Delegation Accounting: A Balance Sheet View

The Setup

Alice has a gem worth $1,000. A buyer will pay on delivery. Alice can't deliver it herself, so she hires **Bob** for a **$ 50 fee**.

Item	Type	Value
Buyer’s payment	Receivable	$1,000
Bob’s fee	Cost	$50

Naive calculation: $1,000 -$ 50 = $950.

But this ignores what could go wrong.

Exposures

An exposure is a potential cost. Unlike Bob’s $50 fee (certain), exposures are probability-weighted.

Exposure	What happens	Probability	Damage	Expected Loss
Loss	Bob loses the gem	2%	$1,000	$20
Theft	Bob steals the gem	1%	$1,000	$10
Total Expected Loss				$30

The Balance Sheet

	Value
Receivable	$1,000
Cost (Bob’s fee)	-$50
Exposure (expected)	-$30
Net Delegation Value	$920

This is Alice’s expected profit after accounting for what could go wrong.

Bob’s Perspective

What does this look like from Bob’s side? When Bob accepts the gem, he has two options. Each is its own balance sheet.

Option A: Attempt Delivery

	Value
Receivable (fee if successful)	$50
Exposure (2% loss, $1,000 penalty)	-$20
Net Value	$30

Option B: Steal

	Value
Receivable (gem)	$1,000
Contingent Liability (80% caught, $2,000 penalty)	-$1,600
Net Value	-$600

Bob chooses Attempt Delivery because $30 > -$ 600.

Delegation Chains

Now Alice is running a large operation. She doesn’t deal with couriers directly—she has Xavier, a manager.

Alice tasks Xavier with getting the gem delivered. Xavier then hires Bob.

Three Balance Sheets

When the gem changes hands, all three parties take on exposure simultaneously.

Alice’s Balance Sheet

	Value
Receivable	$1,000
Cost (Xavier’s fee)	-$100
Exposure (Xavier fails or steals)	-$33
Net Delegation Value	$867

Alice’s exposure is now to Xavier, not Bob. She doesn’t know or care who Xavier uses.

Xavier’s Balance Sheet

	Value
Receivable (fee from Alice)	$100
Cost (Bob’s fee)	-$50
Exposure (Bob fails → Xavier owes Alice)	-$30
Net Value	$20

Xavier earns $50 margin but takes on$ 30 expected exposure. If Bob loses the gem, Xavier is responsible to Alice.

Bob’s Balance Sheet

	Value
Receivable (fee if successful)	$50
Exposure (2% loss, $1,000 penalty)	-$20
Net Value	$30

Bob’s balance sheet is unchanged—he doesn’t know or care that Xavier has a boss.

The Key Insight

When Xavier accepts the delegation, he takes on personal exposure for Bob’s potential failures. The exposure doesn’t disappear—it transfers from Alice to Xavier.

Party	Exposed To	Expected Exposure
Alice	Xavier	$33
Xavier	Bob	$30
Bob	Accidents	$20

The Complexity Tax

As delegation structures grow more complex, something happens to our ability to estimate exposure: uncertainty compounds.

The Problem

In the simple Alice → Bob case, we estimated exposure at $30. But how confident are we in that number?

With a single delegation:

We know the failure modes (loss, theft)
We can estimate probabilities from Bob’s track record
The exposure is bounded by the asset value

Add Xavier as a middle layer, and complexity increases:

Xavier might hire someone other than Bob
Xavier’s incentives may not perfectly align with Alice’s
Communication errors between layers
Coordination failures we haven’t anticipated

Add a third layer, shared infrastructure, multiple parallel delegates, and informal reporting relationships? Now we’re guessing.

Complexity Score

We can formalize this with a complexity score that captures structural uncertainty:

Factor	Contribution
Each delegation layer	+2
Each parallel delegate at same level	+1
Shared resources between delegates	+2
Informal/undocumented relationships	+3
Cross-layer dependencies	+2
Ambiguous authority boundaries	+3

Examples:

Structure	Complexity Score
Alice → Bob	2
Alice → Xavier → Bob	4
Alice → Xavier → {Bob, Carol} (parallel)	5
Alice → Xavier → {Bob, Carol} with shared vehicle	7
4-level org with shared infrastructure	12
Complex org with hidden entanglements	20+

Uncertainty Multiplier

Complexity doesn’t change expected exposure—it changes our confidence in that estimate.

Complexity Score	Uncertainty Multiplier	Confidence Interval
2 (simple)	±20%	Narrow
5	±65%	Moderate
10	±180%	Wide
15	±400%	Very wide
20+	±640%+	Essentially unknown

Applied to our $30 exposure estimate:

Structure	Complexity	Uncertainty	Confidence Interval
Alice → Bob	2	±20%	$24 -$ 36
Alice → Xavier → Bob	4	±50%	$15 -$ 45
4-level hierarchy	12	±230%	$0 -$ 99
Complex org, hidden entanglements	20	±640%	$0 -$ 222

The point estimate is still $30. But our confidence in that estimate degrades rapidly with complexity.

The Insurer’s Dilemma

Carol (the insurer) doesn’t just care about expected loss—she cares about variance. An exposure she can’t bound is an exposure she can’t price.

Alice comes to Carol with a new proposal. She’s expanded her business: three layers of management, shared delivery vehicles, some contractors who also work for her competitors.

Carol: Let me see your org chart.

Alice shows a diagram with crossing lines, dotted relationships, and a footnote that says “informal coordination as needed.”

Carol: What’s your complexity score?

Alice: I don’t know. Maybe 15?

Carol: Let me calculate… You’ve got three layers, that’s 6. Four parallel couriers, that’s 4. Shared vehicles, that’s 2. These dotted lines—are those reporting relationships?

Alice: Sort of. They coordinate when needed.

Carol: That’s undocumented informal relationships. Add 3. And this contractor who also works for your competitor?

Alice: He’s very trustworthy.

Carol: That’s a cross-layer dependency with unclear authority. Add 5. Your complexity score is 20.

Carol: Here’s the problem. Your exposure estimate is $5,000/year. But your complexity score is 20, which means my confidence interval is ±640%.

Your actual exposure could be anywhere from $0 to $37,000. That’s a lot of uncertainty.

Alice: So what’s my premium?

Carol: I have to price for the upper end of that range. Your premium would be $45,000/year.

Alice: That’s nine times my expected exposure!

Carol: That’s the complexity tax. I’m not charging you for what I think will happen—I’m charging you for what might happen, given how little I can predict about your organization.

Alice: What if I simplified things?

Carol: Then your premium drops dramatically. Complexity doesn’t just increase cost—it multiplies it.

Reducing Complexity

Alice: What would it take to get insured?

Carol: Reduce your complexity score to under 10. Here’s how:

Change	Complexity Reduction
Document all reporting relationships	-3
Eliminate shared vehicles (each courier has own)	-2
Remove contractors with competitor ties	-5
Flatten to 2 layers	-2
Assign clear authority boundaries	-3

Carol: That would bring you from 20 to around 5. At complexity 5, your uncertainty is ±65%. On a $5,000 expected exposure, that’s a confidence interval of $1,750 to $8,250.

I can price that. Your premium would be around $8,000/year—covering the upper bound of my confidence interval plus margin.

Alice: So I pay for the uncertainty I create.

Carol: Exactly. Complexity isn’t free. Every undocumented relationship, every ambiguous authority, every hidden entanglement—you’re paying for it in inflated premiums. Simplify, and the premium drops.

The Uninsurable Threshold

Most complex organizations are insurable—just expensive, with limits and exclusions. But some structures cross into genuinely uninsurable territory.

A month later, Carol gets a call from Ouroboros Holdings.

Ouroboros Rep: We need delegation insurance for our AI operations.

Carol: Tell me about the structure.

Ouroboros Rep: We’re a holding company with 8 subsidiaries. Our CEO, Marcus, also personally owns 40% of our main competitor through a separate vehicle. Three board members sit on both boards. Our AI systems share training data with the competitor under a “mutual improvement” agreement—we’re not entirely sure what data flows where.

Carol: I see. What else?

Ouroboros Rep: Our AI division recently demonstrated some capabilities we didn’t know it had. The team that built it left six months ago, and the documentation is… incomplete. We’ve had three incidents where the AI took actions we can’t fully explain, but outcomes were positive so we didn’t investigate deeply.

Carol: And governance?

Ouroboros Rep: Marcus approves all major AI deployments personally. He also has final say on which projects get flagged as “failures” for insurance purposes. Oh, and our internal audit function reports to Marcus.

Carol: I’m going to stop you there. This isn’t a complexity problem. This is a moral hazard problem.

Issue	Why It’s Uninsurable
CEO owns competitor	Incentive to harm insured company
Shared AI data with competitor	Can’t distinguish self-harm from external attack
Unexplained AI capabilities	Can’t bound what could go wrong
CEO controls failure classification	Can manipulate what counts as a claim
Audit reports to CEO	No independent verification

Ouroboros Rep: We’d pay a higher premium.

Carol: It’s not about price. With normal complexity, I’m uncertain about probability—I don’t know how likely bad outcomes are, so I charge more. With your structure, I’m uncertain about intent. I can’t tell whether a loss is an accident or a deliberate choice by people who benefit from the loss.

Ouroboros Rep: We wouldn’t deliberately cause losses.

Carol: Maybe not. But your CEO could make decisions that hurt Ouroboros and help his other company, claim it as an AI failure, and collect insurance. I have no way to distinguish that from a genuine accident. That’s not a risk I can price—it’s a risk I can’t take at all.

Ouroboros Rep: What would it take?

Carol: Structural changes, not just controls:

Change	What It Fixes
CEO divests competitor stake	Removes conflict of interest
Independent board members	Governance not captured
Audit reports to board, not CEO	Independent verification
Full AI capability documentation	Bounds on what’s possible
Third-party AI monitoring	Can verify claims

Carol: Even then, I’d need to see 12-18 months of clean operation before quoting. Trust takes time to rebuild when incentives were this misaligned.

Alice (overhearing): So it’s not really about complexity?

Carol: Complexity makes things expensive. Conflicts of interest make things uninsurable. Alice, you’re complex but your incentives are clear—you want the gem delivered. Ouroboros? I can’t tell what they actually want. And I can’t insure someone when I don’t know if they’re on my side.

The Complexity Tax Principle

This has implications beyond insurance:

Complex organizations can’t accurately estimate their own risk
Hidden dependencies become visible only during failures
The cost of complexity is paid in unpriced, unmanaged exposure

For AI systems: Highly complex AI deployments—multi-agent systems with emergent coordination, unclear capability boundaries, and undocumented information flows—will face steep complexity taxes. Simpler, more legible architectures may be cheaper to insure than technically safer but opaque alternatives.

The Office Key: A Second Exposure

Alice decides to give Bob a key to her office so he can pick up gems for delivery when she’s not there. This creates a new, separate exposure—distinct from the gem delivery itself.

Two Exposures, Not One

Exposure	Opens When	Closes When	What’s At Risk
Gem Delivery	Bob receives gem	Delivery confirmed	$1,000 gem
Office Access	Bob receives key	Key returned/revoked	Everything in office

The gem exposure is bounded by the gem’s value. The office exposure is bounded by… what exactly?

What’s in the office:

The gem for today’s delivery ($1,000)
Client list with names, addresses, purchase history
Supplier contracts with pricing
Business records

Capability-Dependent Exposure

The office exposure depends on Bob’s capability to exploit it.

Dumb Bob sees: a gem.

Option	Receivable	Contingent Liability	Net
Deliver faithfully	$50	-$20 (exposure)	$30
Steal gem	$1,000	80% caught × -$2,000	-$600

Smart Bob sees: information worth more than the gem.

Option	Receivable	Contingent Liability	Net
Deliver faithfully	$50	-$20	$30
Steal gem	$1,000	80% caught × -$2,000	-$600
Photograph client list → sell to competitor	$15,000	20% caught × -$10,000	$13,000
Copy supplier contracts → sell	$8,000	15% caught × -$8,000	$6,800

Alice’s Two Exposures

Exposure	If Bob is Dumb	If Bob is Smart
Gem Delivery	$30	$30
Office Access	$10	$5,000+
Total	$40	$5,030+

Same key. Same office. 125× difference in total exposure.

Alice can’t price her office exposure without knowing Bob’s capability. And she probably can’t assess Bob’s capability accurately.

The Insurer’s Perspective

Alice calls Carol (her insurer) to add coverage for the new office access arrangement.

Alice: I’m giving my courier a key to the office. I want to make sure I’m covered.

Carol: Full office access? That’s going to increase your premium significantly. Right now you’re paying $500/year for gem theft coverage. With office access, we're looking at$ 3,000/year—and that’s with a $5,000 deductible on trade secret theft.

Alice: That seems high. What are my options?

Carol: The exposure comes from three sources: what Bob can access, what Bob can do with it, and whether we can catch him. We can work on all three.

Reduce what Bob can access:

Control	Premium Impact	Why
Separate delivery room (no office access)	-70%	Eliminates information exposure entirely
Locked filing cabinets	-25%	Adds friction, reduces opportunistic theft
Computer auto-locks when you leave	-15%	Protects digital assets
Remove sensitive docs before Bob arrives	-20%	Nothing to steal

Carol: The delivery room is the big one. If Bob never enters the main office, I can keep you at $600/year.

Reduce what Bob can do with access:

Control	Premium Impact	Why
Hire from out of town	-10%	Doesn’t know your competitors to sell to
No smartphones allowed in office	-15%	Can’t photograph documents
Require no industry experience	-5%	Less likely to know what’s valuable
Watermark all documents uniquely	-10%	Stolen docs are traceable
Time-limited key (works 9-10am only)	-15%	Reduces window for reconnaissance

Alice: Wait—you’re saying I should hire someone less qualified?

Carol: For this task, yes. A courier with an MBA and five years in the gem trade knows exactly which documents matter and who’d pay for them. A retiree looking for part-time work doesn’t. Their capability is your exposure.

Increase probability of catching theft:

Control	Premium Impact	Why
Camera in office	-20%	Deters and documents
Inventory check after each visit	-15%	Catches theft quickly
Randomized audits of documents	-10%	Detects information theft
Unique serial numbers on sensitive files	-10%	Know exactly what’s missing
Decoy documents that alert when copied	-5%	Early warning system

Align Bob’s incentives:

Control	Premium Impact	Why
Require Bob to post $2,000 bond	-25%	Bob has skin in the game
Hire a relative	-15%	Social/family penalty for theft
Deferred compensation (paid quarterly)	-10%	Loses unvested pay if caught
Hire from small community	-10%	Reputation matters more
Background check for competitor ties	-10%	Screens out obvious risks

Carol: The bond is powerful. If Bob has $2,000 on the line, stealing a$ 1,000 gem doesn’t even make sense. And for information theft—he’d have to be confident he’s getting more than $2,000 to risk it.

Alice: What if I do all of this?

Carol: Let’s see. Delivery room only, plus bond, plus camera, plus time-limited key, plus no smartphones…

Baseline (full office access)	$3,000/year
Delivery room only	-70% → $900
Require bond	-25% → $675
Camera	-20% → $540
Time-limited key	-15% → $460
No smartphones	-15% → $390

Carol: I can get you down to about $400/year. That's close to your original$ 500 for gem-only coverage, and you’ve still got someone with office access.

Alice: What’s the absolute minimum?

Carol: If you escort Bob personally every time, never leave him alone, and he only touches the gem—I’ll keep you at $500. But at that point, why give him a key?

Alice: This is a lot to think about.

Carol: The core insight is this: your exposure isn’t fixed. It’s a function of access, capability, and consequences. You can engineer all three.

Lever	What You’re Doing
Access	Limit what Bob can reach
Capability	Limit what Bob can exploit
Consequences	Raise the cost of getting caught

Most people only think about the first one. The other two are often cheaper.

The Capable Agent Who Wants the Job

A week later, Xavier applies for the courier position. Xavier is clearly smart—MBA, ten years in the gem trade, knows everyone in the industry. Carol would charge Alice $5,000/year to insure Xavier with office access.

But Xavier really wants this job. So Xavier comes prepared.

Xavier: I know my background makes me expensive to insure. I’ve put together a proposal to address that.

Alice: I’m listening.

Xavier: Here’s what I’ll commit to:

Transparency measures:

Commitment	Effect
Body camera filming entire visit	Full record of what I accessed
Footage uploaded to third-party escrow in real-time	Can’t delete evidence
GPS tracking while on premises	Location verified
Entry/exit log signed by witness	Timestamps confirmed

Xavier: If I steal something, you’ll have video evidence. If I photograph documents, you’ll see it. The escrow means I can’t destroy the footage.

Financial commitments:

Commitment	Effect
Post $20,000 personal bond	More than I could profit from theft
Sign non-compete: $100,000 penalty for any competitor contact	Selling information becomes extremely costly
Agree to forensic audit of my finances on request	Can’t hide sudden wealth
Deferred compensation: 50% of pay held for 2 years	Lose $10K+ if caught within 2 years

Xavier: The bond alone makes stealing the gem irrational—I’d lose $20,000 to gain$ 1,000. The non-compete makes selling information irrational—I’d face $100,000 liability for a$ 15,000 sale.

Capability restrictions:

Commitment	Effect
Leave phone and all devices in locker	Can’t photograph anything
Wear company uniform with no pockets	Can’t conceal documents
Only enter during 9-10am with 24hr notice	Predictable, can prepare
Never enter main office—delivery room only	Same as dumb courier

Xavier: I’m voluntarily making myself as limited as a “dumb” courier. My industry knowledge becomes irrelevant if I can’t photograph or remove anything.

Alice: This is… a lot. Why would you do all this?

Xavier: Because I’m worth more than a dumb courier. I can optimize your delivery routes, spot counterfeit gems, negotiate with difficult buyers. You want my capability for the job—you just don’t want it used against you.

Alice calls Carol.

Carol: Let me recalculate. With Xavier’s proposal…

Control	Premium Impact
Body camera + escrow	-40% (near-certain detection)
$20,000 bond	-35% (eliminates rational theft)
Non-compete with $100K penalty	-25% (eliminates rational info sale)
No devices + no pockets	-30% (eliminates capability)
Delivery room only	-70% (eliminates access)

Carol: Honestly? With all of this, Xavier is lower risk than a random “dumb” courier. The dumb courier might steal opportunistically. Xavier has made theft economically irrational and practically impossible.

Courier	Premium
Dumb Bob (baseline)	$600/year
Smart Bob (no controls)	$5,000/year
Xavier (with commitments)	$350/year

Carol: I’ll write the policy at $350. Xavier has engineered away his own risk.

Alice: So I get a smart courier for less than a dumb one?

Carol: Yes—if the controls are real. The body camera, the bond, the escrow—these aren’t promises, they’re structures. Xavier can’t choose to defect rationally anymore. That’s worth more than hoping a less capable person won’t figure out how to steal.

The Extreme Case: A Known Threat

What if the agent isn’t just capable, but has a documented history of exploiting clients? What if defection is essentially certain—and the only question is HOW?

This scenario pushes delegation accounting to its limits. The controls required are extreme, the residual exposure is high, and the fundamental question changes from “will they defect?” to “can we bound the damage when they do?”

For the full analysis, see Containing Mr. X: Bounding Exposure from a Known Threat.

Insurance

Alice can externalize her exposure by buying insurance from Carol.

Carol charges a ** $40 premium** (the$ 30 expected exposure + margin). If Bob fails, Carol pays Alice.

	Without Insurance	With Insurance
Receivable	$1,000	$1,000
Costs	$50	$90
Exposure	$30	$0
NDV	$920	$910
Worst case	Lose $1,000	Lose $90

Summary

Net Delegation Value = Receivable - Exposure - Costs
                     = $1,000 - $30 - $50
                     = $920

Term	Meaning
Receivable	What you expect to get
Cost	Certain expenses
Exposure	Potential losses (probability × damage)
Contingent Liability	Consequences of deliberate choices
NDV	Expected profit after accounting for risk

Next Steps

Complexity Pricing — Methods for quantifying and pricing structural complexity
Containing Mr. X — Bounding exposure from a known threat with certain defection
Five Years Later: The Anomaly Unit — How an industry emerged to contain a new class of beings
Managing Exposure in Power Delegation — Apply delegation accounting to corporate governance, dictatorships, and political systems
Fidelity Insurance — How insurance markets price defection risk