Measure-improve swarm · Offline-attribution-intelligence agent · Marketing-mix-modeling skill · Build pillar · Published July 12, 2026

How to build per-location MMM for multi-store marketing budget allocation

Multi-location franchise and multi-store operators work above a strong MMM + Bayesian-inference + foot-traffic + ad-platform + econometric-controls primitives layer (Robyn from Meta + LightweightMMM from Google + PyMC-Marketing + Recast + Stan + PyMC + NumPyro + TensorFlow Probability for open-source MMM; Marketing Evolution + Analytic Edge + Nielsen MTA + Neustar MarketShare + IRI MMM + Fospha + Ascential Edge for commercial MMM; Placer.ai + SafeGraph + Foursquare + Veraset + Cuebiq + Near + Unacast for foot-traffic; Google Ads + Meta + TikTok + Amazon Ads + Microsoft Ads + Pinterest + Snap + Reddit + LinkedIn + Local Service Ads for ad platforms — each vendor ships sophisticated primitives that the orchestration sits above). The orchestration that sits above those primitives — per-location spend-sales joining across 25+ channels, per-location Bayesian MMM with hierarchical pooling at DMA / region / banner level, per-location saturation-adstock with operator-counsel-reviewed parameters, per-location halo- cannibalization decomposition paired with incrementality tests, per-location counterfactual optimization with budget constraints encoded as policy-as-code, per-location out-of- sample validation, multi-LLM pre-publish substantiation check, feedback loop, per-MMM-run compliance gate — is operator-side architecture. Account-rolled-up MMM treats the operator as a single buyer; per-location MMM treats every store as a modeling unit that borrows strength from peers without losing per-location identification. This guide explains how to architect the marketing-mix-modeling skill on the offline- attribution-intelligence agent end-to-end.

Start with the AI Readiness Assessment Explore Fractional CMO with AI Swarm Take the 3-question fit quiz

What you will build

A per-location spend-sales joining layerthat ingests 25+ channels (Google Ads + Meta + TikTok + Amazon Ads + Microsoft Ads + Pinterest + Snap + Reddit + LinkedIn + Local Service Ads + OOH + radio + TV + CTV + streaming audio + direct mail + print + influencer + affiliate + PR + organic social + email + SMS + referral + trade promotion) and joins to per-location store-level signal (sales by receipt, foot-traffic from Placer / SafeGraph / Foursquare / Veraset / Cuebiq, loyalty redemptions, call tracking, unified-attribution-stream handoff, pricing-promotional calendar, weather-seasonal controls, competitive-density-mapping handoff, DMA trading- area overlay, and econometric controls — CPI, unemployment, disposable income, housing starts, fuel price).
A per-location Bayesian MMM layer with hierarchical pooling at DMA / region / banner level on operator-selected engine (Robyn, LightweightMMM, PyMC- Marketing, Recast, Stan, PyMC, NumPyro, or TensorFlow Probability), MCMC sampling, prior elicitation from domain experts, posterior uncertainty, credible intervals, MAP estimation, and WAIC / LOO cross-validation.
A per-location saturation-adstock layer(saturation: Hill, Michaelis-Menten, S-curve; adstock: geometric, Weibull, double-exponential) with operator- counsel-reviewed half-life, peak-effect delay, and decay rate per channel.
A per-location halo-cannibalization layerthat decomposes cross-channel halo (TV halo on search, OOH halo on local search, CTV halo on foot-traffic), cross- channel cannibalization (paid search cannibalizing organic, paid social cannibalizing organic social), cross-location DMA spillover, trading-area overlap, and multi-banner internal cannibalization, anchored on incrementality tests (geo experiments, conversion-lift tests, Ghost Ads, Meta PSI Conversion Lift, Google Conversion Lift).
A per-location counterfactual optimization layer that generates counterfactual spend scenarios and solves budget allocation via Lagrangian methods against per-channel marginal ROAS and decreasing-marginal-returns curves, respecting budget constraints (overall, channel floor / ceiling, DMA floor / ceiling) via policy-as-code (OPA Rego, AWS Cedar, Casbin, Cerbos, Oso).
A per-location out-of-sample validation layer (rolling-window out-of-sample test, holdout period, walk- forward validation, MAPE, WAPE, R-squared, AIC, BIC, posterior predictive check) with domain-expert review handoff.
A multi-LLM pre-publish substantiation checkthat cross-checks MMM output claim substantiation, causal- vs-correlation disambiguation, confidence-interval presentation, and uncertainty quantification across a multi- vendor LLM ensemble against operator-labeled holdouts.
A feedback loop comparing realized vs predicted ROAS, realized vs predicted incrementality, and realized vs recommended spend, with saturation-curve, adstock, halo-cannibalization, and prior recalibration plus MMM model drift detection and retraining-trigger logic.
A per-MMM-run compliance gate anchored on FTC substantiation doctrine for ROAS/incrementality claims, FTC MARS multi-location claim consistency, FDD Item 19 Financial Performance Representations when MMM outputs reach franchisees or prospective franchisees, SEC Reg G + Item 7 MD&A + Item 303 of Regulation S-K when MMM outputs surface in public filings, and causal-vs-correlation disclosure, extended to FINRA Rule 2210 + CCPA/CPRA + GDPR + EU AI Act + NIST AI RMF + ISO 42001/27001 + SOC 2 Type II + Sarbanes- Oxley + state UDAP via policy-as-code (OPA Rego, AWS Cedar, Casbin, Cerbos, Oso) that operator counsel reviews.
Cross-skill handoffs and an audit trail to siblings on the offline-attribution-intelligence agent and broader swarm, with audit trail to operator-controlled WORM storage at per-statute retention windows operator counsel sets.

Where the orchestration above MMM + Bayesian + foot-traffic + ad-platform primitives compounds at multi-location scale

The vendor primitives are strong. Open-source MMM libraries ship Bayesian-inference engines. Commercial MMM vendors ship account-rolled-up MMM workflows. Foot-traffic vendors ship per-location visit signal. Ad platforms ship per-campaign telemetry. The orchestration above those primitives is what compounds at multi-location franchise and multi-store scale.

The first operationally distinctive constraint is FTC substantiation doctrine applied to ROAS and incrementality claims. MMM-derived ROAS and incrementality figures surfaced as advertising, sales-collateral, or franchisee-recruitment claims must be substantiable. The per-MMM-run gate ties every published number to the underlying MMM run, prior set, posterior distribution, and validation evidence.

The second distinctive constraint is FTC MARS multi-location claim consistency. When MMM outputs differ across locations for the same operator brand, FTC scrutiny of claim consistency applies. The gate routes per-location MMM output divergence to operator-counsel-reviewed reconciliation.

The third distinctive constraint is FDD Item 19 Financial Performance Representations. When MMM-derived ROAS, payback, or budget recommendations are shared with the franchisee council or prospective franchisees, Item 19 governs how those representations can be made (basis, time period, geographic scope, substantiation), and operator counsel maintains the Item 19 disclosure register. The gate routes franchisee-visible MMM outputs to the FDD Item 19 workflow before release.

The fourth distinctive constraint is SEC Reg G + Item 7 MD&A + Item 303 of Regulation S-K when MMM-derived metrics surface in public filings, earnings releases, or investor decks. Reg G governs non-GAAP financial measures and requires reconciliation to the most directly comparable GAAP measure when applicable. The gate routes publicly- disclosed MMM outputs to securities-counsel review.

The fifth distinctive constraint is causal-vs-correlation disambiguation. MMM produces correlational estimates whose causal interpretation depends on identification assumptions (no-confounding, stable-unit-treatment-value, conditional ignorability). FTC Reasonable-Basis Doctrine and standard scientific practice both require that claims framed as causal be backed by causal-identification evidence. The gate enforces a causal-vs-correlation disclosure on every published MMM output and pairs MMM-derived estimates with incrementality-test results when available.

Beyond the five anchors, the gate also covers FINRA Rule 2210 communications with the public when investment-grade operators publish MMM-derived performance; CCPA/CPRA + GDPR + PIPEDA + CASL + LGPD + DPDP when MMM joins customer identity for personalization; EU AI Act Articles 13/14 when MMM-driven budget allocation drives automated decisioning; NIST AI RMF + ISO 42001 + ISO 27001 + SOC 2 Type II; Sarbanes-Oxley Section 302/404 for marketing-spend accounting controls; state UDAP for state-AG scrutiny of cross-location representations. The gate is policy-as-code (OPA Rego, AWS Cedar, Casbin, Cerbos, Oso); operator counsel reviews rule updates.

The real ecosystem the orchestration sits above

MMM and Bayesian-inference primitives

Robyn (Meta), LightweightMMM (Google), PyMC-Marketing, Recast, Stan, PyMC, NumPyro, TensorFlow Probability for open-source MMM and Bayesian inference; Marketing Evolution, Analytic Edge, Nielsen MTA, Neustar MarketShare, IRI MMM, Fospha, Ascential Edge for commercial MMM. Strong primitives for Bayesian MMM and account-rolled-up MMM workflows. The per-location hierarchical pooling layer sits above this layer.

Foot-traffic and ad-platform primitives

Placer.ai, SafeGraph, Foursquare, Veraset, Cuebiq, Near, Unacast for foot-traffic; Google Ads, Meta, TikTok, Amazon Ads, Microsoft Ads, Pinterest, Snap, Reddit, LinkedIn, Local Service Ads for ad platforms. Strong primitives for per-location visit signal and per-campaign telemetry. The per-location spend-sales joining layer sits above this layer.

Incrementality-test and experimental-design primitives

Geo experiments (Meta Lift, Google Geo Experiments, geo-uplift methods), conversion-lift tests, Ghost Ads, Meta PSI Conversion Lift, Google Conversion Lift. Strong primitives for causal identification. The MMM halo- cannibalization layer pairs MMM-derived estimates with these tests for substantiation.

Compliance-tooling primitives

Hyperproof + Drata + Vanta + Thoropass for SOC 2 / ISO control evidence; OneTrust + TrustArc + Ketch + Securiti + BigID for privacy program tooling; FDD-disclosure management tooling that operator counsel maintains. Strong primitives. The per-MMM-run compliance overlay coordinates them via a policy-as-code gate (OPA Rego, AWS Cedar, Casbin, Cerbos, Oso) that operator counsel reviews.

How the architecture is built

Spend-sales joining substrate. Pull spend telemetry from ad platforms via APIs and CDC. Join to per- location store-level signal (sales by receipt, foot-traffic, loyalty, call tracking, unified-attribution-stream, pricing- promotional calendar, weather-seasonal controls, competitive- density mapping, DMA overlay, econometric controls). Land the joined data in the operator data warehouse (Snowflake, Databricks, BigQuery, Redshift, Postgres).
Per-location Bayesian MMM. Choose an engine (Robyn, LightweightMMM, PyMC-Marketing, Recast, Stan, PyMC, NumPyro, TensorFlow Probability). Specify hierarchical pooling at DMA / region / banner level. Elicit priors from domain experts. Run MCMC sampling. Diagnose convergence (R-hat, effective sample size, divergent transitions). Surface posterior credible intervals and MAP estimates. Compare candidate specifications with WAIC and LOO.
Saturation-adstock decomposition. Fit saturation curves (Hill, Michaelis-Menten, S-curve) and adstock decay (geometric, Weibull, double-exponential) per channel. Anchor half-life, peak-effect delay, and decay rates against operator-counsel-reviewed domain expectations.
Halo-cannibalization decomposition.Decompose cross-channel halo, cross-channel cannibalization, cross-location DMA spillover, trading-area overlap, and multi-banner internal cannibalization. Pair with incrementality-test results (geo experiments, conversion- lift tests, Ghost Ads, Meta PSI Conversion Lift, Google Conversion Lift) for causal-identification evidence.
Counterfactual optimization. Generate counterfactual spend scenarios. Solve the budget allocation via Lagrangian methods against per-channel marginal ROAS and decreasing-marginal-returns curves. Respect budget constraints (overall, channel floor and ceiling, DMA floor and ceiling) via policy-as-code (OPA Rego, AWS Cedar, Casbin, Cerbos, Oso). Surface to forward-looking- recommendations handoff.
Out-of-sample validation. Run rolling- window out-of-sample tests, holdout periods, and walk- forward validation. Report MAPE, WAPE, R-squared, AIC, BIC, and posterior predictive checks. Hand off to domain-expert review.
Multi-LLM pre-publish substantiation check.Ensemble multiple vendor LLM APIs for MMM output claim substantiation, causal-vs-correlation disambiguation, confidence-interval presentation, and uncertainty quantification. Run self-consistency checks. Extract chain- of-thought to the audit trail.
Per-MMM-run compliance gate. Express the gate as policy-as-code on OPA Rego, AWS Cedar, Casbin, Cerbos, or Oso. Encode the five distinctive anchors (FTC substantiation, FTC MARS, FDD Item 19, SEC Reg G, causal- vs-correlation disambiguation) plus the broader compliance surface. Operator counsel reviews every rule update.
Feedback loop. Compare realized vs predicted ROAS, realized vs predicted incrementality, and realized vs recommended spend. Recalibrate saturation curves, adstock parameters, halo-cannibalization decomposition, and priors. Detect MMM model drift and trigger retraining.
Cross-skill handoffs. Hand off to siblings on the offline-attribution-intelligence agent and broader swarm.
Audit trail. Emit a per-MMM-run canonical audit record to operator-controlled WORM storage with per-statute retention windows operator counsel sets (IRS 7yr, FTC 7yr, SEC 6yr, SOX 7yr).

Frequently asked

What does per-location MMM do that an account-rolled-up MMM run does not?

Open-source MMM libraries (Robyn from Meta, LightweightMMM from Google, PyMC-Marketing, Recast, Stan, PyMC, NumPyro, TensorFlow Probability) ship strong primitives for Bayesian MMM. Commercial MMM vendors (Marketing Evolution, Analytic Edge, Nielsen MTA, Neustar MarketShare, IRI MMM, Fospha, Ascential Edge) ship strong primitives for account-rolled-up MMM analysis. Foot-traffic vendors (Placer.ai, SafeGraph, Foursquare, Veraset, Cuebiq, Near, Unacast) ship strong primitives for per-location visit signal. Ad platforms (Google Ads, Meta, TikTok, Amazon Ads, Microsoft Ads, Pinterest, Snap, Reddit, LinkedIn, Local Service Ads) ship strong primitives for per-campaign spend and impression telemetry. Per-location MMM sits above this layer for multi-location franchise + multi-store operators, and adds: a per-location spend-sales joining layer that pulls 25+ channels (Google Ads, Meta, TikTok, Amazon Ads, Microsoft Ads, Pinterest, Snap, Reddit, LinkedIn, Local Service Ads, out-of-home, radio, TV, CTV, streaming audio, direct mail, print, influencer, affiliate, PR, organic social, email, SMS, referral, trade promotion) into a per-location store-level signal — sales by receipt, foot-traffic from Placer/SafeGraph/Foursquare, loyalty redemptions, call tracking, unified-attribution-stream handoff, pricing-promotional calendar, weather-seasonal controls, competitive-density-mapping handoff, DMA trading-area overlay, and econometric controls (CPI, unemployment, disposable income, housing starts, fuel price); a per-location Bayesian MMM layer that runs hierarchical Bayesian pooling at DMA / region / banner level so that location-specific posteriors borrow strength from peers without losing per-location identification, with operator-selected Bayesian engine (Robyn, LightweightMMM, PyMC-Marketing, Recast, Stan, PyMC, NumPyro, or TensorFlow Probability), MCMC sampling, prior elicitation from domain experts, posterior uncertainty, credible intervals, MAP estimation, and WAIC / LOO cross-validation; a per-location saturation-adstock layer that fits the saturation curve (Hill, Michaelis-Menten, S-curve) and the adstock decay (geometric, Weibull, double-exponential) per channel with operator-counsel-reviewed half-life, peak-effect delay, and decay rate; a per-location halo-cannibalization layer that decomposes cross-channel halo (TV halo on search, OOH halo on local search, CTV halo on foot-traffic), cross-channel cannibalization (paid search cannibalizing organic, paid social cannibalizing organic social), cross-location DMA spillover, trading-area overlap, multi-banner internal cannibalization, and pairs the decomposition with incrementality tests (geo experiments, conversion-lift tests, Ghost Ads, Meta PSI Conversion Lift, Google Conversion Lift); a per-location counterfactual optimization layer that generates counterfactual spend scenarios, solves the budget allocation via Lagrangian methods against per-channel marginal ROAS and decreasing-marginal-returns curves, respects budget constraints (overall budget, channel floor / ceiling, DMA floor / ceiling) encoded as policy-as-code (OPA Rego, AWS Cedar, Casbin, Cerbos, Oso) that operator counsel reviews; a per-location out-of-sample validation layer (rolling-window out-of-sample test, holdout period, walk-forward validation, MAPE, WAPE, R-squared, AIC, BIC, posterior predictive check) with domain-expert review handoff; a multi-LLM pre-publish check that cross-checks MMM output claim substantiation, causal-vs-correlation disambiguation, confidence-interval presentation, and uncertainty quantification; a per-MMM-run compliance gate (covered in the next answer); a feedback loop comparing realized vs predicted ROAS, realized vs predicted incrementality, and realized vs recommended spend, recalibrating saturation curves, adstock parameters, halo-cannibalization decomposition, and priors; and an audit trail to operator-controlled WORM storage at per-statute retention windows.

What are the operationally distinctive compliance anchors for per-location MMM, and how does the per-MMM-run compliance gate cover them?

Five anchors sit at the operational center of multi-location MMM that off-the-shelf MMM compliance overlays often miss. Anchor 1 — FTC substantiation doctrine (Pfizer 1972 plus the broader Reasonable-Basis Doctrine) for ROAS and incrementality claims. MMM-derived ROAS figures and incrementality estimates surfaced as advertising claims, sales-collateral claims, or franchisee-recruitment claims must be substantiable. The per-MMM-run gate ties every published ROAS or incrementality number to its underlying MMM run, prior set, posterior distribution, and validation evidence so counsel has the substantiation record. Anchor 2 — FTC MARS multi-location claim consistency. When MMM outputs differ across locations for the same operator brand, FTC scrutiny of claim consistency applies; the gate routes per-location MMM output divergence to operator-counsel-reviewed reconciliation. Anchor 3 — FDD Item 19 Financial Performance Representations. When MMM-derived ROAS, payback, or budget recommendations are shared with the franchisee council or prospective franchisees, Item 19 governs how those representations can be made (basis, time period, geographic scope, substantiation), and the operator counsel-approved Item 19 disclosure register applies. The gate routes franchisee-visible MMM outputs to the FDD Item 19 workflow before release. Anchor 4 — SEC Reg G + Item 7 MD&A for public-filing surfacing. When MMM-derived metrics surface in public filings, earnings releases, or investor decks, SEC Regulation G (non-GAAP financial measures), Item 7 MD&A, and Item 303 of Regulation S-K can apply; the gate routes publicly-disclosed MMM outputs to securities-counsel review. Anchor 5 — causal-vs-correlation disambiguation. MMM produces correlational estimates whose causal interpretation depends on identification assumptions (no-confounding, stable-unit-treatment-value, conditional ignorability). FTC Reasonable-Basis Doctrine and standard scientific practice both require that claims framed as causal be backed by causal-identification evidence; the gate enforces a causal-vs-correlation disclosure on every published MMM output. Beyond the five anchors, the per-MMM-run gate also covers FINRA Rule 2210 communications with the public when investment-grade operators publish MMM-derived performance; CCPA/CPRA + GDPR + PIPEDA + CASL + LGPD + DPDP when MMM joins customer identity for personalization; EU AI Act Articles 13/14 when MMM-driven budget allocation drives automated decisioning; NIST AI RMF + ISO 42001 + ISO 27001 + SOC 2 Type II; Sarbanes-Oxley Section 302/404 for marketing-spend accounting controls; state UDAP for state-AG scrutiny of cross-location representations. The gate is policy-as-code on OPA Rego, AWS Cedar, Casbin, Cerbos, or Oso, with operator counsel reviewing rule updates.

How does the per-location spend-sales joining, Bayesian MMM, and saturation-adstock decomposition actually work?

The per-location spend-sales joining layer ingests per-channel spend (impressions, reach, CPM, CPC, cost, spend) across the 25+ channel surface and joins it to per-location store-level signal. Sales by receipt come from the POS receipt-joining handoff. Foot-traffic comes from Placer.ai, SafeGraph, Foursquare, Veraset, Cuebiq, Near, or Unacast. Loyalty redemptions come from the loyalty-management agent. Call tracking comes from the call-tracking integration sibling. The unified-attribution-stream handoff carries online touchpoint context. The pricing-promotional calendar carries discount events. Weather-seasonal controls and econometric controls (CPI, unemployment, disposable income, housing starts, fuel price) reduce confounding. DMA trading-area overlay and competitive-density-mapping handoff capture local-market structure. The per-location Bayesian MMM layer fits a hierarchical Bayesian model with pooling at DMA, region, and banner levels — so per-location posteriors borrow strength from peer locations without losing per-location identification — on an operator-selected engine (Robyn, LightweightMMM, PyMC-Marketing, Recast, Stan, PyMC, NumPyro, or TensorFlow Probability). MCMC sampling generates the posterior. Prior elicitation captures domain knowledge (industry benchmarks, prior MMM runs, qualitative expectations). Posterior uncertainty and credible intervals carry through every downstream claim. MAP estimation surfaces point estimates for tactical use. WAIC and LOO cross-validation compare candidate model specifications. The per-location saturation-adstock layer fits the saturation function per channel (Hill, Michaelis-Menten, or S-curve) and the adstock decay per channel (geometric, Weibull, or double-exponential), with operator-counsel-reviewed half-life, peak-effect delay, and decay rate parameters that anchor the curves against domain expectations.

How do the halo-cannibalization, counterfactual optimization, and feedback loop coordinate with the rest of the swarm?

The halo-cannibalization layer decomposes cross-channel halo effects (TV halo on search, OOH halo on local search, CTV halo on foot-traffic) and cross-channel cannibalization (paid search cannibalizing organic, paid social cannibalizing organic social) so that channel-level ROAS estimates do not double-count or omit cross-channel interactions. Cross-location DMA spillover and trading-area overlap capture between-location interactions. Multi-banner internal cannibalization captures within-portfolio interactions for operators running multiple banners. Incrementality tests (geo experiments, conversion-lift tests, Ghost Ads, Meta PSI Conversion Lift, Google Conversion Lift) provide causal-identification evidence that anchors the MMM-derived estimates against direct experimental measurement. The counterfactual optimization layer generates counterfactual spend scenarios and solves the budget allocation via Lagrangian methods against per-channel marginal ROAS and decreasing-marginal-returns curves, respecting budget constraints (overall budget, channel floor and ceiling, DMA floor and ceiling) encoded as policy-as-code (OPA Rego, AWS Cedar, Casbin, Cerbos, Oso) that operator counsel reviews. The out-of-sample validation layer runs rolling-window out-of-sample tests, holdout periods, walk-forward validation, and reports MAPE, WAPE, R-squared, AIC, BIC, and posterior predictive checks with domain-expert review handoff. The feedback loop compares realized vs predicted ROAS, realized vs predicted incrementality (against subsequent incrementality test results), and realized vs recommended spend, and recalibrates saturation curves, adstock parameters, halo-cannibalization decomposition, and priors. The skill hands off to siblings on the offline-attribution-intelligence agent (attribution rollup, forward-looking recommendations, lost-call recovery, foot-traffic attribution, receipt joining, identity resolution, per-location ROAS significance gating, per-location MMM, per-location attribution rollup, per-location budget recommendation) and across the broader swarm (per-location cross-channel attribution rollup for franchise reporting, budget-creative pairing, per-location bid-budget management, competitive-density mapping, unified-attribution-stream, per-location metric ingestion, real-time multi-location inventory state monitoring, cross-channel action coordination, brand-voice management, forbidden-phrase library, claims-allowlist substantiation).

What does Completions report on a Tier 3 engagement that covers per-location MMM?

Tier 3 engagements report against a pre-engagement baseline that the Tier 1 assessment establishes for the operator stack. The reporting cycle covers six workstreams: (1) per-location spend-sales joining coverage observed across the 25+ channel surface and per-location store-level signal, with per-source ingestion completeness and CDC lag reported; (2) per-location Bayesian MMM surface observed across hierarchical pooling structure (DMA / region / banner), MCMC diagnostics (R-hat, effective sample size, divergent transitions), posterior credible interval widths, and WAIC / LOO cross-validation diagnostics; (3) per-location saturation-adstock surface observed across operator-counsel-reviewed half-life, peak-effect delay, and decay rate parameters per channel, with per-channel goodness-of-fit diagnostics reported; (4) per-location halo-cannibalization surface observed across cross-channel, cross-location, and cross-banner decompositions paired with incrementality-test diagnostics; (5) per-location counterfactual optimization surface observed against operator-counsel-reviewed budget constraints, with per-channel marginal ROAS and budget-recommendation diagnostics reported; (6) per-MMM-run compliance gate pass rate observed across FTC substantiation + FTC MARS + FDD Item 19 + SEC Reg G + causal-vs-correlation disclosure + FINRA Rule 2210 + CCPA/CPRA + GDPR + EU AI Act + NIST AI RMF + ISO 42001/27001 + SOC 2 Type II + Sarbanes-Oxley + state UDAP scope. Caveats: per-source vendor API rate limits + per-source ingestion completeness + foot-traffic-vendor sample quality + incrementality-test execution timing + LLM-vendor availability + per-statute retention windows shifting with operator counsel policy sit outside Completions control and are reported alongside observed performance; attorney-client privilege on counsel-reviewed FDD Item 19 disclosure rules, SEC Reg G non-GAAP-measure disclosures, and substantiation records is preserved through every layer. Completions does not commit to fixed numeric SLAs on coverage, MAPE/WAPE, R-squared, credible-interval widths, or compliance pass rate when those KPIs depend on vendor performance, sample quality, or counsel policy decisions.

Engage Completions

Start with the AI Readiness Assessment (Tier 1, 2-3 weeks). If the operation is ready to absorb the marketing-mix- modeling skill on the offline-attribution-intelligence agent, the assessment hands off to the AI Swarm Setup Sprint (Tier 2, 4-8 weeks). If the operation needs ongoing orchestration after Tier 2 hand-off, the skill continues under Fractional CMO with AI Swarm (Tier 3, 6-month minimum, 1-2 days/wk embedded). Operator owns every artifact at every tier. Operator can in-house at any time.