← Marketing Insights
Marketing Insights · Talk · NABE TEC Amsterdam 2026

The Economics of Smart Bidding

Untargeted tROAS is not a coherent optimization standard — it’s a bid. Here’s how ad platforms use it to capture advertiser surplus, and how to take control back.

“Smart Bidding” is how platforms describe their automated strategies: maximize conversion value at a target return on ad spend. For 99.9% of advertisers it isn’t optional — Performance Max, App campaigns, and Demand Gen require it, and paid social runs on it entirely. The question is not whether to use it, but how to avoid handing the platform your surplus while you do.

Mechanism design has been extracting advertiser surplus for years

This isn’t new. The last major paper to put these “tricks” in the public domain — Lahaie & Pennock, 2008 — showed how compressing relevance gaps in the auction increases platform revenue and decreases advertiser surplus, with relatively small efficiency loss. If value and relevance are positively correlated (they are), the platform captures more. The modern Smart Bidding stack is a more sophisticated version of the same idea.

The correct optimality condition

An advertiser’s problem is well-defined: maximize return subject to a budget. The right optimality condition is the equimarginal principle — the marginal return per dollar of spend must be equal across all campaigns and countries.

max Σ Ric(bic)  s.t.  Σ Cic(bic) ≤ B   ⟹   dRic/dCic = λ  ∀ i, c

What a single tROAS target actually imposes is different: it forces the average return ratio to a constant across every unit — similar to cost-plus pricing, where margin is equal across products despite very different elasticities. That is not the same as equalizing marginal return, and the gap between them is exactly where money leaks.

The key result: a single tROAS target overspends low-elasticity campaigns (whose true ROAS should be higher) and underspends high-elasticity ones (whose true ROAS should be lower). Efficient spend is left on the table, and the differences across countries, categories, and ad formats can be large.

The constraint-slack mechanism

Surplus ROAS — the gap between your target and the true marginal economics — lets the platform operate in negative-marginal-ROI territory on your behalf. Bid higher and you win impressions in more auctions, at progressively lower average relevance and conversion rate. The platform is happy to sell you that marginal, low-value volume. This holds even without an explicit price-support incentive.

Measuring elasticity dynamically

The best response is to embrace optimization within the Smart Bidding world — and the foundation is measuring spend-return elasticity as granularly as possible.

  • Stage 1 — a noisy learner. Bid push/pull experiments and bid-change causal inference, clustering experimental units by the query × campaign/ad-group graph to limit spillover. Instrument with competitor auction shocks where you can.
  • Stage 2 — smooth into targets. Set each bidding unit’s tROAS equal to the global target scaled by the ratio of global elasticity to the unit’s elasticity. Higher elasticity → lower target, because return on the margin is higher.
  • Output. Differentiated tROAS bids by campaign × country (or portfolio, or ad group).
Respect concavity. Realistic budget shifts are ±20–40%. Bid up 20% and average ROI falls; your bids stay anchored to average ROI but differentiated by elasticity. Shifts outside your measured window require fresh elasticity estimates.

Limiting constraint slack through structure

Differentiated bids reduce value-based slack directly. Campaign structure handles the rest:

  • Quality fencing. Extract high-performing search terms from broad match and route them to exact-match campaigns with their own higher targets. Let broad catch new terms; don’t give it slack on known-good ones.
  • Conservative bids on broad. Broad campaigns semantically overlap stronger, targeted ones. Higher targets limit cannibalization.
  • Negative keywords at scale. The operational implementation of a participation threshold — actively block low-marginal-ROI term categories.
  • Tiered shopping & value segmentation. Group by expected value and tier bids so the best inventory triggers on mutually eligible queries, creating a natural participation hierarchy.

In conclusion

  • tROAS / tCPA is not a coherent optimization standard — it’s a bid.
  • Untargeted, it lets the platform capture your surplus via the constraint-slack mechanism.
  • Differentiating targets on measured marginal ROI directly reduces that slack.
  • The right campaign structure produces the right bidding units and handles semantic overlap.
This is the kind of system we build with clients — measured on your data, structured for your catalog, and owned by your team. Book an assessment →

Need Help Choosing Plan

Describe your company's business activities and needs

Your information has been forwarded to our solutions team, a team member will contact you within 24 working hours.

Core

€500

per month

Features

Performance and Scale

Custom ML Modeling

Integrations

Support

Pro

€2500

per month

Everything in Core, plus:

Performance and Scale

Custom ML Modeling

Integrations

Support

Enterprise

Negotiated

Everything in Pro, plus:

Performance and Scale

Custom ML Modeling

Integrations

Support

Core

€500

per month

Features

Performance and Scale

Custom ML Modeling

Integrations

Support

Pro

€2500

per month

Everything in Core, plus:

Performance and Scale

Custom ML Modeling

Integrations

Support

Enterprise

Negotiated

Everything in Pro, plus:

Performance and Scale

Custom ML Modeling

Integrations

Support