Consultants are paid for judgment under uncertainty, not for fluent summaries of source documents. When I design research workflows around Opus 5.2 I therefore treat the model as a high-speed evidence organizer and gap detector, not as a junior analyst authorized to reach conclusions. The workflow below is the version I use and recommend for typical consulting research cycles—bounded source sets, a clear decision the research must support, and a client who will ask “how do you know?” The process is deliberately conservative. It produces less polished narrative and more traceable reasoning. That is the feature.
Why Consulting Research Needs a Different Standard
Client Questions Target the Evidence, Not the Prose
A client can accept a less elegant sentence. A client will not accept a recommendation whose supporting evidence cannot be located or whose gaps were never acknowledged. The workflow must optimize for traceability and gap visibility.
Sources Are Almost Always Incomplete and Sometimes Conflicting
Real consulting source packs contain contradictions, outdated figures, and missing causal links. A workflow that assumes clean, consistent material will systematically overstate confidence.
The Consultant Remains Personally Accountable
No model output travels to the client without the consultant’s name on it. The process must make that accountability operational through mandatory review points rather than treating it as an after-the-fact formality.
The Workflow in Sequence
Stage 1: Frame the Decision the Research Must Support
Before any source is processed, write one sentence that states the decision or recommendation the research is intended to inform. This sentence becomes the filter for relevance and the anchor for later gap analysis. Without it, synthesis drifts toward general summary.
Stage 2: Inventory and Light-Tag the Source Set
List every source, note its date, type, and apparent reliability, and flag any known limitations. This inventory is later used to detect over-reliance on weak material and to explain why certain questions remain open.
Stage 3: Extract Claims With Source Pointers
For each major source, extract the claims that bear on the decision. Every claim must carry a pointer back to the location in the source. Free-floating claims are rejected. Opus 5.2 is used here for speed of extraction under strict citation rules.
Stage 4: Build an Explicit Conflict and Gap Map
Compare extracted claims across sources. Surface contradictions, chronological inconsistencies, and questions the sources simply do not address. This map is often the most valuable artifact the workflow produces. It prevents the false coherence that clients later punish.
Stage 5: Draft a Cautious Synthesis and Recommendation
Only after the conflict-and-gap map exists is a synthesis attempted. The recommendation must restate the strength of evidence and the remaining uncertainties. Language that erases gaps is treated as a failure.
Stage 6: Human Review Against the Original Sources
The consultant verifies every material claim and every residual uncertainty against the primary sources. Model fluency is never accepted as a substitute for this check. Time for the review is budgeted in advance.
Instruction Patterns That Keep the Model Inside Useful Bounds
Require Source Pointers on Every Extracted Claim
The prompt forbids claims without location references. This single rule eliminates a large class of unverifiable statements and makes the later human review dramatically faster.
Separate Observation From Interpretation in the Output Structure
Force the model to place directly supported claims in one section and interpretive or inferential statements in another. Mixing the two is the most common path to over-confident recommendations.
Mandate an Explicit “What the Sources Do Not Settle” Section
This section is non-negotiable. Its presence keeps residual uncertainty visible and gives the client a clear view of where judgment, rather than evidence, is operating.
The following table summarizes the stages, primary outputs, and the review focus at each point:
Stage | Primary Output | Review Focus |
|---|---|---|
1 | Decision framing sentence | Clarity and scope of the target decision |
2 | Source inventory with limitations | Completeness and known weaknesses |
3 | Claims with source pointers | Verifiability of every claim |
4 | Conflict and gap map | Honesty about contradictions and absences |
5 | Cautious synthesis and recommendation | Evidence strength and uncertainty language |
6 | Final human-verified deliverable | Full source-backed accountability |
Common Failure Modes Specific to Consulting Research
Premature Narrative Coherence
The model produces a smooth story that quietly resolves conflicts the sources left open. Defense: the conflict-and-gap map is required before any synthesis is generated.
Over-Weighting the Most Fluent Source
Longer or better-written documents receive disproportionate attention. Defense: the source inventory includes explicit reliability and recency notes that the synthesis prompt must acknowledge.
Recommendation Language That Outruns the Evidence
The final section drifts into advice the sources cannot support. Defense: the prompt requires every recommendation to restate the supporting evidence strength; human review restores hedging where needed.
How to Introduce the Workflow in a Consulting Practice
Begin With Internal or Low-Stakes Research Cycles
Use the full process on internal strategy questions or low-visibility projects first. The discipline becomes habitual before it is applied to high-visibility client deliverables.
Make the Gap Map a Client-Facing Artifact When Appropriate
In many engagements the honest display of what remains unknown is more valuable than a forced conclusion. Clients who see the gaps early are less likely to feel misled later.
Track the Ratio of Extraction Time to Review Time
If review consistently dominates, the extraction instructions or the source quality need attention. The goal is net reduction in the consultant’s total effort to reach a defensible position.
Closing the Research Workflow
A consulting research process built around Opus 5.2 should make evidence and its limits more visible, not less. The model accelerates extraction and comparison. The consultant retains responsibility for framing, verification, and the final judgment that goes to the client.
Useful remains a higher bar than impressive. A research workflow earns its place when the resulting recommendation can be traced, the gaps are acknowledged, and the consultant can defend every material claim under questioning. The sequence above is designed to clear that bar more consistently than any single-pass “research and recommend” request.
