What is a captured decision worth?

The foundational theory of organizational memory. Walsh and Ungson define it as information from an organization’s history that can be brought to bear on present decisions, and locate it in five internal “retention bins” — individuals, culture, transformations (procedures and systems), structures, and physical ecology — plus external archives such as former employees and regulators. Two implications matter for this page. First, memory is distributed and retrieval is unreliable: an organization can possess knowledge it cannot find at the moment of deciding. Second, the bin that holds decision rationale is disproportionately the individual, which means rationale exits with people. The paper is conceptual rather than quantitative; we cite it for the framework that makes “where does the reasoning live?” an answerable question, not for effect sizes.

A study of weekly productivity data from 36 pizza franchise stores over roughly a year and a half. Unit costs fall with cumulative experience — classic learning by doing — but the arresting result is depreciation: operating knowledge decays at an estimated rate of roughly 17 percent per week, implying a half-life of only a few weeks unless knowledge is continually renewed through production. The study also finds knowledge transfers between stores owned by the same franchisee but not between different franchisees’ stores, evidence that knowledge moves along relationships and shared routines rather than mere proximity. This is the sharpest available number for how fast un-externalized organizational knowledge evaporates, and it anchors the organizational forgetting curve drawn above.

The classic study of organizational forgetting, using production data from the WWII Liberty ship program. Knowledge acquired through production depreciated rapidly: the estimates imply that of the knowledge stock available at the start of a year, only about 3.2 percent would remain a year later if not replenished by continued production. The paper shows that the conventional cumulative-output learning curve “significantly overstates the persistence of learning.” Shipyards that began production later started at higher productivity — evidence of initial knowledge transfer — but ongoing learning did not transfer between organizations once production was underway. Together with Darr et al., it establishes that organizational knowledge is a rapidly depreciating asset, not a durable archive.

The landmark Science review of organizational learning curves. Productivity reliably improves as organizations accumulate production experience, and the pattern is documented across aircraft, ships, refineries, power plants, and more. The finding this page leans on is the variance: organizations differ enormously in how fast they learn, and the authors attribute that variation to organizational forgetting, employee turnover, and incomplete transfer of knowledge across products and organizations — that is, to how well an organization retains and moves what it learns. If learning rates were fixed, memory infrastructure would be decoration. Because they vary with retention and transfer, the mechanisms Record targets are precisely the ones this literature identifies as the difference between fast and slow learners.

An analysis of 48 months of turnover data across 268 stores of a large U.S. retail chain. On average, higher employee turnover reduced store performance, measured by profit margin and customer service. The key result is the moderator: in stores with high process conformance — where managers enforce standardized, documented operating procedures — turnover’s damage was substantially mitigated, while low-conformance stores bore the full cost. This is direct field evidence for the central economic claim of this page: externalizing how work is done buffers an organization against the knowledge that walks out the door. Codification converts a fragile asset held in heads into a durable one held by the organization.

The original “hidden profile” experiment. Four-person groups chose among three candidates; the objectively best one could only be identified by pooling information scattered across members. With full shared information, 83 percent of groups chose correctly. When the decisive information was distributed — each member holding a unique piece — only 18 percent did. Discussion gravitated to what everyone already knew and to information supporting prior preferences, so talking together perpetuated rather than corrected the distortion. The relevance here: decision quality depends on assembling dispersed evidence, and groups demonstrably fail to do this on their own. A system that gathers and attaches evidence before the call is compensating for a precisely documented failure mode of human deliberation.

Two surveys and eleven interviews of Microsoft developers, asking where engineering time actually goes. “Understanding the rationale behind a piece of code” ranked as the single most serious problem — 66 percent of developers agreed, ahead of understanding the code itself. Eighty-two percent agreed it takes substantial effort to understand why code is implemented the way it is, and 73 percent struggled to tell whether code was a temporary workaround. The mechanism is the one this page is about: rationale lived almost entirely in individuals’ memories, recovered by code archaeology and interrupting teammates, because design documents were inadequate and stale. It is the best direct measurement we have of professionals paying daily interest on undocumented “why.”

The founding experiment on escalation of commitment. Business students allocated R&D funds in the role of a corporate executive; the study manipulated whether they were personally responsible for an earlier investment and whether it had failed. People committed the most additional resources to a previously chosen course of action precisely when they were personally responsible for its negative consequences — throwing good money after bad to justify the past. The economic link to decision capture: escalation feeds on ambiguity about what was originally decided, why, and under what success criteria. A record with pre-stated assumptions and reversal conditions is the cheapest known institutional defense against self-justifying reinvestment.

An analysis of 271 observations of 122 best-practice transfers inside eight companies, asking why firms fail to replicate their own successes. Contrary to the conventional wisdom blaming motivation — hoarding, not-invented-here — the biggest barriers were knowledge-related: the recipient’s limited absorptive capacity, causal ambiguity, and arduous relationships between source and recipient. Causal ambiguity is the direct hook for this page: when nobody recorded why a practice works, even a willing organization cannot copy it from itself. The finding reframes knowledge transfer as an information problem that capture infrastructure can actually address, rather than an incentive problem that it cannot.

The HBR article that preceded the 2021 book Noise. In the flagship “noise audit,” experienced underwriters at an insurance company each priced identical realistic cases. Executives predicted their judgments would differ by about 10 percent; the median difference was roughly 50 percent — about five times expectations — with similar results for claims adjusters. The authors distinguish noise (random inconsistency) from bias and argue it imposes large, invisible costs because identical inputs produce different outputs depending on who decides and when. Their prescription is decision hygiene: structured, documented reasoning and consistent decision records. The audit result is among the strongest evidence that undocumented judgment is not merely unexplained — it is measurably inconsistent.

A rigorous modern replication of Ebbinghaus’s 1880s forgetting experiment: 70 hours of learning and relearning nonsense-syllable lists at retention intervals from 20 minutes to 31 days. The replication closely matches the original curve, confirming steep early forgetting: in Ebbinghaus’s data, savings fall from roughly 58 percent after 20 minutes to about a third after one day and around 21 percent after 31 days; the replication’s 31-day figure was lower still. We cite it to make a comparative point drawn in the chart above: individual memory decays on a curve science has known for 140 years — and the organizational forgetting literature finds that organizations, without capture, forget on a curve that is steeper.

McKinsey’s survey of more than 1,200 managers at global companies, published alongside its “Decision making in the age of urgency” research. Sixty-one percent of respondents said that at least half the time they spend on decision making is used ineffectively, and fewer than half said decisions are timely. McKinsey’s stated arithmetic: for a typical Fortune 500 company, ineffective decision time is equivalent to about 530,000 days of managers’ time per year, roughly $250 million in annual wages. The companion survey found only 57 percent of decisions were judged high quality. These are self-reported survey measures from a consultancy with an interest in the topic — a caveat we keep — but they are the most direct available sizing of decision-process waste inside large firms.

Bain’s decade-long research program covering more than 1,000 companies, summarized in the book Decide & Deliver and the HBR article “The Decision-Driven Organization.” Bain reports a clear correlation — at a minimum 95 percent statistical confidence level — between decision effectiveness (quality, speed, yield, and effort of key decisions) and business performance, and finds top-quintile decision-effectiveness companies deliver average total shareholder returns nearly six percentage points higher than others. One precision note we carry deliberately: the famous “95%” is a confidence level on the correlation, not a correlation coefficient of 0.95, though looser phrasings circulate. The direction is what matters here: how organizations decide is tightly associated with what they earn.

MGI’s sizing of how interaction workers actually spend the workweek, based on task-level analysis: about 28 percent of the week managing email and roughly 19 percent searching for and gathering information — finding internal knowledge or tracking down colleagues who have it. Nearly half the week, in other words, goes to communication overhead and hunting for things the organization already knows. MGI estimated that a searchable record of internal knowledge could cut information-search time by as much as 35 percent, and that full adoption of social technologies could raise interaction-worker productivity 20–25 percent. The 35 percent search-reduction estimate is the closest thing in the literature to a direct prior for what decision-capture infrastructure should be worth, and our central scenario stays deliberately below it.

A YouGov-conducted survey of over 1,000 U.S. employees, modeled to enterprise scale. Headline findings: U.S. knowledge workers waste an average of 5.3 hours every week either waiting for vital information from colleagues or recreating institutional knowledge that already exists; 42 percent of institutional knowledge is unique to the individual who holds it; and inefficient knowledge sharing costs the average large U.S. business an estimated $47 million per year in lost productivity — about $2.7 million annually per 1,000 employees. New hires spend roughly 28 inefficient hours per month during onboarding for the same reason. Vendor-commissioned research, so we treat the dollar models as indicative rather than precise — but the time-use figures are consistent with MGI’s independent task-level analysis.

A survey of 13,123 knowledge workers across seven markets. Workers reported spending about 60 percent of their time on “work about work” — searching for information, switching between apps, chasing status, duplicating effort — rather than the skilled work they were hired to do; at organizations above 5,000 employees the figure rises to 63 percent. Unnecessary meetings cost respondents 157 hours per year, and new hires spent nearly twice as much time on duplicated work as tenured colleagues — a clean signature of missing organizational memory, since duplication is what rediscovery looks like from the inside. Vendor-commissioned but large-sample, multi-country, and directionally consistent with the McKinsey and Microsoft time-use data cited alongside it.

A survey of 31,000 people in 31 countries combined with Microsoft 365 telemetry. The telemetry shows the average employee spends 57 percent of working time communicating — meetings, email, chat — and only 43 percent creating. Respondents ranked inefficient meetings as the single biggest disruptor of productivity. The failure modes they named are the ones decision capture addresses directly: 56 percent said it is hard to summarize what happened in a meeting, and 55 percent said next steps at the end are unclear. That is, the expensive part of communication is not the talking but the failure to convert it into a durable, retrievable outcome — which is precisely the conversion a decision record performs.

Deloitte’s survey of roughly 9,000 organizations across 119 countries. Seventy-five percent said creating and preserving knowledge across evolving workforces is important or very important to their success over the following 12–18 months — yet only 9 percent said they were “very ready” to address it, one of the largest importance-readiness gaps of any trend measured that year. Fifty-two percent cited workforce movement — turnover, restructuring, alternative work arrangements — as the force pushing them toward knowledge-management strategies, and most respondents still defined knowledge management as documenting and disseminating rather than connecting knowledge to action. We cite it as the cleanest measurement of the gap between what organizations say memory is worth and what they have built.

The paper that made management measurable. Using a double-blind interview instrument scoring practices in monitoring, targets, and incentives across 732 medium-sized manufacturers in the U.S., U.K., France, and Germany, Bloom and Van Reenen show that management scores are strongly associated with firm productivity, profitability, market value, and survival. Every country exhibits a long tail of very badly managed firms, and poor management persists where competition is weak. The relevance to this page: the practices the instrument scores are substantially practices of recording and reviewing — tracking performance, documenting processes, following up on commitments. Organizational memory is not a soft amenity in this literature; it is a measured component of the management input that predicts productivity.

The randomized controlled trial behind the +17 percent bar in the chart above. Large Indian textile plants were randomly assigned free consulting on 38 management practices, most of which involve systematic recording and monitoring: quality-defect tracking, inventory records, standardized operating procedures. Treated plants raised productivity by 17 percent in the first year through improved quality and efficiency and lower inventory, and within three years opened more production plants. The authors conclude that informational barriers — firms not knowing or not believing the value of these practices — were the main reason profitable practices had not already been adopted. It is the strongest causal evidence available that instituting recording practices, by itself, produces double-digit productivity gains.

An asset-pricing measurement of organizational knowledge. The authors capitalize firms’ cumulative SG&A spending as a proxy for organization capital — the intangible capital embodied in key employees, processes, and know-how — and find that firms with high organization capital earn average stock returns 4.6 percent per year higher than low-organization-capital firms. Their model explains why: organization capital is partly portable in the heads of key talent, so shareholders bear the risk that it walks out the door, and markets price that fragility. For this page the paper does two jobs: it establishes that un-externalized organizational knowledge is a large asset class, and that its residence in individuals rather than systems is a priced risk, not a neutral storage choice.

A study of defect data from a major automaker’s assembly plant producing about 190,000 cars in a year. Average defects per car fell from 75 in the first week of production to 25 within eight weeks, with each tenfold increase in cumulative output roughly halving defect rates — quality improvements conservatively worth about $106 in added profit per car. The finding that matters most here: the accumulated learning was embodied in the organization’s routines and processes rather than in individual workers — high absenteeism and turnover did not degrade quality once processes had captured the knowledge. It is the mirror image of the forgetting studies: when learning is institutionalized, the organization keeps it even as the people change.

The canonical demonstration that lightweight structured documentation changes outcomes. A 19-item WHO surgical safety checklist was introduced in eight hospitals across eight countries, with 3,733 patients studied before implementation and 3,955 after. In-hospital deaths fell from 1.5 percent to 0.8 percent — a roughly 47 percent relative reduction — and inpatient complications fell from 11.0 to 7.0 percent, with effects holding in both high- and lower-income settings. A checklist is the smallest possible decision record: it externalizes what must be confirmed, by whom, before an irreversible action. If seconds of structure per operation can move mortality, the burden of proof shifts to anyone claiming that structure around consequential business decisions is not worth its friction.

The Stanford/University of Washington experiment behind the “3× faster” bar above. Comparing deep-learning speech recognition against the built-in smartphone keyboard, speech input was 3.0 times faster than typing in English and 2.8 times faster in Mandarin; after error correction, speech error rates were also lower than keyboard entry in both languages. This is the friction number that makes the 20 percent capture premise mechanically plausible: the historical failure of decision logs was an authorship tax paid in typing time, and speech collapses that tax to under a minute for a typical rationale. Capture at the moment of judgment becomes cheaper than the first interruption caused by its absence.

A staggered rollout of a generative-AI assistant to 5,172 customer-support agents. Access raised issues resolved per hour by 15 percent on average — and by about 30 percent for the least experienced agents — while improving customer sentiment and employee retention; the most experienced agents gained little. The mechanism, as the authors document, is organizational memory: the assistant works by capturing and disseminating the tacit conversational patterns of top performers to everyone else. It is the most rigorous evidence to date for the last link in this page’s chain — that once organizational knowledge is captured in machine-usable form, AI systems can distribute it at near-zero marginal cost, with the gains concentrated among the people who had the least access to it before.