The Missing Layer: Why The Trade Desk's Performance Mode Can't Work for B2B
They handed bidding, pacing and inventory selection to an AI co-pilot. For B2B, it exposes a foundational data problem programmatic has never solved.
The biggest DSP on the open internet just handed bidding, pacing and inventory selection to an AI co-pilot. For consumer advertisers, it’s a genuine simplification. For B2B, it exposes a foundational data problem programmatic has never solved.
What you’ll learn in this article
Why The Trade Desk’s new Performance Mode, now in closed beta, is a well-engineered product aimed at the wrong problem for B2B marketers
The three-layer data hierarchy (global account → buying committee → individual stakeholder) that B2B actually runs on, and why every DSP in the market operates on only one layer
How auto-optimisation against 10–18 month B2B sales cycles systematically pushes spend toward the wrong impressions, no matter how strong the underlying AI
What a B2B “adaptation layer” sitting on top of a DSP actually has to do: identity resolution, committee clustering, HVA scoring, curated supply, and pipeline tie-back
Why the Publicis and Omnicom fee audits are the surface debate, and where the deeper architectural argument for B2B actually sits
In early April 2026, Digiday confirmed what four ad executives had already been whispering: The Trade Desk’s new Koa Adaptive Trading Modes had moved into closed beta. The product was announced to investors in September 2025, and its a fascinating aside that an all-in-one priced solution has launched into a headwind around transparency. Anyhow, the outcome (pun intended) is that advertisers can now choose between Performance Mode - where TTD’s Koa AI automatically manages bidding, pacing, inventory prioritisation and data application, with media, data and tech fees rolled into a single CPM - and Control Mode, where traders keep manual control over every lever.
The product was first announced in September 2025 alongside Audience Unlimited, TTD’s third-party data marketplace overhaul. Performance Mode bundles Audience Unlimited at no extra cost; Control Mode charges tiered rates of 3.3% or 4.4% of impression costs. As of February 2026, Control Mode is the platform default but Performance Mode is the one TTD is actively steering advertisers towards.
The timing is awkward. Trading Modes have gone live in beta in the same quarter the industry has spent auditing TTD for fee opacity — Publicis first, Omnicom second, with Publicis publicly advising clients to avoid TTD’s stack while the review ran. That backdrop is what most of the trade press has focused on. It is not the interesting story but frankly, simply reading the paragraph above this one - these complex named products with 3.3% extra fees here and 4.4% there gets to the heart of the debate triggered in these audits - the fee accruals are complex and the traders with their hands on their keyboards getting battered internally to drive better outcomes, can easily end up toggling on far more fee generative products then they maybe realise. Long gone are the DSP days of a bid reduction and some data costs.
The actually interesting story, if you work in B2B, is that Performance Mode will optimise brilliantly against a universe that was never designed to carry the structure B2B actually runs on. The fee debate is a surface argument. The data architecture underneath is the real one.
TradeDesk remains the best DSP for B2B marketers, but as I have constantly argued here and on LinkedIn, there is no genuinely turnkey DSP for B2B - and that boils down to data hierarchy at its core. Before that though, what is Performance Mode?
What Performance Mode actually is
Setting the commercial debate to one side, the product itself is well engineered.
Koa — TTD’s optimisation AI — has been evolving for years. What Performance Mode does is give it full agency (control): dynamically optimising bids and allocation within buyer-set guardrails, pulling in Audience Unlimited, Predictive Clearing, Identity Alliance, Prism and measurement in concert, all continuously optimised toward the outcomes the advertiser has defined. TTD’s own framing is that Performance Mode “activates the best performance features for every impression” while keeping strategy “firmly in human hands.”
For a consumer brand buying open-internet reach against tens of millions of addressable users, this is a genuine step forward. The state of the art in DSP-led optimisation is, on balance, better than what most trading teams can achieve manually across fragmented inventory. Public commentary from TTD’s partners supports that reading. This levels the playing field between what say an in-house brands team can achieve with the TradeDesk and what specialist B2C trading arms like MiQ and Goodway can achieve. Its codified the best traders and weaponising vast machine learning and AI to pour over vastly more data then any human can use. Its genuinely very exciting for what programmatic can achieve in driving performance alongside other consumer channels.
So this is not a piece arguing that Performance Mode is a bad product. It isn’t.
It is a piece arguing that it’s the wrong shape of product for B2B — not because of the optimisation layer, but because of what sits beneath it.
The foundational data model that programmatic has never built
Every programmatic platform, TTD included, is built around a single unit of targeting: the individual ID. This construct was the centre of attention around the cookie-apocolapse, movement towards unique IDs and other identity solutions and the eventual reversal of Chrome’s deprecation. This was existential because programmatic is built around individual a-placement auctions to individual users.
So whether you are identifying that individual user using cookies, mobile device IDs (MAIDS), hashed emails, CTV identifiers, Unified ID 2.0 — the format changes, but the unit doesn’t. The DSP sees an individual, applies audience logic to that individual, bids on that individual’s impression, and measures an outcome attached to that individual.
That model works because consumer advertising is, fundamentally, an individual-level problem. One person researches an air purifier. One person buys a new pair of jeans. One person installs an app.
B2B is not an individual-level problem. B2B is a hierarchical, multi-tier account problem, and the hierarchy is non-trivial. It is in-fact very complex and thre i sno evidence with any changes int eh way of the world and work with the like sof AI, that this is getting any simpler.
A typical complex B2B purchase, by the available evidence, looks like this:
The global account. The company you’re trying to win. Often multi-entity, multi-region, with subsidiary relationships and parent-company rollups that matter for deal attribution.
The buying committees inside that account. Gartner’s widely cited research puts the typical B2B buying group at six to ten decision-makers for complex solutions, and Forrester’s 2024 State of Business Buying puts the average complex B2B purchase at 13 stakeholders, with 89% of buying decisions crossing multiple departments. In enterprise technology, committees regularly exceed 20 people. Crucially, these are not one committee, they are several: procurement, IT, finance, security, the end-user business unit, legal. Each runs its own parallel evaluation. Potentially multiple committees live for multiple procurement exercises all running at once
The individual stakeholders inside each committee. The people the DSP can actually see. Each with their own role, seniority, influence weight, content preferences, and position in the buying cycle.
Three layers minimum. In regulated, multi-entity enterprise deals, sometimes four or five.
Programmatic’s data model has one layer. The individual.
There is no native concept of account in the DSP. No native concept of buying committee. No way to express, in the platform’s optimisation logic, that three engaged stakeholders inside one target account are worth more than thirty random individuals scattered across the open web. No way to tell the bidder that a finance VP engaging after IT and the end-user BU have both engaged is a materially different signal than a finance VP engaging in isolation.
This is not a flaw in The Trade Desk specifically. It’s the shape of the entire programmatic stack. DV360, Xandr, Amazon DSP, Yahoo DSP — same single-layer model. It is the inherited architecture of a consumer-advertising industry that never had to think in terms of accounts because it was never pricing against accounts.
Why Performance Mode magnifies the problem rather than solving it
Here is where the argument sharpens - or to put it better, the problem becomes real if you are trying to use this for B2B
When you hand optimisation to an AI co-pilot, you are making a bet that the data the AI is optimising against is the right data. In consumer, it is. Koa is optimising toward an outcome (a purchase, an app install, a lead form completion) that is genuinely attached to an individual, inside a short feedback loop, with a clear signal. Their tracking captures it.
In B2B, the optimisation target is fundamentally different:
The outcome is account-level, not individual-level. The “conversion” that matters is a deal closing inside the target account — potentially 10 months later on average for B2B, 12+ months for enterprise deals with large committees, and 3–6 months for mid-market software rising to 9–18 months for enterprise. No real-time bidder can feedback-optimise against that timescale.
The short-term proxy signals are misleading. Clicks, lead forms, content downloads, demo requests. Koa will optimise against these because they’re all it has. But in B2B, the individuals clicking are frequently not the individuals in the buying committee — they’re adjacent researchers, competitive analysts, consultants, or junior staff running background work. Optimising toward “more clicks from this kind of person” pulls spend away from the harder-to-reach decision-makers who actually matter. This is before you realise how the DSP uses a load of segments from data vendors like Eyeota - AND before you realise the tooling like Predictive clearing and just how much that pulls against trying to find the right person in the right account at the right moment. Its really very flawed for this use case
The addressable universe is structurally narrow. Not 50 million consumers. Five hundred to five thousand target accounts, containing perhaps 20,000 to 50,000 relevant individuals across committees. Inside that narrow universe, auto-optimisation against cheap CPMs will systematically prefer the impressions that are easiest to win — which are almost always the wrong ones.
The result is predictable: Performance Mode, applied to a B2B brief, will deliver strong-looking dashboards against consumer-grade metrics while the account-level pipeline underneath stays flat. The AI is not failing. It is doing exactly what it was built to do. It just has nothing useful to chase, because the data model beneath it doesn’t hold the structure that B2B outcomes actually live in.
This yet again is why sophisticated teams need to reject vanity metrics in B2B and with this type of auto-optimisation running behind the scenes, you’ll be burning budget and zeroing in on terrible outcomes
The adaptation layer: what has to exist before any optimiser adds value
This is the work.
Before any DSP-side AI can meaningfully optimise for B2B, someone has to build an adaptation layer that re-hierarchises the data. That layer has to do several things the DSP cannot:
Resolve individuals into accounts. Every impression, every engagement signal, every content interaction needs to be tagged not just to a person but to a company, and ideally to a business unit within that company. This is non-trivial identity work — it draws on firmographic resolution, IP-to-company mapping, B2B identity graphs, CRM integration, and increasingly AI-driven entity resolution against unstructured intent data.
Assemble individuals into committees. Once accounts are resolved, individuals have to be clustered into the functional committees doing the buying. IT, finance, procurement, business unit owners, champions. The committee is the unit that predicts pipeline progression — not the individual.
Score engagement at the account level, not the impression level. An engagement from three stakeholders in two committees is a qualitatively different signal from an engagement from three stakeholders in the same committee, which is different again from twenty engagements from one junior researcher. The scoring logic has to reflect committee shape, seniority, and cross-committee coverage.
Tie the whole thing to pipeline. Marketing-influenced pipeline, not clicks. That requires CRM integration deep enough to close the loop on 6–18 month cycles — which means building the kind of longitudinal view no DSP has, because DSPs are built around short-loop optimisation.
Curate the supply side. Not every impression is worth the same bid inside a 2,000-account universe. Curated deal IDs, premium B2B supply, private marketplaces built around professional context — all matter far more in B2B than in consumer, because the cost of wasting budget on irrelevant impressions is an order of magnitude higher.
Once that adaptation layer is in place, and only then, the things B2B marketers actually need become possible:
Account-level progression, not individual clicks. Measuring whether a target account is advancing through stages of awareness, consideration and preference, across the buying committee, not just whether one person filled in a form.
High-value action (HVA) scoring. Identifying which specific engagement patterns correlate with pipeline creation months later, and weighting the bidding and targeting accordingly.
Marketing-influenced pipeline, measured properly. Closing the loop from impression to committee engagement to opportunity to closed-won revenue, at the account level.
Automated bidding at the account level. Not at the impression level. Bidding logic that says “this account has shown early-stage committee engagement but no finance-side touch yet, push spend toward finance-titled inventory this week” — the kind of logic no DSP-native AI can generate because no DSP-native AI knows what a committee is.
None of these capabilities can be built inside the DSP, because the DSP’s data model doesn’t carry the hierarchy. They have to be built on top.
This is not a criticism of The Trade Desk. It’s a product boundary.
It’s worth being precise here. The Trade Desk is the best independent DSP in the market. Koa is a genuinely strong optimiser. Audience Unlimited is a real simplification of a previously painful data-activation workflow. Performance Mode will, by every indication, deliver meaningful efficiency gains for the consumer advertisers it was designed for.
What it cannot do and what no DSP can do, is carry the foundational data structure B2B depends on. That structure has to be built externally and applied to the DSP from outside.
This is the category FunnelFuel has been operating in for the 4 years. A managed-service layer sitting on top of TTD (and other best-of-breed infrastructure: Index Exchange on the supply side, Bombora on the intent data side, CRM systems on the pipeline side), with the account/committee/individual hierarchy modelled explicitly, HVA scoring running against it, account-level reporting sitting on top, and curated B2B supply underneath. This is not building another DSP. It is adapting the worlds best DSP’s to a problem they weren’t designed for.
The reason this matters right now, in the week Performance Mode has entered beta, is that a lot of B2B marketers are about to be sold a story. The story will say: “Turn on Performance Mode, let the AI drive, the results will improve.” For consumer, it will. For B2B, without the adaptation layer underneath, the AI will confidently optimise toward the wrong thing — and the flatness in pipeline will arrive six months later, uncorrelated in the dashboards with the decision that caused it.
Performance Mode is a great product for the problem it was built to solve.
In B2B, the problem is one layer deeper.
That layer has to be built before any optimiser, however good, has anything useful to chase.
Mike Harty is the founder of FunnelFuel and publishes The B2B Stack. FunnelFuel runs managed-service B2B programmatic on best-of-breed adtech infrastructure, including The Trade Desk, with account-level analytics, HVA scoring, curated deal IDs and the adaptation layer described above.
Further reading / sources
The Trade Desk is changing how advertisers buy — and what they can see — Digiday, 6 April 2026
The Trade Desk Announces Major Overhaul of Digital Advertising Data Marketplace — TTD Investor Relations, 29 September 2025
Performance Mode: AI-powered trading — The Trade Desk
The Trade Desk and Publicis split, ad tech community divided — Ad Age, 23 March 2026
Omnicom launches audit of The Trade Desk’s fees — Ad Age, 24 March 2026
Publicis vs. The Trade Desk isn’t really about transparency — it’s about who gets the margin — Digiday
Gartner research on B2B buying committees (6–10 decision-makers, complex solutions)
Forrester, The State of Business Buying, 2024 (13 stakeholders average)
6sense Buyer Experience Report 2025 (10-month average B2B cycle; 12+ months enterprise)
Demandbase 2025 research on buying group complexity