Sources and provenance

#How to read this lesson

Every material claim across this course already carries an inline [Established], [Directional], or [Speculative] tag where it's made — see the confidence-tag system in How to use this course for what each tag means. This lesson doesn't repeat every individual claim; it collects the underlying sources behind those tags in one place, organised by tier, so you can see the shape of the evidence base at a glance — how much of this course rests on primary data versus a consistent pattern versus a single unverifiable blog post — rather than encountering that only one claim at a time.

Two source-quality notes that recur across this course and are worth stating once, here, rather than repeating in every lesson:

• A source having a name and a URL is not the same as a source being primary. Several of the sources below are agency-marketing blogs, tool-vendor content, or SMMA-course-adjacent material — real, checkable pages, but not disclosed data or a named study. They're listed as [Speculative] for that reason, not omitted, so you can see exactly what this course chose not to build a claim on.
• Where sources disagree with each other in a way that suggests they're repeating one unverified number rather than measuring independently — the "good churn" range in Unit economics and survivors, the personal-injury-law retainer figures in Choosing a niche — that disagreement is named explicitly in the lesson itself rather than smoothed into a single confident number here.

#Established — primary sources, directly checkable

These are named government data, a platform's own published pricing or product documentation, or a named court/regulatory docket — the tier this course treats as fact, with the standing caveat that anything time-sensitive (a price, a CPM, a rule's current status) should be re-verified before you rely on it, since primary sources themselves change.

• US Bureau of Labor Statistics, Business Employment Dynamics programme — the general new-business survival data (≈76.8% one-year survival, ≈51.2% five-year survival) used in Unit economics and survivors as the grounded comparison point against the untraceable "92% of agencies fail" claim. Covers all new employer establishments, not agencies specifically — a floor for comparison, not a direct agency-failure figure.
• Federal Trade Commission, CAN-SPAM Act requirements and penalty schedule (ftc.gov) — the six-requirement cold-email compliance floor and the per-email penalty structure cited in Compliance and the guru economy. Re-verify the specific inflation-adjusted 2026 penalty figure against the FTC's current published number before relying on it.
• Federal Trade Commission, FTC v. Air AI Technologies case materials — the named, docketed 2025–2026 enforcement action (filed August 2025, settled with a permanent ban announced March 2026) cited in Compliance and the guru economy as the closest verified example of Business Opportunity Rule enforcement against exactly this course's adjacent client base.
• Federal Trade Commission, proposed Earnings Claim Rule (January 2025) — the named regulatory proposal cited in the same lesson.
• Telephone Consumer Protection Act (TCPA), general federal framework — the well-established cold-calling/texting consent regime cited in Compliance and the guru economy.
• Meta's own stated product direction for Advantage+ — the shift toward automated campaign setup, cited in Root mechanism and Platform shifts, 2026 as Meta's disclosed roadmap; the specific self-reported performance figures Meta attaches to it (revenue, ROAS improvement) are listed under Directional below, since they're the platform grading its own product.
• GoHighLevel's own published pricing tiers — cited in Tools, KPIs, and kill switches for the CRM/automation-platform cost table; SaaS pricing changes without much notice, so re-verify against the vendor's current page before budgeting.

#Directional — disclosed-operator data or a consistent independent pattern

Real data with a named methodology that isn't a controlled primary study, or a pattern that holds up across multiple independent sources without a single one confirming the exact figure. Treated as a reasonable planning input, not a guarantee — and specifically not something to quote to a prospect as a precise number.

• Databox, survey of its own agency-customer base — the retainer-pricing-by-client-type figures (SMB $1,500–$5,000/mo; enterprise $8,000–$25,000/mo) in Unit economics and survivors. A real, named, disclosed-operator-adjacent source — Databox surveying its own software's users — but a self-selected sample of agencies that already use one reporting tool, not a random sample of the industry.
• Databox, "To Niche or Not to Niche?" survey (published October 2022; 87 respondents — 51 agencies, 36 client-side companies) — cited in Choosing a niche for the finding that 77.27% of surveyed agencies already self-identify as niche, and the near-even client preference split between service- and industry-specialised agencies. Real named methodology, but older than the rest of this course's 2026 research pass, and Databox's own published findings do not include a pricing-premium or growth-rate multiplier — a distinction this course draws deliberately, since 2026 content elsewhere attaches unsourced multipliers to this same underlying (real) trend.
• Haus, 640-test incrementality study on Meta Advantage+ — cited in Platform shifts, 2026 as independent evidence that platform-reported ROAS overstates real incremental performance. Haus sells incrementality-measurement services, a real commercial interest in this finding, but a 640-test evidence base is materially stronger than a blog assertion or the platform's own self-reported benchmark.
• Cold-email and cold-calling conversion-rate benchmark content (a 2025 email-benchmark report; a cold-calling study spanning 200,000+ calls) — cited in Unit economics and survivors for the funnel arithmetic behind the "weeks to first client" estimate.
• Cold-email-tooling vendor cost content — the ~$5,000/month "done properly at volume" deliverability-infrastructure figure in the same lesson, flagged there for a real conflict of interest (vendors selling an alternative to raw cold email have an incentive to make cold email look expensive).
• Agency-industry churn and retention content (multiple independent blogs) — the 18%-annual-retainer-churn / 56-month-average-lifespan figures, the by-agency-size breakdown, and the 43%-of-churn-in-first-90-days concentration, all in Unit economics and survivors. The specific "good churn" percentage disagrees across sources in a way flagged directly in that lesson as folklore-tier even though the broader retention pattern is consistent.
• Agency-industry cost-comparison content (multiple independent breakdowns) — the in-house-hire-versus-retainer cost comparison in Root mechanism.
• SMMA/agency startup-cost breakdowns (multiple independent sources) — the capital-tier table in Capital and timeline, with one source describing a materially different, more heavily-staffed model flagged explicitly as not comparable to the entry-level tier.
• Ad-tech vendor CPM benchmark tracking (a single vendor, undisclosed methodology) — the $13.48 average Meta CPM and Reels-discount figures in Platform shifts, 2026; direction (CPMs rising) treated as more reliable than the precise number.
• Trade-press coverage of TikTok's 2025–2026 ownership change and algorithm shift toward search and longer watch-time (multiple independent sources) — in the same lesson; the ownership change itself is well-corroborated, the specific "added compliance-review layer" detail rests on a single trade-press source and is flagged as less solid than the general direction.
• agencyanalytics.com, "How To Pick Your Agency Niche" and searchlab.nl, "How to Pick a Niche" (framework portions only) — the six-criterion niche-scoring framework in Choosing a niche. Both are agency-marketing-adjacent sites with no named study behind the framework, but the criteria themselves are a reasoned, checkable structure rather than a number to trust at face value — used here as a way to organise judgment, not as data.
• Niche-specific retainer figures corroborated by two or more independent sources — medical aesthetics/med spa ($2,000–$10,000/mo, clustering $4,000–$8,000), cosmetic dental/dermatology ($2,000–$6,000/mo), and the real-estate retainer-plus-ad-spend structural split, all in Choosing a niche.
• A single aggregator-reported statistic on blended in-house-plus-agency marketing outcomes (≈42% better results) in Root mechanism — no named underlying study, listed here as Directional on the strength of being consistent with the broader mechanism argument rather than on its own precision.
• Secondary compliance-law summaries of a January 2025 TCPA consent-attribution rule change — cited in Compliance and the guru economy; the change itself is corroborated across multiple compliance-focused sources, but this research relied on secondary summaries rather than the FCC's own rule text directly.

#Speculative — single-source, unsourced, or actively debunked

Claims this course encountered, named specifically, and did not build anything on. Listed here so you recognise and discount them if you see them repeated elsewhere — the point of naming a bad source is to inoculate against it, not to pretend this course never encountered it.

• "92% of agencies fail" / "90% of marketing agencies fail." Traces to a widely-shared LinkedIn post that itself concedes the figure is "just opinionated." Other circulating figures in the same space ("75% of small agencies fail within five years," "34% of advertising agencies close within five years") disagree with each other and with the 92% figure, with no shared methodology behind any of them. Discarded in Unit economics and survivors; the BLS data above is the grounded comparison point used instead.
• A "class-action lawsuit against SMMA" course sellers, over unfulfilled income promises. Found only in low-quality aggregator search content with no named plaintiff, filing, or court docket. Named in Unit economics and survivors so it isn't repeated as fact — this course found no evidence it's false, only that it found no evidence it's true, which is a meaningfully different and weaker claim than the confident tone it's usually repeated with.
• "Specialist agencies convert at 67% vs 20% for generalists, charge 2–3x more, grow 3x faster, and 84% of agencies now specialise." Traced in Choosing a niche through its actual citation chain: the number originates, unsourced and undisclaimed, in a blog post by Phantom (phantomleads.ai) — a company selling a lead-generation/prospecting tool, with a direct commercial interest in agencies believing specialisation is dramatically more effective. searchlab.nl and at least one other site repeat the figures citing Phantom's post as their only source. No named study, survey, or primary disclosure sits behind any version of the claim. Discarded as a number; the real, named Databox survey above is used instead for the underlying (and better-supported) direction.
• Niche-specific retainer figures with only single-source backing and no disclosed methodology — the personal-injury-law figures (which additionally disagree with each other by more than 10x across two sources), roofing, restaurants, franchises, financial advisors, insurance agencies, and mortgage brokers, all in Choosing a niche. Reported as rough planning ranges, explicitly not as benchmarks to price against.
• SMB AI-adoption rate (≈22% in 2024 to ≈38% in 2026) and AI-agents-market-doubling figures, in Is SMMA dead?. Neither traces to a primary study; the general direction (rising SMB AI adoption) is independently plausible, the specific numbers are not independently confirmed.
• The "AI agencies run 70–90% margins vs 30–50% for SMMA, with SMMA down to 11–20% net in 2026" comparison, in the same lesson. Traces to a company selling AI-automation-agency tooling and training — a direct financial incentive to make the alternative it sells look structurally superior to the one it's replacing. Named as marketing content, not a neutral benchmark, the same treatment this course's own compliance lesson gives promised-income figures from mentorship programmes generally.
• TikTok's reported "84% of searches happen during the exploration phase" figure, in Platform shifts, 2026. Could not be traced to a named study; the broader direction (TikTok weighting search and discovery more heavily) is corroborated independently and carries a Directional tag on its own.
• SMMA-course-marketing content generally (the "$1,000–$2,000/month sweet spot for a new agency" framing, and any course or programme promising a first client in days rather than weeks) — used throughout this course only to note where such content's numbers cluster lower than general agency-industry figures, never as standalone evidence, on the grounds argued in Unit economics and survivors: course-seller content describes its own under-differentiated, entry-level students, not the achievable ceiling of the model.

#What this list is not

This is not a claim that every source above was checked against its own underlying data — for the Directional and Speculative tiers, that's precisely the point: this course could not verify them further, and says so rather than dressing up a blog post as a study. Where this course's own confidence tag says [Established], that claim was checked against a primary source directly. Everywhere else, treat the number as a planning input to be re-verified against your own real quotes, your own outreach data, and your own market — not a figure to repeat as fact to a client, a lender, or in your own marketing.