Most firms retain research outputs but lose the reasoning that made them useful. The reusable asset is not the old report; it is the chain connecting the question, assumptions, evidence, disagreement, decision, and what happened next.
A strategy team receives a question it has seen before.
Perhaps the company is reconsidering a market it assessed two years ago. A consulting team is entering an industry another office recently studied. An investment group is revisiting a theme that generated several prior diligence exercises. A policy team needs to understand a regulation its colleagues analyzed when the proposal first appeared.
The organization is not starting from zero. Somewhere there are presentations, transcripts, market studies, spreadsheets, emails, expert notes, and perhaps a final recommendation.
Yet the new team often behaves as though it is.
It searches for the latest market data. It reconstructs the competitive landscape. It commissions another round of interviews. It rediscovers disagreements that prior researchers already encountered. Only late in the process does someone remember a colleague who “did something on this last year.”
The obvious diagnosis is poor retrieval.
Often, it is not.
The deeper problem is that organizations preserve documents more reliably than they preserve research memory.
A final report tells a future team what the previous team concluded. It rarely preserves enough of the structure behind that conclusion: the precise decision being considered, assumptions that were uncertain, sources that carried disproportionate weight, disagreements that were never resolved, questions deliberately left open, confidence attached to major claims, and subsequent evidence that strengthened or weakened the original view.
Strip away that structure and a report becomes difficult to reuse safely. The next researcher must either trust the old conclusion without reconstructing its basis or reconstruct the entire inquiry before deciding what remains valid.
That accumulated reconstruction cost can be thought of as research debt.
Like technical debt, it does not necessarily appear as an explicit line item. It appears as duplicated research, repeated expert calls, lost provenance, slow onboarding, contradictory internal conclusions, and teams spending expensive time establishing what the organization should already know.
AI makes this problem more consequential, not less. Generative systems are making it dramatically easier to produce and search large volumes of analytical material. The scarce organizational capability increasingly becomes the ability to determine which prior reasoning is reusable, under what conditions, and what now needs to be refreshed.
The practical unit of institutional research memory should therefore not be the document.
It should be a reusable research object.
Repositories do not automatically create memory
Organizations have been building knowledge repositories for decades. The problem of reuse has survived them.
That is partly because knowledge retention is only one component of organizational learning. Linda Argote, Sunkee Lee, and Jisoo Park divide the process into four distinct activities: search, knowledge creation, knowledge retention, and knowledge transfer. An organization can therefore become quite good at storing what it has produced without becoming equally good at getting prior knowledge into a new decision.
The distinction becomes clearer in research on transactive memory systems.
Transactive memory is often summarized as knowing “who knows what.” It treats collective memory not simply as a common warehouse of facts but as a system in which knowledge is distributed and people understand where expertise resides. Ren and Argote's review of 76 studies found a substantial literature connecting such systems to learning and group performance, while also noting important questions about extending team-level transactive memory to whole organizations.
Empirical studies help explain why that distinction matters.
Lewis, Lange, and Gillis found that groups with an established transactive memory system and experience across multiple related tasks were better able to develop an abstract understanding that transferred to new tasks. But the effect was conditional: the authors did not find equally strong support after experience with only one prior task, and stability in who held expertise mattered.
In a different setting, Reagans, Argote, and Brooks studied joint-replacement teams in a teaching hospital. They found distinct performance contributions from individual experience, broader organizational experience, and experience working together, the last of which improves the team's ability to coordinate.
These settings are not research departments, and their findings should not be imported mechanically. But they point to an important organizational-memory principle:
It is not enough for knowledge to exist somewhere. Future users need a workable route back to the knowledge, the people, and the context that gave it meaning.
That is precisely where document repositories often fail. A presentation called Market_Entry_Final_v7.pptx may preserve the answer. It does not necessarily preserve why one source was believed over another, which assumption the team considered fragile, which former executive disputed the apparent consensus, or which question remained unanswered when the steering committee made its decision.
Search can find the file.
It cannot recover reasoning that was never stored.
The final output is the wrong unit of reuse
The traditional knowledge-management instinct is to save deliverables.
That makes sense operationally. Deliverables are visible, polished, and comparatively easy to classify. They are also often the point at which the richest parts of the inquiry have already been compressed away.
Research is a transformation process.
A team begins with a decision or uncertainty. It frames questions. It forms assumptions. It searches. It decides which sources are credible. It encounters contradiction. It makes judgment calls about what evidence means. It develops confidence in some propositions and leaves others unresolved. Eventually, all of that becomes a recommendation, memo, presentation, model, or briefing.
The final artifact is useful precisely because it compresses that process.
Compression, however, destroys information needed for later reuse.
M. Lynne Markus's influential work on knowledge reuse identified this problem from another direction. Different kinds of knowledge reusers have different needs, and repositories built around what the original producer naturally records often fail to meet those needs. Reuse can consequently require substantial rework, while the original knowledge producer has limited incentive to repurpose material for unknown future users.
A future analyst rarely wants merely to know, “What did the previous team conclude?”
They need to know:
What were they trying to decide?
Which facts were observations, and which were interpretations?
What assumptions had to hold?
Which sources were primary?
Where did informed people disagree?
What remained unresolved?
How certain was the team?
What action followed?
What happened afterward?
Without those connections, old research has ambiguous status. It is simultaneously too valuable to ignore and too opaque to trust.
The reusable unit should therefore be a minimum research object containing at least:
- Question and decision context: What uncertainty was being investigated, and what decision would the research inform?
- Source provenance: What evidence supported the work, when was it collected, and where did it originate?
- Live assumptions: Which propositions were being treated as true but remained uncertain?
- Unresolved questions and disagreement: What was not settled, and where did credible sources diverge?
- Confidence: How strongly did the researchers believe the central conclusions, and why?
- Owner: Who understands the inquiry well enough to interpret or update it?
- Refresh trigger: What change in time, market conditions, regulation, evidence, or decision context should cause the work to be revisited?
The deliverable can sit on top of this object.
It should not substitute for it.
Research debt compounds invisibly
Research debt starts small.
A team fails to link an expert transcript to the claim it informed. The consequence is negligible.
Six months later, another team encounters the same claim but cannot determine whether it came from one operator, five interviews, a public report, or a colleague's intuition. It searches again.
Then a third team finds both sets of research. Their conclusions differ, but there is no clear record of whether conditions changed or the teams simply interpreted the evidence differently.
Now someone must reconcile them.
The organization has paid for the same question several times: first to investigate it, then to rediscover it, then to understand why its own answers conflict.
BCG describes a version of this problem in M&A. Its 2026 research argues that insights from one deal often fail to carry into the next because data are fragmented and knowledge dissipates as teams move on. BCG highlights the opportunity to capture past interactions, decisions, rationales, assumptions, and subsequent outcomes in reusable form so that future transactions can learn from them.
The same mechanism appears across research-intensive organizations.
A management-consulting practice repeatedly investigates the economics of the same sector because knowledge is organized by engagement rather than question.
A public-markets team has years of company and industry research but cannot reconstruct how its original expectations compared with subsequent evidence.
A government-affairs organization produces excellent country updates while losing the assumptions behind earlier regulatory scenarios.
A corporate strategy group retains board presentations but not the rejected alternatives or evidence that caused the recommendation to change.
The cost is not just repeated desk research.
Research debt removes the organization's ability to distinguish new learning from rediscovery.
If a team reaches the same conclusion six months later, did evidence strengthen the thesis? Or did a new group simply reconstruct the same analysis from scratch?
If the conclusion changed, did the world change? Did new evidence arrive? Did different experts get consulted? Or did a different analyst frame the question differently?
Those are valuable organizational-learning questions.
A folder of finished reports often cannot answer them.
AI makes context quality more important, not less
One might expect AI to eliminate research debt by making everything searchable.
It will help considerably with retrieval.
It may also make the underlying problem worse.
The volume of generated research is increasing, while the marginal cost of creating another summary, market map, competitor profile, scenario, or interview synthesis is falling. That means an organization can accumulate analytical artifacts faster than humans could ever maintain a traditional knowledge library.
Finding text becomes easier.
Knowing what deserves to be reused becomes harder.
Recent organizational-learning research is beginning to examine this problem at the system level. Figge, Anderson, and Lewis model organizations in which AI learning interacts with individual and collective learning, explicitly incorporating both learning and forgetting. Their work is computational rather than field evidence about corporate research functions, but its important conceptual move is to reject the idea that AI productivity can be understood solely at the individual- task level. AI becomes part of a broader learning system in which different stocks of organizational knowledge can reinforce, or undermine, one another over time.
This suggests a useful test for AI research infrastructure.
Do not ask only:
Can it generate a good answer?
Ask:
Does completing this inquiry leave the organization better prepared for the next related question?
An AI system that produces 100 excellent reports but leaves no durable structure connecting questions, assumptions, evidence, and later outcomes may increase individual productivity while contributing little to institutional learning.
In fact, it can accelerate research debt by creating more outputs that future teams must evaluate.
The important AI capability is therefore not merely generation or semantic retrieval. It is maintaining relationships across inquiries.
This question was asked because of that decision.
This claim rested on these sources.
This conclusion depended on these assumptions.
This experts disagreed for these reasons.
This uncertainty remained open.
This decision followed.
This later evidence changed the view.
That structure turns an archive into something closer to a learning system.
Reuse without context creates a different failure: stale certainty
There is an obvious danger in this argument.
Starting from prior research is not always better than starting over.
Markets change. Technologies mature. regulations evolve. Competitive structures shift. Management teams change. Expert knowledge ages. A framework that worked in one client, country, or business model can fail badly in another.
Institutional memory can become institutional inertia.
Lee and Van den Steen formalized part of this tradeoff in their research on knowledge-management systems. Their model finds that formal knowledge systems become particularly valuable when organizations repeatedly encounter similar issues, but also that recording and disseminating existing practices can under some conditions discourage experimentation. They describe a potential “competency trap” in which accessible prior know-how pushes people toward established approaches rather than continued exploration.
That is the strongest case against indiscriminate reuse.
And it changes what good research memory should look like.
A poor system says:
We researched this in 2024. Here is the answer.
A good system says:
We researched this in 2024 for this decision, under these conditions. These were the assumptions.
This evidence was current as of this date. These questions remained unresolved. These developments would invalidate or require refreshing the work.
The goal is not to preserve answers indefinitely.
It is to preserve enough context to know what can safely be inherited and what must be questioned again.
Extracted images (52):
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_1.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_10.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_11.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_12.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_13.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_14.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_15.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_16.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_17.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_18.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_19.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_2.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_20.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_21.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_22.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_23.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_24.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_25.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_26.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_27.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_28.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_29.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_2_table_1_v2.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_3.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_30.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_31.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_32.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_33.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_34.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_35.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_36.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_37.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_38.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_39.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_4.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_40.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_41.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_42.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_43.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_44.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_45.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_46.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_47.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_48.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_49.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_5.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_50.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_51.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_6.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_7.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_8.jpg`
- `parsed-documents://20260825-194715-378791/Enquire_AI_Blog_Collection.pdf/images/page_9.jpg`# Document parsed from: split_b.pdf
This is one reason explicit hypotheses matter. MIT CISR's research on organizational learning argues that teams learn more effectively when they unpack initiatives into underlying assumptions that can be tested separately. An explicit hypothesis makes it easier to determine what evidence has actually changed rather than simply labeling the eventual project a success or failure.
Research memory should do the same thing.
It should preserve not merely what the organization knew, but the conditions under which it believed it knew it.
Design for updating, not copying
A reusable research object therefore needs a lifecycle.
Imagine a new team encounters prior work on the market it is investigating.
The wrong workflow has two extremes.
At one extreme, the team ignores the old work and begins again.
At the other, it imports the old conclusions directly because “we already have research on this.”
A better workflow begins with triage.
What remains structurally reusable?
Definitions, historical analysis, primary-source collections, interview guides, causal frameworks, and descriptions of institutional structure may remain useful for years.
What has a shorter half-life?
Market size, pricing, competitor behavior, regulation, executive opinions, operating metrics, and expert observations may require frequent updating.
What was always uncertain?
The old research should expose its unresolved questions and low-confidence assumptions immediately instead of making the new team infer them from polished prose.
What has happened since?
New evidence should attach to the old question, allowing researchers to see whether the prior analysis has been confirmed, weakened, superseded, or rendered irrelevant.
This changes the mindset from reuse the answer to resume the inquiry.
That is a much higher standard for knowledge management.
It is also a more defensible one.
Diagnose research debt before buying another search tool
Heads of research and knowledge management can test how serious the problem is without building an elaborate measurement system.
Take a sample of recurring questions from the last year and ask the next researcher to reconstruct the organization's prior state of knowledge.
Measure five things.
1. Prior-work discovery rate
Can researchers reliably identify the relevant previous inquiries, or do they depend on someone remembering that the work exists?
2. Provenance recovery
Once an important prior conclusion is found, how easily can the researcher trace it back to the evidence that supported it?
A repository can score well on discovery and poorly on provenance.
3. Reasoning recovery
Can the researcher tell which assumptions drove the conclusion, which alternatives were rejected, and what uncertainties remained?
This is the difference between finding an output and recovering an inquiry.
4. Repeated-work rate
How much new research recreates evidence already collected inside the organization?
Not all repetition is waste. Sometimes deliberate replication is exactly what confidence requires. The diagnostic should distinguish purposeful retesting from accidental rediscovery.
5. Refreshability
Can the team identify which part of the prior work needs updating without redoing everything?
If every new question requires reopening the whole analysis, the organization has stored an artifact but not created a reusable research asset.
These measures should be interpreted together.
A team with a high reuse rate but no refresh discipline may simply be circulating stale analysis.
A team with low reuse may either have poor research memory or operate in an environment where every problem is genuinely novel.
Research debt is therefore contextual. The best candidates for structured memory are domains where questions recur while answers evolve.
Capital markets, policy monitoring, recurring diligence, sector research, competitive strategy, and professional-services delivery all fit that pattern unusually well.
Measure whether knowledge travels
The ultimate test is not how many documents are stored or how often an internal search tool is used. It is whether experience changes later work.
Organizational-learning research provides a useful conceptual standard. Reagans, Argote, and Brooks found that organizational experience and experience working together contributed separately to performance; Lewis, Lange, and Gillis found that knowledge structures could support transfer to related tasks under particular conditions.
For research organizations, the analogous question is:
Does inquiry N make inquiry N+1 materially better?
That could mean the second team reaches the frontier of existing knowledge faster.
It could mean it knows which question not to ask again.
It could discover a contradiction between new and old evidence sooner.
It could improve an interview guide because earlier respondents exposed a weak assumption.
It could identify a researcher or expert who already understands the issue.
Or it could recognize that the old conclusion no longer applies, and explain exactly why.
BCG's current M&A work describes closing this feedback loop explicitly: compare outcomes with earlier assumptions, identify which diligence findings proved predictive, understand where models diverged from reality, and feed those lessons into subsequent decisions.
That is a useful definition of institutional learning far beyond M&A.
The organization should not merely accumulate research.
Its next question should inherit the learning from the last one.
Where Enquire fits: make the inquiry itself persistent
This is also where Enquire's current product direction connects naturally to the problem.
Enquire describes its Research Center as a workspace for building and refining understanding as conditions change, and its current product materials explicitly describe an Evolving Research Context in which context is preserved across inquiries. The company also says organization-level Research Centers support collaborative synthesis and that insights can accumulate rather than requiring research to restart with each question.
Its current positioning combines that persistence with structured AI research and expert perspective, allowing prior questions, emerging disagreements, and new evidence to remain connected as research develops.
The useful application is not simply a better archive of Enquire reports.
It is to make the research context itself persistent.
A team investigating a sector can begin with what previous researchers established, see where expert perspectives diverged, identify assumptions still requiring scrutiny, add new external evidence, and update the inquiry rather than opening a clean page.
That design does not solve the stale-knowledge problem automatically. Dates, provenance, confidence, unresolved questions, ownership, and refresh conditions still have to be visible.
Nor should an old expert perspective become authoritative simply because it is preserved.
The advantage is narrower: maintaining enough continuity that new research can interrogate the old work instead of first reconstructing it.
Stop asking how much knowledge you store
The paradox of institutional knowledge is that many organizations possess more of it than they can effectively use.
Their servers contain the research.
Their people remember pieces of it.
Their search systems can increasingly locate it.
Yet the next project still begins with reconstruction because the relationships that make the knowledge reusable were discarded when the last project ended.
AI will make storing another answer nearly free.
That makes the distinction more important.
The organizations that develop durable research memory will not necessarily be those with the largest repositories, longest transcripts, or most capable enterprise search.
They will be the ones that preserve the smallest useful structure around each important inquiry:
- What were we trying to decide?
- What did we believe?
- What evidence supported it?
- Where did informed people disagree?
- How confident were we?
- What remained unresolved?
- What did we do?
- What evidence would make this research stale?
Preserve that chain, and old research becomes something a future team can challenge, update, and extend.
Lose it, and the organization acquires another document while quietly taking on more research debt.
The question for a research leader is therefore not, “Can our people find what we already produced?”
It is more demanding:
When this question returns in a year, what will the next team inherit besides our final answer?
Sources and further reading
- Yuqing Ren and Linda Argote, “Transactive Memory Systems 1985–2010,” Academy of Management Annals experts.umn.edu
- Kyle Lewis, Donald Lange, and Lynette Gillis, “Transactive Memory Systems, Learning, and Learning Transfer,” Organization Science pubsonline.informs.org
- Ray Reagans, Linda Argote, and Daria Brooks, “Individual Experience and Experience Working Together,” Management Science pubsonline.informs.org
- Linda Argote, Sunkee Lee, and Jisoo Park, “Organizational Learning Processes and Outcomes,” Management Science pubsonline.informs.org
- M. Lynne Markus, “Toward a Theory of Knowledge Reuse,” Journal of Management Information Systems jmis-web.org
- Deishin Lee and Eric Van den Steen, “Managing Know-How,” Management Science pubsonline.informs.org
- Jeanne W. Ross and Nils O. Fonstad, “Learn from Hypotheses, Not Failures,” MIT CISR cisr.mit.edu
- Jens Kengelbach et al., “AI Is Turning M&A into a High-Impact Learning Machine,” BCG bcg.com
- Patrick Figge, Edward Anderson, and Kyle Lewis, “AI-human learning systems,” Strategic Organization journals.sagepub.com
- Enquire, current product overview enquire.ai