From 50 Spreadsheets to One Source of Truth: Building Institutional Memory

The scene: Family office, 15 years of private fund investing, Tuesday afternoon.

The question: "Why did we invest in Fund XYZ back in 2018? What were our assumptions?"

The answer search:

• Check Excel file: Portfolio Model 2018 v3 FINAL.xlsx
• No, try: Portfolio Model 2018 v3 FINAL REVISED.xlsx
• Maybe email thread? Search for "Fund XYZ"
• Old IC memo? Where did we store those?
• Ask Sarah (who joined in 2020, wasn't here for that decision)
• Answer: Lost

This happens constantly in family offices.

When decisions made today affect wealth that will pass to grandchildren, institutional memory isn't a nice-to-have—it's your competitive advantage.

---

The Spreadsheet Sprawl Problem

A Typical Family Office After 10 Years

What you have:

• 50+ Excel files with various names:
  • Portfolio Model 2015.xlsx
  • Portfolio Model 2016 v2.xlsx
  • Portfolio Model 2017 FINAL.xlsx
  • Portfolio Model 2017 FINAL REVISED.xlsx
  • Portfolio Model Q3 2018 - Sarah.xlsx
  • Portfolio Model Q3 2018 - John.xlsx
  • Portfolio Scenarios 2020 COVID.xlsx
  • NEW Portfolio Model 2022.xlsx

What's in each:

• Different assumptions (which were current at the time)
• Different fund lists (portfolios evolve)
• Different formulas (each person builds differently)
• Different formats (no consistency)

What's lost:

• Why you made specific decisions
• What you expected at the time (vs what actually happened)
• How GPs performed vs their promises
• Lessons learned from past commitments

---

The Knowledge Transfer Problem

Scenario 1: Analyst Leaves

Sarah managed the portfolio for 5 years. She built the Excel models. She knew:

• Which assumptions were conservative vs aggressive
• Why certain funds got larger allocations
• Historical context for each commitment decision
• Workarounds for model quirks

Sarah leaves.

John takes over. The Excel files still exist, but:

• Formulas are opaque (no documentation)
• Assumptions are buried in cells (no explanation)
• Decision rationale is lost (not in the spreadsheet)
• Institutional knowledge walked out the door

Cost: 6-12 months for John to rebuild mental model. Decisions are worse during this period.

---

Scenario 2: Multi-Generational Transition

Family office manages $200M across 3 generations:

• Generation 1 (founders): Made initial commitments 1995-2010
• Generation 2 (current): Managing portfolio 2010-2025
• Generation 3 (next): Will take over 2025-2040

Questions Generation 3 needs to answer:

• "Which GP relationships delivered consistently?"
• "What strategies worked in different market cycles?"
• "Where did we make mistakes and why?"
• "How should we adjust for changing markets?"

Where are the answers?

• Generation 1's decisions: Documented in emails, old memos, verbal lore
• Generation 2's decisions: Excel files (50+ versions)
• No systematic institutional memory

Cost: Each generation re-learns lessons. Mistakes repeat. Relationships are lost.

---

What Institutional Memory Should Capture

1. Decision Context

When you commit $5M to a fund, record:

• Why this fund? (What made it compelling)
• What were we assuming? (Growth rates, exit timing, market conditions)
• What was our thesis? (Why we think this will work)
• What alternatives did we consider?
• What was the opportunity cost?

Years later, you need this to:

• Evaluate GP performance (vs what they promised)
• Learn from the decision (what worked, what didn't)
• Decide on re-ups (do we still believe the thesis?)

---

2. Assumptions at Decision Time

Example: 2018 Commitment

You modeled:

• Target IRR: 20%
• Deployment: 4 years
• Hold period: 6 years
• Exit timing: Years 5-8
• Market assumption: 10% equity growth

Reality (2024 - 6 years later):

• Actual IRR: 14% (underperformed)
• Deployment: 6 years (slower than expected)
• Still holding (exits delayed 2+ years)
• Market: 15% growth (tailwind, fund still underperformed)

Learning:

• GP deploys slower than promised (pattern across multiple funds)
• Exit timing optimistic by 2-3 years
• Even in good market, this GP underdelivered
• Decision: Don't re-up

But you need the 2018 assumptions to make this comparison. If they're lost in Portfolio 2018 v3 FINAL.xlsx on an old laptop, you can't learn.

---

3. Variance Analysis

Compare forecast vs actual:

Why variance analysis matters:

• Identifies patterns (this GP always deploys slow)
• Informs future modeling (adjust assumptions)
• Improves decision quality over time

But you need both forecast AND actual in one place.

---

4. Manager Track Record

Across multiple funds with same GP:

Pattern: GP consistently:

• Over-promises IRR by 5-7%
• Delivers but below expectations
• Deployment speed varies (3-6 years)

Decision impact:

• Maybe commit to Fund IV but model 15% IRR (not 20%)
• Expect 5-year deployment (not 3)
• Adjust allocation accordingly

This is institutional memory in action: Learning from 10 years of data to make better decisions.

---

The Single Source of Truth

What if instead of 50 Excel files, you had one system where:

• All funds in one place (current + historical)
• All assumptions documented and dated
• All decisions traceable with rationale
• All actuals captured from GP reports
• All forecasts preserved (can compare to actual later)
• All analysis accessible (not locked in someone's head)

This is what institutional memory looks like.

---

Example: Evaluating a Re-Up Decision (2025)

GP pitches Fund IV:

• "Target 20% IRR"
• "Deploy $50M over 3 years"
• "Start distributing Year 5"

You ask Nagare Portfolio Companion:

"How did this GP perform on previous funds vs their promises?"

AI responds with analysis:

Fund I (2015 vintage):
  Promised: 25% IRR, deployed in 3 years
  Delivered: 18% IRR, deployed in 3 years
  Variance: -7% IRR (missed target but deployment on schedule)

Fund II (2018 vintage):
  Promised: 22% IRR, deployed in 4 years
  Delivered: 14% IRR (ongoing), deployed in 6 years
  Variance: -8% IRR, +2 years slow deployment

Pattern:
  - Consistently over-promises IRR by 6-8%
  - Deployment timing varies (50% of funds slow)
  - Delivers positive returns but below expectations

Recommendation:
  - If committing to Fund IV, model 15% IRR (not 20%)
  - Assume 4-5 year deployment (not 3)
  - Adjust allocation expectations accordingly

This took 10 seconds.

Manually: Would take 2-3 hours digging through old files, emails, IC memos.

More importantly: You COULDN'T do this if the data is scattered across 50 Excel files and old emails.

---

What Changes with Institutional Memory

1. Better Manager Selection

Without memory:

• GP pitches Fund IV
• You vaguely remember their previous funds did "ok"
• Can't quantify performance vs promises
• Decision is gut feel

With memory:

• Query historical performance
• Compare to promises
• Identify patterns (slow deployment, conservative exits)
• Decision is data-driven

---

2. Learning from Mistakes

Without memory:

• 2019: Committed to over-valued late-stage VC fund → Lost money
• 2024: Similar pitch from different GP
• You don't remember the 2019 mistake
• You repeat the mistake

With memory:

• Portfolio Companion flags similarity: "This looks like Fund ABC from 2019"
• Shows what went wrong: Overpaid for growth that didn't materialize
• You avoid repeating the mistake

---

3. Continuous Improvement

Your portfolio strategy should improve over time:

Year 1-3: Make commitments, learn what works Year 4-6: Refine based on early data Year 7-10: Mature strategy based on full cycle Year 10+: Sophisticated, data-driven approach

But this only works if you CAPTURE and ANALYZE the learnings.

Without institutional memory:

• Each year is independent
• No systematic learning
• Strategy doesn't improve

With institutional memory:

• Track what works (allocate more)
• Track what doesn't (allocate less)
• Refine assumptions based on variance analysis
• Strategy gets better every year

---

4. Multi-Generational Continuity

Family office planning is 50+ years:

Generation 1 (1995-2025):

• Built initial portfolio
• Learned which strategies work
• Established GP relationships

Generation 2 (2025-2055):

• Needs to learn from G1's decisions
• Continue what worked
• Avoid what didn't
• But only if the knowledge transfers

With Excel files:

• Knowledge is tribal (oral history)
• Files are scattered and incomprehensible
• Each generation starts over

With institutional memory:

• All decisions documented
• All learnings captured
• All relationships preserved
• Knowledge transfers seamlessly

---

How Nagare Builds Institutional Memory

1. Single Source of Truth

Everything in one place:

• All funds (active + exited + historical)
• All transactions (capital calls, distributions, NAV updates)
• All assumptions (forecasts, scenarios, what-ifs)
• All calculations (preserved via snapshots)

No more:

• "Which Excel file has the 2019 data?"
• "Did we update the model after that exit?"
• "Where did we document our assumptions?"

---

2. Snapshots: "What We Knew When"

Problem: 2019 forecast vs 2024 actual comparison is impossible if you've overwritten the 2019 forecast.

Solution: Calculation Snapshots

Every time you run a forecast, Nagare can snapshot:

• All assumptions at that moment
• All projections generated
• All market data used
• Time-stamped and preserved

Example:

Snapshot: "Q3 2019 IC Meeting - Fund XYZ Commitment Decision"
  Date: 2019-09-15
  Assumptions: 20% IRR, 4-year deployment, 2.5x TVPI
  Projections: Capital calls Q4 2019 - Q3 2023
  Market data: S&P at 2,950 (pre-COVID)

Years later (2024):

• Pull up the 2019 snapshot
• Compare to what actually happened
• Analyze variance
• Learn from the difference

This is time-travel for your assumptions.

---

3. Portfolio Companion Remembers

Ask questions about history:

"Show me all funds where we over-estimated IRR by more than 5%"

AI searches:

• All historical forecasts
• All actual results
• Calculates variance
• Generates report

"What assumptions did we make for Fund XYZ in 2019?"

AI retrieves:

• 2019 snapshot
• Shows assumptions, projections, rationale
• Compares to what actually happened

"Which GPs consistently deliver on their promises?"

AI analyzes:

• Promised IRR vs delivered IRR (all funds, all GPs)
• Ranks by reliability
• Shows you who to trust

This would take DAYS to compile manually from 50 Excel files. Portfolio Companion does it in seconds.

---

4. Variance Attribution

Not just "we were wrong"—WHY we were wrong:

2019 Forecast:

• IRR: 20%
• Deployment: 4 years
• TVPI: 2.5x

2024 Actual:

• IRR: 14%
• Deployment: 6 years
• TVPI: 1.9x

Variance attribution:

• Deployment variance: +2 years slow → Impact: -3% IRR
• Exit timing variance: +1.5 years delayed → Impact: -2% IRR
• Exit multiples: 15% lower than forecast → Impact: -1% IRR
• Total: -6% IRR variance

Insight: Deployment speed was the biggest issue (3% of 6% variance).

Learning: For this GP, model 5-6 year deployment going forward (not 4 years).

This analysis requires having BOTH forecast and actual. Excel files get overwritten. Institutional memory preserves both.

---

Real Impact: The Compounding Effect

Year 1: Make 5 commitments with assumptions Year 3: 2 funds underperforming, 3 on track Year 5: Analyze variance, identify patterns Year 7: Next vintage—model with learned assumptions Year 10: Portfolio performs 15-20% better because assumptions are refined

Without institutional memory:

• Each decision uses same assumptions
• You don't learn from underperformance
• Mistakes repeat

With institutional memory:

• Assumptions improve based on data
• Avoid repeating mistakes
• Compound learning over decades

On a $200M portfolio, 15-20$30-40M over 10 years.

That's the value of institutional memory.

---

The Excel Death Spiral

How Excel files lead to knowledge loss:

Phase 1: Proliferation (Years 1-3)

• Each quarter, you update the model
• "Save as" → New version
• Eventually: 10+ versions per year
• 30 files after 3 years

Phase 2: Divergence (Years 4-7)

• Two analysts working on models
• Each makes their own copy
• Models diverge (different assumptions, formats)
• "Which is the right one?"
• 60+ files, no canonical source

Phase 3: Abandonment (Years 8-10)

• Models become too complex to maintain
• Formulas break when adding funds
• People stop updating older versions
• Give up and start new model from scratch

Phase 4: Archaeology (Years 10+)

• Need historical data for decision
• Have to dig through 100+ old Excel files
• Files don't open (Excel version compatibility)
• Formulas are broken
• Institutional memory is effectively lost

This is the Excel death spiral.

It's not a problem with Excel the tool—it's a problem with Excel as a database.

---

Single Source of Truth Design

What a proper system needs:

1. Unified Data Model

All funds in one schema:

• Fund metadata (name, vintage, strategy, GP)
• Capital structure (commitment, calls, distributions)
• Performance metrics (NAV, IRR, TVPI, DPI)
• Portfolio companies (holdings, valuations, exits)

No more:

• Fund A in Sheet1, Fund B in Sheet2 with different column structure
• Copy-paste errors when moving data
• "Did we update this everywhere?"

---

2. Time-Series Everything

Every data point has a timestamp:

• NAV as of 2024-09-30
• Forecast created on 2019-09-15
• Assumption changed on 2022-03-10

Why:

• Can ask "What was NAV in Q3 2023?" → Instant answer
• Can ask "What did we forecast in 2019?" → Snapshot preserved
• Can compare forecast-vs-actual over time

Excel: Overwrites old data. History is lost. Institutional memory: Preserves all versions. History is queryable.

---

3. Versioned Assumptions

Your assumptions evolve:

2019 Assumptions:

• Growth equity deploys in 3-4 years
• Exits in years 6-8
• IRR 18-22%

2024 Refined Assumptions (after 5 years of data):

• Growth equity deploys in 4-6 years (slower than expected)
• Exits in years 7-10 (delayed)
• IRR 15-18% (more conservative)

System should:

• Preserve 2019 assumptions (for variance analysis)
• Use 2024 assumptions (for new forecasts)
• Show evolution (how your thinking changed)

Portfolio Companion can answer: "How have our deployment assumptions changed over time?"

---

4. Decision Traceability

Every commitment should have:

• IC memo (why we invested)
• Forecast at decision time (what we expected)
• Market conditions (context)
• Alternatives considered (opportunity cost)
• Vote/consensus (who supported, who had concerns)

Years later:

• Pull up the decision
• See full context
• Compare to actual outcomes
• Learn from the decision

---

Implementation: How to Build This

What You Need:

1. Centralized database (not scattered Excel files)

• All funds in structured format
• Time-series transactions and NAV
• Versioned forecasts

2. Snapshot capability

• Preserve assumptions at decision time
• Compare forecast vs actual later
• Track how assumptions evolve

3. Query/analysis tools

• Answer questions about history
• Compare GPs, vintages, strategies
• Identify patterns across years

4. Documentation capture

• IC memos linked to decisions
• GP correspondence
• Rationale preserved

---

Nagare Approach:

Single Source of Truth:

• All funds in PostgreSQL database (not Excel files)
• Transaction history preserved (never overwritten)
• Calculation snapshots (time-travel to any past forecast)

Query with Portfolio Companion:

• "Show me all 2019 commitments and how they performed"
• "Which GPs delivered on their IRR promises?"
• "How have our growth equity assumptions changed?"

Variance Analysis Built-In:

• Compare forecast to actual automatically
• Attribution (why variance occurred)
• Pattern detection across funds

---

The 50-Year View

Private fund investing is multi-generational:

Your decisions today:

• Commit to 2025 vintage funds
• Model 10-year horizon (2025-2035)
• But the capital compounds for 50+ years

Your grandchildren (2075):

• Inherit the portfolio
• Need to evaluate 2040-2050 vintage re-ups
• Should learn from your 2025-2045 decisions

If institutional memory exists:

• "Grandpa committed to GP XYZ in 2025, they delivered"
• "Strategies that worked: Growth equity in tech, credit in healthcare"
• "Mistakes to avoid: Over-paying for late-stage VC in frothy markets"
• 50 years of compounded learning

If institutional memory is lost:

• Start from scratch every generation
• Repeat mistakes
• Lose GP relationships
• Wealth doesn't compound as well

---

Institutional Memory as Competitive Advantage

Why some family offices perform better than others:

Hypothesis: It's not just capital size or access to GPs.

It's institutional memory:

• They've invested through 3+ market cycles
• They've learned which strategies work
• They've identified reliable GPs
• They've refined assumptions based on decades of data
• They make better decisions because they remember past decisions

New family office (no memory):

• Guesses at assumptions
• Believes GP promises at face value
• No pattern recognition
• Learns slowly (starting from zero)

Mature family office (strong memory):

• Calibrated assumptions (refined over years)
• Evaluates GP claims against historical delivery
• Recognizes patterns across market cycles
• Learns quickly (builds on past learnings)

Over 20 years, the difference compounds to 20-30% better performance.

Institutional memory is your competitive moat.

---

How to Start Building

If you're in spreadsheet sprawl now:

Step 1: Consolidate (1-2 weeks)

• Import all current funds into Nagare
• Load historical transactions (capital calls, distributions)
• Link to GP reports and IC memos

Step 2: Snapshot Current State (1 day)

• Create baseline snapshot (where you are today)
• Document current assumptions
• Preserve current forecasts

This becomes your starting point.

Step 3: Go Forward with Memory (Ongoing)

• All new decisions captured
• All new data flows into central system
• All assumptions documented
• Institutional memory begins building

Step 4: Backfill History (Optional, 2-4 weeks)

• Go back through old Excel files
• Extract key decisions and assumptions
• Import historical snapshots
• Recover past institutional memory

---

Summary

50 Excel files across 10 years:

• Knowledge is scattered
• Context is lost
• Learning doesn't compound
• Institutional memory is weak

Single source of truth:

• All data in one place
• All decisions traceable
• All assumptions preserved
• Institutional memory is strong

Impact:

• Better manager selection (learn from track records)
• Avoid repeating mistakes (pattern recognition)
• Continuous improvement (refine assumptions)
• Multi-generational continuity (knowledge transfers)

On $200M portfolio over 20 years:

• 15-20% performance improvement
• $30-40M additional wealth
• That's the value of institutional memory

---

Try It Yourself

Nagare preserves institutional memory through:

• Single database (not scattered Excel files)
• Calculation snapshots (time-travel to past forecasts)
• Portfolio Companion (query history in natural language)
• Variance analysis (compare forecast vs actual)

Start building institutional memory today:

Start Free →

---

Related Reading:

• Three Questions Your Portfolio Model Should Answer
• Institutional Memory: Your Competitive Edge
• How to Evaluate GP Performance Over Time (coming soon)