Sources and provenance

This course was built via web research in August 2026, specifically to cover a distinct angle from AI Agency's own source brief: an operator automating their own business's operations, not a consultant selling AI implementation to clients. None of it shares source material with AI Agency's original research brief.

#Established (disclosed methodology, primary report, or official vendor page)

• Intuit, "2026 AI Impact Report" — quickbooks.intuit.com (survey of 34,000+ business owners plus anonymized data from 5.3 million+ QuickBooks businesses across the US, Canada, UK, and Australia, developed with University of Chicago economists)
• Gartner, "Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027" (press release, June 2025, poll of 3,400+ organizations actively investing in the technology) — gartner.com, quoted with full methodology via martech.org and searchengineland.com after this research's own direct fetch of gartner.com was blocked
• IBM Newsroom, "IBM report: 13% of organizations reported breaches of AI models or applications, 97% of which reported lacking proper AI access controls" (press release, July 2025, Cost of a Data Breach Report with the Ponemon Institute) — newsroom.ibm.com
• Sogolytics, 2026 consumer-trust survey (n=1,011 US adults), cited via CX Today and Retail Customer Experience trade coverage
• n8n.io/pricing, zapier.com/pricing, make.com/en/pricing — official vendor pricing pages, fetched directly, August 2026 snapshot
• Intercom (Fin) and Zendesk, disclosed per-resolution pricing ($0.99 and roughly $2.00 respectively)

#Directional (consistent secondary-source pattern, no single primary disclosure)

• n8n 2.0 / native LangChain integration / 70+ AI nodes, released January 2026 — digitalapplied.com, jahanzaib.ai, finbyz.tech, and other 2026 trade write-ups; not independently confirmed against n8n's own changelog in this research pass
• Make AI Agents, launched February 11, 2026 across all plans, and the AI-agent credit-consumption multiplier (roughly 43–50 credits per AI-agent execution versus a few for a classic step) — usecarly.com, dynalord.com, thinkpeak.ai, and other 2026 Make pricing guides
• Zapier Central/Agents 2026 pricing restructuring (activities billed separately from tasks) — lindy.ai, nocode.mba, usecarly.com
• Intercom Fin / Zendesk AI resolution-rate gap between vendor-claimed (70–80%) and independently documented production figures (44–53%), and the shared "any conversation that closes without human pickup counts as resolved" definition behind it — theaiagentindex.com, clonedesk.ai
• Small-business automation failure-mode patterns (tool sprawl, integration brittleness, "the first 100 real transactions surface most failure modes") — automatonagency.com, layer3labs.io, imversion.com, thinkbot.agency
• Payback-period ranges (6–14 months typical, 4–6 months fastest documented, versus an 18–24-month baseline three years earlier) and the 5–10%-of-build-cost annual maintenance-budget figure — kemenystudio.com, semnexus.com, alicelabs.ai, and other 2026 automation-consulting writeups. All three named sites are themselves AI-implementation/automation agencies selling the build service their own ROI content is describing — Alice Labs' own site advertises a specific "median 4.2-month payback period" blended from its own delivered projects alongside real Deloitte/IBM citations, which is the vendor-self-published-benchmark pattern this course explicitly discards elsewhere (see the ROI multipliers below). The 6–14-month range used in ROI and payback is the conservative end of what these vendor sources report, not their own headline figures, and is flagged there as optimistic for that reason.
• Administrative-task time-use breakdown (expense logging, scheduling, invoicing, data entry as the leading categories) — timeetc.com's operator-time survey coverage, cross-referenced against secondary aggregators citing NFIB's Small Business Economic Trends survey; the NFIB citation itself was not independently fetched from nfib.com in this pass
• McKinsey/S&P Global Market Intelligence figures on agentic AI production adoption (roughly 23% of organizations scaling an agentic system, roughly 31% of enterprises with at least one agent in production) — cited via a secondary aggregator (digitalapplied.com), not independently confirmed against a primary McKinsey report in this research pass

#Speculative / marketing-tier — named and explicitly discarded, not used as claims in any lesson above

• "280–520% ROI in year one," "300–1000% ROI in year one," "35% average operational cost reduction," "$3.50 returned per $1 invested in AI customer service" — traced to SEO-style content (raasautomazioni.it, ai-crescent.com, yotomations.com, orbilontech.com, adai.news, deantek.co) with no disclosed sample size or methodology. See ROI and payback for why these are named rather than silently omitted.
• The unresolved AI-agent-pilot-failure-rate conflict: figures of 88%, 89%, and 77% for "the share of AI agent pilots that never reach production" all circulate across 2026 aggregator content, attributed inconsistently to Forrester/Anaconda, Gartner/IDC, and McKinsey respectively by different secondary sites — sometimes for the same underlying number. This research could not resolve which attribution, if any, is accurate, and none of the three specific percentages appears anywhere in this course's lessons as a result. Only the more narrowly-traceable Gartner figure — over 40% of agentic AI projects (not "pilots," a distinct claim) canceled by end of 2027, sourced to a named press release with a disclosed 3,400+-organization poll — is used in KPIs and kill switches and Failure modes, because its methodology could actually be traced. Treat any "88%/89%/77% of pilots fail" claim you encounter elsewhere as unverified until you find its actual primary source.