Expert Networks & AI Disruption in Professional Research

This programme explored how finance, consulting, healthcare, and public-sector professionals use expert networks (GLG, Tegus, AlphaSights, Third Bridge), how usage is changing, core pain points, and openness to AI alternatives — AI-moderated interviews, transcript reuse, AI trained on transcripts, and synthetic experts.

Research group: 10 senior U.S.-based professionals across buy-side investing (equity/credit), corporate ops/strategy, healthcare provider finance/ops, and public-sector K-12 ops.

What they said: Teams follow a disciplined playbook — tightly scoped decisions, primary/public sources first, operator/peer references and small pilots — treating expert networks as transcript-first inputs and reserving live calls for last-mile, high-stakes gaps. Spend has been repriced and repurposed: more transcript mining; fewer, sharper calls; tighter procurement, MNPI/PHI controls, and demand for auditable outputs. The value in calls is "scar-tissue" detail; the friction is screening, scheduling, pricing opacity, and weak deliverables.

• Disciplined, transcript-first playbook. Teams tightly scope decisions (one-pagers, kill criteria), mine primary/public sources, and lean on operator/peer references and small pilots; expert networks are ancillary, late-stage gap-closers. Evidence: Kimberly Strait's one-pager framing; Cleotilde Seymour's base case from filings and regulator data; Dale Gomez's "read 8–12 transcripts before the first call."
• Spend is repriced and repurposed, not simply cut. Transcript consumption is up; live calls are reserved for last-mile, high-stakes inflections given budget/procurement scrutiny and compliance/MNPI risk. Evidence: Paula Thomas ("No bundles, no annuals"); Sonia Daniel ("MNPI/PHI paranoia," calls only at regulatory/staffing pivots); Dale Gomez (transcripts up, calls flat to slightly down).
• Calls deliver scar-tissue value — and heavy friction. High-leverage when surfacing tacit realities that change models (e.g., retrofit bottlenecks), but burdened by screening misfit, scheduling/cancels, opaque pricing and fees, and compliance overhead. Evidence: Sonia Daniel's "Tuesday-at-2 p.m. reality"; Jason Talbot's mis-screened "operator"; Kimberly Strait on rate bloat and procurement drag; Dale Gomez on a single GLG call killing an OEM thesis.
• AI welcomed as a co-pilot, not the pilot. Trusted for logistics, verbatim capture, timestamps, extraction, and cross-call contradiction finding; distrusted for autonomous interviewing, reading subtext, or final compliance calls. Conditions: verbatim audio + timestamps, tenant-isolated data custody, MNPI guardrails, audit trails, and human sign-off. Evidence: Jason Talbot ("co-pilot"); Kimberly Strait ("No summary-only deliverables"); Paula Thomas (AI for scheduling/prep); Cleotilde Seymour's refusal to delegate end-to-end; Michelle Thao's requirement for raw recordings and expert attestations.

• Healthcare provider finance/ops (Sonia Daniel, Michelle Thao): auditability and MNPI/PHI drive transcript-first, with sparse, decisive calls when staffing, credentialing, or reimbursement assumptions swing the P&L.
• Public-sector K-12 / rural ops (Herbert Proctor): paid networks are largely untenable due to procurement optics and open-records risk; rely on peers, board packets, and pilots.
• Buy-side investors (Dale Gomez, Jason Talbot, Cleotilde Seymour): broad transcript canvases (8–12) plus 1–3 razor-scoped calls when they can materially move position sizing; strong demand for provenance and de-duplication.
• Corporate ops / management analysts (Kimberly Strait, Paula Thomas, Chelsea Chavez, Anthony Strand): prefer operator references, scripted demos, and pilots; networks used for late-stage confirmation. Bilingual / code-switching needs surfaced by Chelsea Chavez for accurate capture with operator populations.

Recommendations

• Build a provenance-first transcript-intelligence layer. Quote-level search with timestamps, role/tenure/region metadata, freshness, numeric extraction with units, de-duplication across sources, contradiction panels, and audit-ready exports.
• Offer a human-led AI call co-pilot. Automate scheduling, consent, recording, live bookmarking, MNPI flagging, and post-call citations — while keeping a human interviewer in control.
• Ship compliance and procurement enablers. Tenant isolation and no-train data custody; consent artifacts; MNPI flags; simple per-seat / day-pass / per-project pricing.

Risks & measurement guardrails

• Licensing/rights on third-party transcripts: prioritize client-owned ingestion; negotiate rights; embed export controls.
• Hallucination / false precision: enforce quote-only for decision-grade outputs, show ranges with N and recency, contradiction banners.
• Stale or biased corpus: default recency filters, role/region weighting, coverage maps that prompt fresh calls when needed.
• Security/compliance failures: SOC 2 pathway, HIPAA-adjacent controls, SSO, audit logs, retention controls.

KPIs

Calls avoided per project (≥1.5) · prep-time reduction (≥50%) · citation coverage (100% of decision-grade outputs) · compliance incidents (0; all flags logged) · pilot-to-paid conversion (≥60%).

• 0–30 days: launch quote-first search with timestamps/recency; deliver "Audit Pack" exports; publish procurement-ready pricing and security kit.
• 30–60 days: run pilots in buy-side, healthcare ops, and corporate ops; add freshness and contradiction banners; finalize data-custody policies.
• 60–90 days: ship Transcript Intelligence v1 (numeric extraction, de-dup, coverage map, CSV/memo exports); beta the human-led co-pilot.
• 90–180 days: add client connectors (S3/SharePoint/GDrive) and first network partnerships; harden the compliance engine; introduce bilingual capture where operator populations require it.