistent key that remains stable across sessions and devices. When that key resolves at the household level, it enables two things that cookie-based buying cannot do reliably:

• Cross-device frequency management: A household exposure cap that applies across the CTV, the tablet, and the desktop, rather than three separate per-device caps that sum to over-exposure without any platform knowing it.
• Attribution across surfaces: Connecting a CTV ad impression to a downstream web conversion on a different device, using the same household key as the common thread rather than a probabilistic fingerprint.

Cookie-based DSPs approximate this with IP matching and probabilistic device graphs. The approximation degrades in two real conditions: shared IPs (apartment buildings, co-working spaces) and mobile, where IP is dynamic and cookies are sandboxed per-app. People-based identity is a more stable input, but it depends on the publisher or advertiser collecting consensual authenticated signals in the first place. That supply of authenticated impressions is the actual constraint on scale.

Platform comparison: capability dimensions across The Trade Desk, Viant, Amazon DSP, and Yahoo

The four platforms most associated with people-based or household buying take meaningfully different approaches to identity, CTV, and measurement. The table below names each approach honestly; every cell is method-level or primary-sourced. Where a capability is unverified from primary vendor documentation, the cell says so rather than inferring.

One thing the table makes clear: these platforms are not equivalent substitutes. The Trade Desk and UID2 are the most portable across publishers and DSPs. Amazon DSP is the strongest choice where commerce data is the audience signal. Viant's Household ID is the most granular residence-level approach for CTV-heavy programmes. Yahoo's position in this field depends on Logan Wright login-graph scale that the other three cannot replicate from Yahoo properties specifically.

Why household identity matters most in CTV buying

CTV is structurally cookieless. There is no browser to drop a pixel on, no third-party cookie to set, and no app-level ad ID (like IDFA or GAID) that travels across apps on the same device. Each streaming app on a connected TV is, for tracking purposes, an isolated environment unless the identity layer solves the cross-app linkage.

Household identity solves this in a way that per-device identity cannot. A household watching a streaming app is likely doing so on a shared television. Multiple household members may use the same physical device across different sessions. Per-person targeting at the device level will miss or misattribute. Household-level targeting sets frequency caps, measures reach, and attributes outcomes at the unit that actually matters for most brand campaigns: the home.

The operational consequence is concrete. Without a household layer, a campaign capped at 5 impressions per device can deliver 5 impressions to the CTV, 5 to the tablet, and 5 to the laptop of the same household, for a total of 15 exposures to the same set of people. With a household layer, the cap applies across all three devices and total delivery stays at or below the intended 5.

For direct-response objectives, household attribution connects the CTV impression to the conversion. When a viewer sees a CTV ad and later converts on their phone, the household key is the common thread. Without it, the conversion registers as organic or dark social and the CTV campaign receives no credit.

How to evaluate a people-based DSP: the dimensions that determine fit

The four platforms differ enough that platform selection is a genuine decision, not a commodity choice. The right dimensions to evaluate depend on your programme's actual requirements, not on which platform has the most polished sales materials.

Five dimensions drive most of the selection logic:

• Identity source: Where does your audience come from, and which identity approach matches that source? If your advertiser has a large first-party CRM, UID2 is the most portable activation path. If your audience is defined by purchase behaviour, Amazon DSP's first-party commerce graph is the strongest match. If your programme is CTV-heavy and residence-level frequency management is the core objective, Viant's Household ID is purpose-built for that case.
• Inventory access: Are you buying named publisher environments (streaming networks, broadcast apps, premium SVOD) or optimizing across the broadest possible CTV supply? The Trade Desk's direct publisher relationships and OpenPath supply access give it breadth. Amazon DSP offers privileged access to Amazon's own streaming inventory. Viant's Publica integration expands supply to premium CTV publishers. Inventory access determines whether you can actually buy what your campaign requires.
• Measurement architecture: Does your measurement objective require commerce outcomes (Amazon's DPVR, NTB, AMC event-level queries), incremental lift at the household level (Viant's iROAS), cross-channel open-web attribution (UID2's portability), or a combination? The measurement system is not an afterthought; it should be selected before the platform, because the platform determines which measurement is possible.
• First-party data activation: Can you bring your own customer data and match it to the DSP's identity graph? UID2 is the most open path for this: advertisers transmit UID2-based segments to participating DSPs without relying on the DSP's proprietary graph. Amazon DSP supports first-party data uploads matched against Amazon's own identity. Viant accepts first-party data for Household ID matching, though the graph is proprietary and the data does not port out.
• Operational complexity and access model: Amazon DSP's managed service requires a published minimum spend threshold. The Trade Desk and Viant are accessed through direct relationships or managed-service partners. Self-service access varies by market and minimum. For programmes in Singapore, Malaysia, Australia, the US, or Canada, verify current access paths directly with each platform, as managed-service and self-service availability differs by region.

The cross-link to the CTV PMP and programmatic guaranteed setup guide is relevant here: deal type (open auction, PMP, programmatic guaranteed) is a separate decision from DSP selection, but the two interact. Some PMP relationships are exclusive to specific DSPs because of publisher-SSP partnerships. Verify deal availability on your chosen DSP before negotiating the publisher deal.

Outcome measurement in people-based programmatic: what each approach enables

Measurement is where people-based DSPs diverge most sharply from standard programmatic platforms, and where the choice of platform most directly constrains the questions you can answer after a campaign.

Standard programmatic measurement reports impression delivery, click-through rates, and view-through attribution using a last-touch or fractional model tied to device-level cookies. In CTV, none of those signals exist. The platform is reporting into a measurement void unless it has built a cross-device attribution layer on top of its identity graph.

The measurement approaches available across the four platforms reflect their identity architectures:

• Viant (Household ID): Closed-loop household measurement matches residence-level CTV impressions to downstream conversions on any device within the household. Tracks completion rate, view-through conversions typically within a 1-to-7-day window, ROAS, and incremental ROAS. This is the most self-contained measurement stack of the four, and the most CTV-native.
• Amazon DSP (first-party commerce graph): Proprietary metrics including detail page view rate (DPVR), add-to-list, and new-to-brand. Brand lift and offline sales lift studies for eligible advertisers. Omnichannel Metrics (OCM) and Marketing Mix Modeling (MMM) available at higher tiers. Amazon Marketing Cloud provides event-level dataset access for custom SQL-based analytics. This is the deepest commerce-outcome measurement of the four, but it is most powerful for advertisers selling on Amazon or within Amazon's ecosystem.
• The Trade Desk (UID2): Open-web cross-device attribution using UID2 as the common key across publisher environments. Frequency and suppression management across channels. Third-party measurement integrations available through the open marketplace. The Trade Desk's Kokai platform includes Koa AI for campaign optimization. The measurement is strong for open-internet programmes; it is less specialized for CTV-specific outcome attribution than Viant's closed-loop approach.
• Yahoo DSP (first-party login graph): Method-level: Yahoo's logged-in user base across Mail, Finance, and Sports properties provides deterministic attribution for users who encounter Yahoo-property ads. Primary-source documentation for specific measurement products not retrieved at build time. Verify current measurement capabilities directly with Yahoo Advertising.

The practical implication: if your primary objective is CTV-driven brand and performance measurement, Viant or Amazon DSP (depending on whether commerce or household reach is the signal) are the most direct paths. If your measurement objective spans the full open internet and requires first-party data portability, UID2 via The Trade Desk is the more flexible architecture. None of these is wrong; they answer different questions.

For brands running across Singapore, Australia, Canada, the US, and Malaysia, the programmatic CTV partner guide covers the operational model decision alongside measurement architecture. For brand safety considerations in programmatic environments, see the MFA and brand safety in programmatic guide.

DSP evaluation checklist: nine questions before you commit

Before shortlisting a people-based DSP for a programme, work through these questions. The checklist is interactive: check each item as you confirm it with your platform contacts.