T2

See the scoped evidence_register and qualification_protocol. Structural plausibility, descriptive surveys and laboratory benchmarks do not alone identify a local commercial policy effect.

See the scoped evidence_register and qualification_protocol. Structural plausibility, descriptive surveys and laboratory benchmarks do not alone identify a local commercial policy effect.

Revenue is volume times the product of stage conversions times price. In log space, contributions are additive. All rates have unit elasticity for equal proportional improvements, holding other inputs fixed. For equal absolute rate changes, dR/dr_i=R/r_i, but investment return also depends on dr_i/dspend, downstream quality/rates, price, capacity and time. A low rate can be a useful filter; select the best feasible incremental contribution intervention, not min(r_i).

• G2 — The Evaluation Maze — 22 July 2026. Primary report. June online survey of 1,038 decision makers plus 55 interviews. Forty percent name evaluation their longest stage; 39% name security the largest post-selection delay; 49% report a CFO veto in the past year. Source: https://sell.g2.com/2026-buyer-behavior-report
• Shuxian Fan and colleagues — ShoppingReasoningBench — 10 June 2026. Full preprint. Seven experts construct 525 missions, 1,996 turns and 10,863 weighted binary rubrics across five product families; nine models tested. Optional constraints and multi-turn tasks are harder. Source: https://arxiv.org/html/2606.12608v1
• Haobo Zhang and colleagues — Learning from Online User Feedback for Shopping Agents — 12 August 2026. Full preprint. JD.com real-log offline study: 7,001 behavioral sessions from 6,704 users plus 6,121 directive sessions; purchase-aligned and corrective training. Source: https://arxiv.org/html/2608.11604v1
• Carbon — B2B Buyers Report — September 2026. Question-level findings. Commercial survey, 189 quality-filtered decision makers; denominators vary by question. High reported satisfaction coexists with provider reconsideration. Source: https://explorecarbon.com/b2b-buyers-report/findings/
• Xiaolin Li, Qing Li, Danny Wang and Flora Gu — AI Human Collaboration in Sequential Sales Funnels — 13 June 2026. Primary conference abstract. Abstract describes randomized SME lending study with 78,169 prospects: AI initial calls and intent tags, then human follow-up. Source: https://scholars.hkbu.edu.hk/en/publications/ai-human-collaboration-in-sequential-sales-funnels-evidence-from-/

• No fitted local coefficient or complete-policy deployment evidence is supplied.
• 40% is a respondent share, not elapsed journey time. Nine percent say they would permit guarded purchasing and 2% purchasing without pre-approval; those are authority preferences, not observed adoption. Current/planned use remains separate. Sponsor/self-report selection prevents causal conclusions. C2/C4/C6/C8.
• scored recommendations, not actual purchase welfare or commercial effects; one generation per query and excluded categories. Versions and time-sensitive facts matter. C2/C4/C8.
• behavioral training selects successful purchases; issue-resolution evaluation partly uses an LLM judge. Better offline purchased-item retrieval does not establish live conversion, welfare or retention uplift. C4/C6/C8.

All rates have unit elasticity for equal proportional improvements, holding other inputs fixed. For equal absolute rate changes, dR/dr_i=R/r_i, but investment return also depends on dr_i/dspend, downstream quality/rates, price, capacity and time. A low rate can be a useful filter; select the best feasible incremental contribution intervention, not min(r_i).