Est.

Why Organizations Keep Repeating the Same Strategic Mistakes

Knowledge of past failures stays trapped within teams instead of spreading across the organization.

Publication desk · · 11 min read · Updated
Cover illustration for “Organizational Learning Failures After Repeated Strategic Mistakes”
Institutional Memory · September 23, 2026 · 11 min read · 2,419 words

Organizations repeat the same strategic mistakes because lessons from past failures never reach the people who need them. The problem is not bad hiring or careless teams. It is a broken system for passing knowledge from one group to the next.

Picture the same project, six months apart, two different teams, one identical mistake. The first team burned through budget chasing an approach that didn't work. The second team had no idea what the first team had learned. They chased the exact same approach and burned through budget the exact same way. Nobody was negligent. The debrief happened, notes got filed, and lessons got "captured." None of it mattered, because none of it reached the second team before they needed it.

Leadership usually reads that pattern as a people problem: wrong hires, sloppy discipline, a team that didn't do its homework. That diagnosis is almost always wrong. Learning did happen. It just happened inside one team's memory, one chat thread, or one document that got written once and never opened again. It never became something the whole organization could use. That gap — between what one team learned and what the organization as a whole knows — is where strategic mistakes go to repeat themselves. Closing that gap is a structural problem, not a hiring problem.

Why lessons fade before they become shared knowledge

Four things drive the decay, and they compound.

First, knowledge lives in people instead of written records. When the person who made a call leaves the company, or moves to a different team, the reasoning goes with them. Not the decision itself — that's usually recorded somewhere — but the why. Why this approach and not the other three that got discussed. What constraint made the obvious choice impossible. That context rarely survives a departure.

Second, project reviews grade outcomes instead of decision quality. A launch that worked gets stamped "good decision." A launch that flopped gets stamped "bad decision." That's a tempting shortcut. But it trains everyone to look right in hindsight rather than reason well under uncertainty. Those are different skills.

Third, the lessons that do get written down aren't where anyone would find them when needed. The document exists. It's just three folders deep in a drive nobody searches, filed under a project name nobody remembers six months later.

Fourth, leadership turnover resets the whole thing. New leaders arrive, look at old lessons, and decide they're "not relevant anymore." Often they don't check whether the underlying conditions actually changed. Researchers call this organizational amnesia. It tends to come back on a long cycle — roughly whenever enough people who lived through the original mistake have moved on.

When you put those four together, the pattern writes itself. New hires get handed tasks without the trade-off history behind them, so they repeat the trade-offs. The organization runs the same fix twice, holds the same debrief twice, and reacts with the same surprise twice. That is the system working exactly the way it is built to work.

What research on learning from failure actually shows

There's a comforting version of "fail forward" that says failure teaches more than success. That version has real research behind it. Madsen and Desai's study of orbital launch vehicle programs, published in the Academy of Management Journal, found that organizations learn more from failures than from successes. It also found that lessons from failure fade more slowly over time than lessons from success. Good news, as far as it goes.

It doesn't go all the way, though. The same research found that how much an organization can pull from a failure depends on its prior experience and on how big the failure was. A small stumble by a team with deep experience teaches differently than a big one on a team with none.

Then there's another complication. Research out of Carnegie Mellon and Clark University, published in the Strategic Management Journal, found that individual learning from failure follows an inverted curve. Performance improves as failures pile up — up to a point. Then it declines. One reason is that when failures keep coming, and some were clearly outside a person's control, people start blaming outside causes more broadly. That includes failures that actually were their fault. That short-circuits learning right when repeated failure is doing the most damage.

There's one more wrinkle for organizations doing product or innovation work. It may be harder to pull learning from failures that happen later in a project's lifecycle. Without a clear step where the organization actually changes its behavior based on what failed, the learning effort stalls. Similar mistakes come back around.

Where the three-step failure-learning cycle breaks down

Research on knowledge-intensive organizations describes failure-learning as three steps: recognition, sensemaking, and adaptation. Recognition means admitting the thing failed. Sensemaking is the debrief, where the team works out what happened and agrees on a shared account of the causes.

Most organizations get through both of those. The debrief happens. People argue about causes and eventually agree on a version of events. It feels productive, and it is productive, as far as it goes.

Adaptation is the third step, and it's where the whole cycle usually stalls. Adaptation means the lesson actually changes how the next decision gets made. A debrief that produces agreement but doesn't change the approval process, the checklist, or the criteria for the next similar call has produced sensemaking without adaptation. It's a complete, well-documented record of a team understanding its own mistake. And it's sitting in a folder, changing nothing downstream. That's the exact failure from the opening: the notes get filed, and filed is where they stay.

Why mixing up decision quality and outcome quality makes things worse

Decision quality and outcome quality are not the same thing. Treating them as the same is where most reviews go wrong.

Decision quality can only be judged using the information available when the decision was made. Outcome quality reflects everything that happened afterward, including things nobody could control. A good decision can produce a bad outcome. A bad decision can get lucky. If you reward outcomes, you don't build sharper decision-makers. You build people who avoid blame and stop writing down their uncertainty, because documented uncertainty is the first thing used against them later.

The fix changes the first question asked in a review. Instead of starting with "what went wrong," start with "what did we know when we made this call?" Write down the information available at the time, the options that got weighed, the reasoning behind the choice made, and the assumptions underneath it. Only after that context is laid out does the review turn to what new information should update those assumptions. Reversing the order turns the review into a hunt for someone to blame instead of a tool for getting sharper.

What a useful decision record needs to contain

Here's a fair test for whether a decision record — a written account of what was decided and why — is worth anything. Hand it to someone who wasn't in the room. Can that person check whether the conditions behind the original choice still hold? Can they reuse the reasoning or update it, without redoing the whole analysis from scratch?

A record that passes that test has five things in it. The decision that was made. The context at the time — meaning constraints, data, and the options actually considered. The assumptions driving the reasoning. The trade-offs that got rejected, and why. And the conditions that would mean it's worth revisiting the whole conclusion later.

Records fail that test constantly, and they fail in the same way. They capture the conclusion without the context, the outcome without the reasoning, the lesson without the conditions that made it relevant. "We decided not to expand into that market" tells the next team nothing. "We decided not to expand because unit economics required a 20% price premium the local competition wasn't going to allow, and that constraint might not hold anymore" tells them everything.

None of that matters if nobody can find the record when they need it. A lesson buried three folders deep is, in practice, no lesson at all. Being easy to find at the moment of decision is the whole point of documentation.

How separate systems block decision context from reaching the right people

Picture the systems tracking one person through an organization: an application history in one platform, training progress in another, support interactions in a third, performance records in a fourth. Each one holds a slice of that person's story. The reasoning behind a past decision about that person — the why behind a call someone made six months ago — often lives in none of them.

A decision record sitting in a project management tool doesn't reach the customer relationship system. A hard-won lesson in a shared drive doesn't surface inside the workflow where the next nearly identical decision is happening right now. New hires get logins to every one of these systems. What they get is the data. What they don't get is the judgment that shaped it, because that judgment was never attached to the record in a way any system could carry forward.

Connecting context across separate systems doesn't mean piling every record into one giant database and hoping meaning falls out. It means linking what specific data points actually mean for the same workflow, the same person, or the same decision. That way, the relationships between records are clear to whoever is looking.

Making past reasoning shape future decisions, not just sit on a shelf

There's a real difference between a lesson you can consult and a lesson that actually shapes what happens next. A lesson filed away for reference depends on someone remembering it exists. They have to remember where to look, and choose to look — right when they're under deadline pressure and drowning in volume. That's exactly when people don't go check.

Making past reasoning shape future decisions takes a few specific things.

You need decision records that tie together the evidence available, the reasoning applied, the human judgment used, any approval or override that happened, the action taken, and the outcome that followed. You need a separation between who can read a record and who can edit it, because those are different permissions with different accountability. You need named ownership of exceptions, so that when conditions drift from what made an old lesson true, a specific person decides whether the lesson still applies. And you need conditions for reuse built into the workflow itself, so the system checks whether circumstances have changed before letting a past lesson get applied blindly to a new call.

Handled this way, institutional memory — the organization's shared record of what was decided, why, and what happened — becomes a governance tool rather than just a document. The record of what got decided, why, by whom, under what constraints, and with what result becomes a real input into the next decision's approval process. That changes what leaders can rely on. A structure that surfaces context automatically replaces the need for individuals to remember and pass it on by word of mouth. Losing a person erases a relationship, but a structure that captures reasoning in records — rather than in people's heads — can survive it.

Diagram: Where the Failure-Learning Cycle Breaks Down. Visualizes: Illustrate the three-phase failure-learning cycle — Recognition, Sensemaking, Adaptation — as a linear flow, making viscerally clear that most organizations complete the first two…

How AI-connected systems can close the gap between documented lessons and live decisions

The core problem isn't a shortage of lessons. Organizations generate plenty of lessons. The problem is that those lessons aren't accessible inside the exact context where the next decision is being made, at the exact moment someone needs them.

AI-connected workflow systems — software that links data and past decisions across separate tools and surfaces the right piece automatically — address that gap directly. They pull together data scattered across separate systems: business records, documents, past decisions, and past human corrections. Then they surface the relevant piece right as a new decision is forming. That's a different job than search. Search requires you to know a lesson exists and go looking. Surfacing means the system recognizes the workflow or situation in front of it and brings the past reasoning along automatically.

A context graph — a map connecting what different data points mean for the same workflow or the same person across separate systems — is the structure that makes this possible. It lets a new team tackling a familiar problem see the reasoning that came before, without needing to know that reasoning exists or where anyone filed it. Decision traceability is what comes out the other end. Every important decision leaves behind a record of what information was available, what reasoning was applied, whether a human reviewed or overrode it, and what actually happened. That record becomes the raw material the next similar decision draws on.

Start with one workflow, one owner, and a measurable baseline before scaling

Organizations that actually turn failure into shared knowledge — rather than just a folder of debrief notes — don't try to fix everything at once. They start in one bounded workflow, where the baseline is clear, the outcome is measurable, and one person is accountable for both.

That single workflow needs a few things before it can generate real institutional learning. A named owner who can say what the current process does, what a better version looks like, and how "better" gets measured. An agreed baseline, captured before anything changes: what the process produces today, at what cost, with what error rate or how often exceptions come up. A defined measurement window, with one named person responsible for reading the results against that baseline when the window closes. And decision records running from day one, so the reasoning used during the pilot becomes the organization's first real deposit into shared memory — not an afterthought added later.

A single workflow with a clear owner and a measurable outcome is the right place to start. It lets the organization prove the structure works before anyone has to depend on it for something bigger. It also produces the first genuinely traceable lesson — the one future decisions can actually pull from.

From there, the effect compounds. Every new workflow adds another piece to the context map. Each decision record extends the running account of what was known and what happened as a result. Each measured outcome gives the organization evidence it can use — rather than another well-meant lesson that got documented once, filed away, and forgotten by the time the next team needed it most.

Sources

  1. Tepper School Study Finds Not All Failures Lead to Learning
  2. Decomposition of double-loop failure risk in post-innovation failure phase - ScienceDirect
  3. Strategies for Learning from Failure
  4. Failing to Learn? The Effects of Failure and Success on Organizational Learning in the Global Orbital Launch Vehicle Industry | Academy of Management Journal
  5. The Organizational Memory Problem
  6. pmctraining.com
  7. Embracing the “fail fast and learn fast” mindset: conceptualizing learning from failure in knowledge-intensive SMEs | Small Business Economics | Springer Nature Link

More in Institutional Memory