How Buy-Side Firms Handle Expert-Network Call Prep Documents: 7 Retention Models

Pre-call prep documents live in a gray zone. The analyst's question list, the agenda sent to the expert, the background memo from the network, and any pre-reads are neither the recording nor the note nor the invoice, but they can carry the analyst's thesis, restricted-list references, and sometimes the expert's identity. Buy-side firms handle them inconsistently, and the retention model a firm picks determines both its compliance exposure and how reusable the material becomes for downstream AI research systems.

Below are seven models in active use across hedge funds, long-only shops, and RIAs. None is universally correct. The choice reflects the firm's structure, its regulator, and its appetite for institutional memory versus thesis-leakage risk.

1. Delete-on-Completion

The most restrictive model auto-purges prep documents 24 to 72 hours after the call ends. Nothing survives except the compliance-cleared transcript or note. This is common at pod-shop multi-managers where thesis leakage between pods is the dominant risk, and where institutional memory is deliberately fragmented so that no single artifact reveals what a portfolio manager was working on.

The logic ties directly to Chinese-wall policy. If a question list about a semiconductor supplier sits in a shared drive, it can be discovered by an adjacent pod running a correlated book. Auto-purge removes the artifact before that becomes possible. The trade-off is real: the next analyst who picks up the name has no record of what the firm already asked, and the same expert may be re-briefed on questions the firm already answered six months earlier.

2. Attach-to-Call-Record in the Research CMS

The opposite model bundles the prep doc with the transcript and the analyst's note inside the research content management system. Platforms like Bipsync, Sentieo, and AlphaSense Enterprise Intelligence support this natively: the prep doc becomes an attachment on the call record, tagged to the ticker and the expert, and any future analyst covering the name sees what was asked and why.

This is the standard at long-only shops where coverage is continuous, analyst turnover is a real cost, and institutional memory is a genuine asset. A Fidelity or Capital Group analyst inheriting coverage of a mid-cap industrial wants the last five years of expert-call prep as context, not just the final notes. The compliance posture assumes that a well-governed CMS with role-based access is a defensible home for thesis-adjacent material.

3. Analyst-Owned Local Retention

At smaller hedge funds, often under about $2B in AUM, prep documents never enter a central system at all. They live in the analyst's OneDrive, Notion workspace, or a local folder. When the analyst leaves, the prep leaves with them, unless the firm has a specific offboarding process that captures it.

This model is less a deliberate policy than an artifact of scale. Firms below a certain size have not built the CMS infrastructure that makes central retention practical, and compliance is handled by an outsourced CCO who focuses on the transcript and the invoice rather than the pre-call artifacts. The exposure is uneven: some analysts are disciplined, some are not, and the firm's actual retention posture is the average of individual habits.

4. Expert-Network-Side Retention Only

Some firms rely on the expert network itself to hold the agenda. The network distributed it, the network keeps it, and the firm pulls it back on request. Both GLG and Guidepoint publish compliance policies that address document retention on their side.

This shifts the storage burden but concentrates the discovery risk. Expert networks receive subpoenas, and material held by a vendor is discoverable through that vendor. A firm that has decided its own retention footprint should be minimal has, in effect, outsourced the footprint rather than eliminated it. In enforcement contexts, regulators can and do go to the network directly.

5. Compliance-Vaulted With Time-Lock

A fifth model routes prep documents to a compliance archive, typically Global Relay, Smarsh, or Proofpoint, with a hold period of three to seven years and access gated by a compliance ticket. The analyst who wrote the prep cannot pull it back without a documented reason.

The hold periods reflect regulatory floors. SEC Rule 17a-4 requires broker-dealers to retain certain communications for three years, with the first two years in an easily accessible place. The Advisers Act Rule 204-2 sets a five-year retention requirement for registered investment advisers, and the FCA's record-keeping regime and SEBI's parallel rules impose their own timelines. Compliance-vaulting treats prep documents as communications-like artifacts and applies the strictest applicable rule.

This model preserves everything, but it also removes the material from the analyst's workflow. Reuse for AI research systems is possible only after a compliance review, which in practice means the material is retained for defense rather than for productive downstream use.

6. Ingested Into a Vector Index for the AI Research Copilot

The fastest-growing model chunks prep documents into the firm's AI research index , Rogo, Hebbia, or an internal LLM system , so future queries surface prior question lists. An analyst asking the copilot "what did we ask the last channel-check expert on this supplier" gets an answer grounded in the firm's own prep history.

The productivity case is clear. The compliance case is less mature. Embedding prep documents into a vector store creates several problems at once: the material is now retrievable by any user with query access rather than by document-level permissions, restricted-list references may surface in unrelated queries, and the embeddings themselves are difficult to redact or delete cleanly once created. The lock-in to the specific vector provider also becomes structural, since re-indexing years of prep material into a different system is a substantial project.

Firms adopting this model are increasingly separating the ingestion step from the retention decision. Prep documents may be indexed for a defined window, then aged out of the vector store while the source document itself follows one of the other five paths.

7. Redacted-and-Retained as Training Material

The seventh model strips prep documents of ticker, thesis language, and restricted-list references, then keeps the sanitized version as a generic template. A prep memo for a call with a semiconductor equipment channel expert becomes a redacted example of "how we prep for a channel-check call," used to train junior analysts and standardize research quality.

This is the least common model, partly because redaction is manual and expensive, and partly because the value only compounds if the firm has a genuine analyst-training program that uses the material. Where it works, it converts a compliance liability into a training asset, and the residual document is generic enough that its retention creates minimal exposure.

How the Choice Actually Gets Made

Most firms do not sit down and pick one of these seven models cleanly. The retention posture emerges from three factors: the firm's regulatory register (broker-dealer, RIA, offshore fund, or some combination), the CMS and compliance infrastructure already in place, and the cultural weight given to institutional memory versus thesis containment. A pod shop with a strong Chinese-wall culture and a robust compliance archive will land near models 1 and 5. A long-only shop with continuous coverage and a mature CMS will land near model 2. A small hedge fund without central research infrastructure will land at model 3 by default.

The AI copilot question is the one reshaping the landscape. Firms that ingest prep documents into a vector index are effectively adding a sixth model on top of whatever they were already doing, and the interaction between the two retention regimes is where compliance teams are spending increasing time. The SEC's 2017 guidance on electronic messaging for advisers offered one framing for how communications should be captured across channels; the parallel question for AI-indexed research artifacts has not yet received the same regulatory clarity.