Synthesis at Scale: How Top Strategy Firms Extract Maximum Value From Primary Research

The best consulting engagements are not won on frameworks. They are won on proprietary insight. A firm that has spent three weeks in deep conversation with the twenty most knowledgeable operators in a market arrives at the client presentation with something no slide template can replicate: a grounded, tested, evidence-rich point of view that competitors simply do not have.

In a world where every major strategy firm has access to the same secondary research — the same industry reports, the same earnings transcripts, the same analyst briefings — primary research through expert calls is the primary source of differentiated insight. The firms that consistently produce the most compelling strategy work are, without exception, the firms that have built the most rigorous primary research practices.

The problem is not access. Expert network capacity has expanded dramatically over the past decade. Any reasonably resourced engagement team can schedule thirty expert calls in two weeks. The problem is synthesis. As call volume has scaled, the synthesis bottleneck has become the primary constraint on research quality — and most firms have not meaningfully updated their synthesis practices to keep pace.

A team that runs thirty expert calls but synthesizes them poorly produces less value than a team that runs twelve calls with structured synthesis. The calls are the input. Synthesis is the product. And for most engagement teams, the synthesis process is improvised, rushed, and structurally flawed in ways that are entirely preventable.

This article lays out the synthesis infrastructure that top strategy firms are building: the four-layer synthesis model, the daily synthesis cadence, protocols for managing expert volume without losing nuance, and the emerging opportunity in cross-engagement knowledge capture. These are not theoretical frameworks — they are operational practices drawn from the research programs of firms that have solved the synthesis problem at scale.

The Synthesis Bottleneck: Why More Calls Create More Problems

The intuition behind running more expert calls is sound: more data points, more perspectives, more confidence in the findings. In practice, beyond a certain threshold, additional calls without a corresponding upgrade in synthesis infrastructure begin to degrade rather than improve output quality. The reason is structural.

Consider the reading stack problem. Thirty transcripts at an average of eight thousand words each is two hundred and forty thousand words of unprocessed content. Reading this material linearly — the default approach for most teams — takes twenty or more hours of focused work. That is an entire working week for a single analyst, before any synthesis has occurred. Most engagement teams do not have that time, which means the reading stack is rarely fully processed, which means significant portions of the evidence base are simply never incorporated into the final deliverable.

The second failure mode is recency bias. In the absence of a structured synthesis protocol, the last two or three calls disproportionately influence the final output because they are freshest in the team's working memory. An expert interviewed in week one of a three-week program is, cognitively speaking, ancient history by the time synthesis begins. Their views — which may have been the most precisely relevant to the client's question — are systematically underweighted relative to experts interviewed in the final days.

The third failure mode is the authority gradient. When a team includes one or two senior, credentialed experts alongside a larger group of mid-level practitioners, the senior expert views tend to dominate the synthesis regardless of how well-evidenced they actually are. A former C-suite executive who speaks with confidence but has been out of active industry for five years may be less current than a mid-level operator who joined the company eighteen months ago — but their authority signal overrides careful weighing of evidence quality.

Together, these three failure modes — the reading stack, recency bias, and the authority gradient — mean that even research programs with excellent expert selection and excellent call quality frequently produce deliverables that substantially underrepresent the quality of the underlying evidence. The synthesis process destroys value that the call program created.

The solution is not to run fewer calls. The solution is to build a synthesis architecture that keeps pace with call volume — one that processes evidence continuously rather than in a single compressed burst at the end of the program. The four-layer synthesis model is the foundation of that architecture.

The Four-Layer Synthesis Model

The four-layer synthesis model treats the synthesis process as a pipeline with four discrete stages, each operating at a different level of abstraction. The key discipline of the model is that each stage must be completed before the next can begin — and that the pipeline runs continuously throughout the research program, not in a single end-of-engagement burst.

Layer 1: Claim Extraction (Call Level)

Within twenty-four hours of each expert call, the responsible analyst extracts all discrete claims from the transcript or call notes. A claim is a single, falsifiable assertion: a statement of fact, causation, prediction, or evaluation that the expert has made. The target is twenty-five to forty claims per call — a number that seems high until analysts develop the discipline of granular extraction.

Claim extraction is not summarisation. A summary collapses and loses information. Claim extraction preserves every discrete assertion so that nothing is lost in the transition from raw transcript to structured evidence. Each claim is tagged with the expert's identifier, the date of the call, and an initial confidence rating. This is the raw material layer: the foundation on which all subsequent synthesis is built.

The twenty-four-hour constraint is not arbitrary. Research on memory and cognitive processing consistently shows that claims extracted within twenty-four hours of a call are more precisely rendered and more accurately attributed than claims extracted days later. The analyst who waits three days to extract claims from a transcript is working from a degraded memory of the call, even with notes available.

Layer 2: Theme Clustering (Cross-Call Level)

Once claims have been extracted from multiple calls, they are grouped into five to eight thematic clusters that map directly to the client's strategic question. The clusters are not defined in advance and imposed on the evidence — they emerge from the evidence itself, though guided by the team's understanding of the client's decision context.

Claims that do not map cleanly to any cluster are placed in a residual pool — a holding category for evidence that is potentially important but does not yet have a home in the emerging synthesis. The residual pool is not a discard pile. It is a signal register for themes that have not yet crystallised. Its management is discussed in detail in Section 4.

Theme clustering is the layer at which the research program transitions from individual call intelligence to collective intelligence. Individual calls reveal what a single expert knows or believes. Clusters reveal what the expert community collectively knows or believes — and where it agrees, disagrees, or is genuinely uncertain.

Layer 3: Tension Identification (Synthesis Level)

Within each theme cluster, the synthesis team identifies the one to three most important tensions: claims that directly contradict each other, or that reveal a genuine strategic uncertainty. Tension identification is the core analytical act of the synthesis process. It requires the team to resist the temptation to resolve contradictions prematurely and instead to sit with them, examine them, and understand what they mean.

A tension is not simply a contradiction. A contradiction is two experts saying opposite things. A tension is a contradiction that has strategic implications — where knowing which claim is correct, or understanding why experts disagree, would meaningfully change the client's decision. Not all contradictions are tensions, but all genuine tensions are contradictions.

Identifying tensions also serves a secondary purpose: it focuses follow-up call activity. When a tension is identified mid-program, the team can design subsequent calls specifically to probe the tension — recruiting experts with direct relevant experience, asking targeted questions, and deliberately seeking out the minority view to stress-test the emerging consensus.

Layer 4: Insight Articulation (Deliverable Level)

Each tension — whether resolved or unresolved — becomes an insight statement. The standard format is: 'Expert evidence suggests X, but a minority view suggests Y, and the resolution depends on Z.' This format is deliberately structured to convey three things simultaneously: what the evidence says, what the evidence does not settle, and what additional information or analysis would resolve the remaining uncertainty.

Insight articulation is the layer at which primary research intelligence becomes client-relevant strategy content. Findings describe what the evidence shows. Insights describe what the evidence means — and what decisions it should inform. The transition from findings to insights is not automatic; it requires explicit analytical effort and is where the most experienced team members should be most actively engaged.

The four-layer model is a pipeline, not a checklist. Each layer feeds the next, and the pipeline runs continuously. A team running a three-week, thirty-call research program should have completed Layer 1 for every call, Layer 2 across all completed calls, and Layer 3 for at least two or three theme clusters by the end of week two — well before the final calls are conducted.

The Daily Synthesis Cadence

The most common synthesis failure mode is not a failure of method — it is a failure of timing. Teams understand, in principle, that synthesis should happen throughout the engagement. In practice, synthesis is deferred to the end because there is always something more pressing: the next call to prepare for, the next slide to draft, the next client check-in to manage.

The solution is to institutionalise synthesis through a daily cadence that makes it a non-negotiable part of the engagement workflow rather than an activity that competes with other priorities. The daily synthesis session is thirty minutes, held at the end of each call-intensive day, with the full engagement team present.

The agenda is standardised: review claims extracted from the day's calls; update theme cluster assignments for new claims; flag any new tensions identified in the day's material; revise the insight draft to incorporate the day's evidence. The session is not a meeting about the research — it is a working session that advances the synthesis pipeline.

The daily cadence addresses three specific pathologies. Information decay: claims become less vivid and less precisely retrievable within seventy-two hours of the call that generated them; daily synthesis captures claims while they are still fresh. Theme drift: in the absence of daily explicit synthesis, the team's collective framing of the research question shifts imperceptibly over the course of the engagement, often without any explicit decision to change direction; daily synthesis makes framing shifts explicit and deliberate. Recency bias: by incorporating each call's evidence into the synthesis on the day it is conducted, the daily cadence ensures that early-program calls are given equal structural weight to late-program calls.

The insight draft is the central artifact of the daily synthesis process. It is a living document, started on day one, updated at every daily session, and maintained as the team's working hypothesis about what the evidence shows. By the time the final expert call is conducted, the insight draft should be sixty to seventy percent complete — meaning that the final synthesis session is a refinement exercise rather than a from-scratch construction.

Teams that implement the daily synthesis cadence consistently report that total synthesis time decreases even as the quality of the final deliverable improves. This is not paradoxical — it reflects the fact that synthesis work done in small daily increments is cognitively less expensive than synthesis work done in a single large compressed session. Distributed synthesis is both faster and better.

Managing Expert Volume Without Losing Nuance

Structured synthesis carries its own risk: over-aggregation. The discipline of sorting claims into clusters and looking for consensus can create pressure to round off the sharp edges of the evidence — to fit every claim into a theme, to treat every outlier as noise, to privilege the majority view at the expense of the dissenting signal that might be the most important finding in the program.

A claim that resists clustering — that does not fit neatly into any of the established thematic categories — is not necessarily wrong or irrelevant. It may be the early signal of a theme that has not yet crystallised; it may be the only piece of evidence from an expert with unique direct experience of the specific question at hand; it may be a genuine outlier that, if correct, would fundamentally alter the engagement's conclusions.

The residual pool protocol addresses this risk. Claims that do not cluster easily are placed in the residual pool rather than forced into the nearest available cluster. The residual pool is reviewed weekly by a senior team member — someone with enough experience to recognise when an unclustered claim is genuinely anomalous versus when it represents an emerging theme that the cluster structure has not yet caught up with.

Confidence weighting is the second tool for managing expert volume without losing nuance. In a thirty-call program, not all expert views deserve equal weight in the synthesis. A five-factor confidence model provides a structured basis for differential weighting: source proximity (how directly relevant is the expert's experience to the specific question?), temporal relevance (how recent is the experience?), specificity (how precisely does the claim address the question?), corroboration (how many other experts have made the same or similar claim?), and track record (where the expert is known to the firm, has their prior analysis proven accurate?).

Applying the five-factor model does not require precise numerical scoring. In practice, a three-tier classification — high, medium, and base confidence — is sufficient to prevent the authority gradient failure mode and to ensure that synthesis gives appropriate weight to well-evidenced claims from less credentialed sources and appropriate skepticism to confident-sounding claims from highly credentialed sources whose direct experience may be dated or indirect.

Cross-Engagement Knowledge Capture

The most underutilised asset in consulting primary research is the accumulated intelligence from prior engagements. A firm that has conducted primary research in the logistics sector across fifteen engagements over five years holds a claim register of potentially ten thousand discrete expert assertions about that sector. This is a proprietary knowledge asset of extraordinary value — and for most firms, it is essentially inaccessible.

The structural reason for this inaccessibility is that primary research outputs are typically stored as deliverable documents — polished client reports, PowerPoint presentations, executive summaries — rather than as structured evidence repositories. Finding the relevant prior insights requires knowing which prior engagement addressed the relevant question, locating the deliverable, reading it in full, and hoping that the synthesis in that deliverable preserved the specific claim you are looking for. In practice, analysts simply do not do this. The institutional knowledge exists but cannot be accessed within the time constraints of an active engagement.

The opportunity is to build a searchable claim register at the firm or practice level. Each claim extracted from an expert call — tagged with sector, geography, functional area, confidence rating, date, and engagement — is added to a shared claim database that can be queried by analysts on subsequent engagements. The infrastructure requirement is modest; the operational discipline requirement — consistent claim extraction across all engagements — is more significant but achievable for firms that have already implemented the four-layer synthesis model.

The practical value is immediate. An analyst beginning a new engagement in the freight and logistics sector should be able to query: 'What do we know about last-mile delivery economics from prior programs?' and receive a structured summary of every relevant claim the firm has extracted from expert calls in that area, with confidence ratings, dates, and source context. This starting point eliminates the weeks of orientation that currently precede every new engagement in familiar sectors.

The compounding effect is the more significant long-term value. Firms that build cross-engagement knowledge infrastructure get smarter on every engagement, not just within it. The quality of their primary research work in any given sector improves as a function of the number of prior engagements they have conducted in that sector — a structural advantage that deepens over time and is very difficult for competitors to replicate without making the same investment in synthesis infrastructure.

Synthesis for Client Communication

The final synthesis challenge is translation: converting structured intelligence into client-ready communication that conveys both the substance of the evidence and the team's analytical confidence in what it means. This translation is not merely a communication exercise — it is a final analytical act that forces the team to make explicit commitments about what the evidence supports and what it does not.

The first principle of evidence-to-communication translation is to lead with the insight, not the source. Clients do not want to hear what experts said — they want to hear what the evidence means. 'EXP-012 indicated that margins are compressing' is a transcript summary. 'Margin compression is underway across the category, driven by input cost inflation that suppliers cannot pass through to buyers' is an insight. The transition from the first formulation to the second requires the team to take analytical ownership of the claim — to put their own credibility behind it rather than hiding behind the source.

The second principle is to quantify confidence explicitly. Vague confidence language — 'experts believe,' 'there is evidence that,' 'our research suggests' — communicates nothing about the strength of the underlying evidence and is rightly viewed with skepticism by sophisticated clients. Precise confidence language — 'seven of nine experts with direct category experience described this dynamic,' 'this view was corroborated across all five geographic markets we covered' — conveys genuine information about evidential weight and builds the client's confidence in the analysis.

The third principle is to surface the minority view. Engagements that present only the consensus finding are presenting an incomplete picture of the evidence — and, more importantly, are failing to give the client the information they need to make a properly calibrated decision. The standard format for presenting contested findings is: 'X is the consensus view, supported by the majority of our expert conversations; the dissenting view, held by two of nine experts with relevant experience, suggests Y — and the implication if Y is correct is Z.' This formulation respects the consensus while preserving the signal value of the minority view.

Surfacing the minority view also serves a risk management function. When a client later encounters information that contradicts the consensus finding, a deliverable that acknowledged the minority view demonstrates analytical integrity. A deliverable that presented only the consensus finding looks, in retrospect, like the team cherry-picked the evidence. The firms with the strongest long-term client relationships are, without exception, the firms that have built a reputation for complete, calibrated honesty about what their evidence shows — including what it does not settle.

Conclusion

Synthesis at scale is a discipline, not a talent. The teams that do it well have built infrastructure: claim extraction protocols that run within twenty-four hours of every call, daily synthesis sessions that keep the evidence pipeline current, residual pool management that protects against over-aggregation, and cross-engagement knowledge capture that compounds institutional intelligence across engagements.

None of this infrastructure is technically complex. The four-layer synthesis model, the daily synthesis cadence, the residual pool protocol, the five-factor confidence model — these are operational disciplines, not advanced analytical frameworks. They do not require specialist technology or specialized expertise. They require the commitment to treat synthesis as a first-class activity that deserves the same investment of time and structure as the call program itself.

The consulting firms that will consistently produce the highest-quality primary research deliverables over the next decade are not the ones with the largest expert network budgets. They are the ones that have solved the synthesis problem — the ones that have recognised that call volume is not the binding constraint on primary research quality, and have invested accordingly in the processes that convert raw expert intelligence into strategic insight.

The calls are the input. Synthesis is the product. The firms that understand this distinction — and build for it — are the ones that will define the standard for primary research excellence in the years ahead.