Agency AI spend can accumulate across research, adaptation and review long before a campaign ships. Qua would resolve routine questions from approved knowledge first, reserving premium reasoning for work where its expected value exceeds its incremental cost.
Industry thesis — modelled, not deployed. Figures are potential savings.
$294.6bnUS internet advertising revenue, 2024IAB/PwC, April 2026
13.9%US internet advertising growth, 2025IAB/PwC, April 2026
$37bnUS creator advertising spend, 2025IAB/PwC, April 2026
$4.44mAverage breach cost, all industriesIBM, 2025
Marketing & creative agencies — the cheapest trustworthy source answers first; frontier reasoning is paid for only when it earns its price.
Where the money goes today
Marketing & creative agencies: the cost of asking
IAB and PwC's April 2026 Internet Advertising Revenue Report put US internet advertising revenue at $294.6 billion in 2025, up 13.9%. That provides context for the scale of digital production, not an estimate of agency AI demand.
The same report put creator advertising spend at $37 billion in 2025, now a core channel. Agencies therefore produce and adapt far more variants per brief than a single campaign master.
IBM's 2025 Cost of a Data Breach Report put the global average breach cost at $4.44 million across industries. This is not an agency-specific loss estimate, but it gives context to controls around unreleased campaigns, customer lists and commercial briefs.
Three pressures
What makes this industry different
Unpriced Repetition Across Accounts
Teams repeatedly ask about tone, specifications and approved claims. Sending every request to a premium model would turn reusable knowledge into recurring expense.
Client Boundaries Must Hold
One agency can hold competing clients' plans. Access controls and provider restrictions would need to apply before retrieval or model selection.
Premium Reasoning Needs Justification
Complex strategy can warrant deeper reasoning; formatting usually cannot. A routing decision should reflect the task rather than a default model subscription.
Five applications
How Qua would run inside marketing & creative agencies
Each application is a real workflow, mapped to the tier that could answer it. The waterfall tries the cheapest trustworthy source first and only pays a frontier model when the expected value clears the gate.
Approved brand voice lookup
Resolves at Personal Knowledge~40,000 questions/month
Account teams would reuse approved guidance rather than regenerate it. Each answer would carry a receipt showing exact query cost, the premium baseline cost, savings and what stayed private.
Potential saving — 90-100% of query spend.
Apply client access rules and an Enterprise Search maximum-tier clamp.
Return a current, verified Personal Knowledge answer when available.
On a miss, search approved brand guidance inside the perimeter; otherwise abstain.
Campaign asset rights check
Resolves at Enterprise Search~30,000 questions/month
Producers would find usage territories, expiry dates and source agreements in one grounded response. Ambiguous rights would still require legal or rights-team review.
Potential saving — 85-100% of query spend.
Apply account permissions, licence-document access and provider blocks.
Check verified answers, then search authorised asset and rights records inside the perimeter.
Return supporting passages or flag missing evidence without external escalation.
Channel copy adaptation
Resolves at Fast Models~20,000 questions/month
Teams would generate first drafts against explicit length and tone constraints. Human review would remain responsible for publication and factual accuracy.
Potential saving — 70-90% of query spend.
Mask restricted brief details and apply a Fast Models maximum-tier clamp.
Check approved variants, then retrieve the brief and channel specifications.
Use Fast Models only for the remaining adaptation and return a cost receipt.
Client status report drafting
Resolves at Fast Models~5,000 questions/month
Account managers would start from a grounded draft rather than reconstruct the reporting period. Missing metrics would remain visible instead of being filled with generated estimates.
Potential saving — 70-90% of query spend.
Restrict retrieval to the account and mask personal or commercial identifiers.
Check saved answers, then retrieve approved delivery and performance records.
Use Fast Models to draft the narrative when cheaper tiers cannot answer.
Cross-market campaign claims review
Resolves at Pro Models~5,000 questions/month
Complex claim comparisons would receive premium reasoning only when the expected review benefit justifies the additional cost. The output would support, not replace, legal and advertising-standards review.
Potential saving — 0-30% of query spend.
Apply market-specific data rules, masking and approved-provider restrictions.
Check verified answers, search substantiation records and test Fast Models capability.
Allow Pro Models only when the expected-value gate clears; require specialist sign-off.
The modelled ledger
What the same year of questions could cost
A thesis, not a case history. The assumptions are stated so you can replace them with your own numbers — which is exactly what a pilot does in week one.
Ledger rates are the vendors’ own published list prices: $0.0150 per premium question and $0.00065 per fast question at 1,500 input / 700 output tokens.
Modelled annual comparison
Workload: 100,000 questions/month for 12 months, at 1,500 input and 700 output tokens per question.
Published list prices, not estimates — premium baseline xAI Grok 4 at $3.00/$15.00 per 1M tokens = $0.0150/question.
Cheap metered tier xAI Grok 4 Fast at $0.20/$0.50 per 1M tokens = $0.00065/question.
Terminal-tier mix: 70% verified or in-perimeter, 25% Fast Models, 5% Pro Models. Rows exclude Qua fees, retrieval infrastructure, integration and human review.
WorkloadFrontier-onlyWith Qua
Verified answers and in-perimeter search840,000 annual questions resolved with no external model call.$12,600$0
Fast Models (Grok 4 Fast)300,000 annual questions at xAI's published Grok 4 Fast rate ($0.00065/question).$4,500$195
EV-gated Pro Models (Grok 4)60,000 annual questions keep frontier reasoning at the published Grok 4 rate.$900$900
At published vendor prices this thesis models $16,905 of avoided annual inference spend — $18,000 down to $1,095, a 93.9% reduction — before the excluded costs above.
Controls that matter here
Policy runs before routing, not after
Client Data Minimisation
For GDPR Article 5 data minimisation, Qua would apply masking, account permissions and provider blocks before routing. Qua Cloud or customer VPC deployment would remain subject to the agency's lawful-basis, processor and transfer assessments.
Licensed Material Access Boundaries
For the UK Copyright, Designs and Patents Act 1988, retrieval would be restricted to authorised material and retain source references. Open Models would be user-picked only, and access to an asset would not establish permission to reuse it.
Advertising Claims Review Evidence
For CAP Code Section 3 substantiation requirements, responses would point reviewers to approved evidence and flag unsupported claims. Receipts would document routing and cost, not certify that an advertisement complies.
In week one, measure knowledge and search resolution rates against a client-approved answer set.
In week one, compare metered spend with a matched premium-only baseline, including retries and reviewer acceptance.
In week one, test cross-client retrieval denial, masking and provider blocks, then inspect privacy receipts.
Sources
Every figure on this page, traceable
Market figures come from the publishers below. Qua savings are modelled from the vendors’ published list prices — they are not customer results.
[1]Interactive Advertising Bureau / PwC, Internet Advertising Revenue Report: Full Year 2025 (2026). Source US internet advertising revenue reached $294.6 billion in 2025, up 13.9%; this is advertising-market revenue, not agency AI demand.
[2]Interactive Advertising Bureau, Digital Ad Revenue Climbs to Nearly $300B (2026). Source Creator advertising spend reached $37 billion in 2025 and is now treated as a core media channel; this measures spend, not AI workload.
[3]Interactive Advertising Bureau / PwC, Internet Advertising Revenue Report: Full Year 2025 — channel breakdown (2026). Source Social advertising reached $117.7 billion and search $114.2 billion in 2025, with programmatic buying above $162 billion; these are channel totals, not agency AI demand.
[4]IBM, Cost of a Data Breach Report 2025 (2025). Source IBM's 2025 study reported a global average breach cost of $4.44 million across industries; this is not an agency-specific loss estimate.
[5]European Union, Regulation (EU) 2016/679 — General Data Protection Regulation, Articles 5, 25, 28 and 32 (2016). Source GDPR requires minimization, purpose limitation and processor safeguards for covered personal data, supporting controls on customer lists and personal information in client briefs.
[6]US Copyright Office, Copyright Law of the United States — Title 17, Sections 106 and 107 (n.d.). Source Copyright owners hold specified reproduction, adaptation and distribution rights, subject to exceptions such as fair use; material access and reuse must respect applicable rights and licenses.
[7]Federal Trade Commission, FTC Policy Statement Regarding Advertising Substantiation (1984). Source Advertisers must possess a reasonable basis for objective claims before dissemination, supporting documented claims review and substantiation rather than reliance on generated assertions.
