Intelligence Arbitrage

How can one recurring SME decision retain enough intelligence to improve a later choice?

Your business does not lack intelligence. It loses intelligence between the person who notices and the system that acts.

A recurring decision passes through people, spreadsheets, inboxes, software, approvals, and memory. Each handoff can drop context. The next decision then repeats the analysis, applies a different standard, rediscovers an exception, or acts without leaving reusable evidence.

Intelligence arbitrage is the temporary value gap between what reality now makes possible and what the current decision system can perceive, route, or act upon.

The useful move is not to hoard the gap. It is to close one valuable gap with a governed route, observe the result, and retain what should improve the next decision.

Reality

Consider one recurring SME decision: which incoming opportunity should receive scarce specialist time this week?

The customer context sits in a CRM. Delivery capacity lives in a planning sheet. Risk is known by an experienced operator.

Margin is calculated elsewhere. Approval arrives in a message. The final choice may be sensible, but its route is hard to reconstruct.

The losses are predictable:

• analysis is repeated because earlier reasoning cannot be recalled;
• context goes stale between observation and action;
• exceptions stay hidden in one person's memory;
• standards change across teams or weeks;
• decisions leave no receipt that can teach a later choice.

This is a routing problem before it is an automation problem.

The telco precedent

An international telecom routing system faced a similar control problem. A call intent had to move through governed reference truth, eligible carriers, quality and commercial gates, a ranked route, settlement, and observed results:

call intent
→ governed reference truth
→ eligible carriers
→ quality and commercial gates
→ ranked route
→ settlement record
→ observed result

I worked with this pattern in an international telecom operating context in the early 2000s. The system reconciled fragmented data, applied hard routing controls, produced switch-ready plans, settled what happened, and used observed quality and commercial evidence to correct later routes.

That history proves that the routing-control architecture operated in one real telecom setting. It does not prove that transferring the architecture to knowledge work will improve growth, judgment, or customer value. Read the Telco Network Value Engine for the mechanism, historical boundary, and failure modes.

Dream

An intelligence system for SME decisions would preserve the same control shape:

intent
→ current context and precedent
→ eligible capability
→ human-owned standards
→ best viable route
→ attributable action
→ receipt
→ improved next route

The hypothesis is simple:

cycle output becomes next-cycle input improvement

A cycle compounds only when its result changes a later input, rule, capability, or precedent. For example:

• a lesson is promoted into precedent;
• a heuristic is promoted into a tested rule;
• a repeated route is encoded into a workflow;
• a capability is retained for later routing;
• a failed assumption is removed from future decisions.

A faster decision is not proof of compounding. Neither is a larger archive. The later decision must state how recalled evidence changed—or deliberately did not change—its reasoning, route, threshold, or action.

Use this model to locate weak links:

next-cycle decision capability
=
current capability
× context retained
× instruction precision
× route quality
× proof fidelity

This is a testable model, not a measured law or a promise of exponential growth. A weak factor can limit the whole route. The factors are not yet calibrated customer metrics.

Bridge

Transfer the telco architecture into one governed SME decision before building a factory.

Layer	Telco system	SME intelligence system	Control	Receipt
Intent	Destination and service	Consequential outcome	Named beneficiary	Decision identity
Truth	Numbering, prices, contracts	Current context and precedent	Source and freshness	Evidence snapshot
Capability	Available carriers	Humans, agents, tools	Demonstrated ability	Eligible routes
Standards	Reach, quality, capacity	Authority, virtue, risk, quality	Hard gates	Gate results
Route	Ranked carrier path	Selected work path	Explained ranking	Alternatives considered
Settlement	Call and commercial record	Action, owner, cost, consequence	Attribution	Outcome receipt
Learning	Quality and margin review	Updated precedent or rule	Later recall	Changed later decision

The layers have distinct jobs:

• Intelligence Arbitrage finds a valuable gap in the current route.
• A decision engine chooses one bounded route through SETPOINT → OBSERVE → GATE → RANK → ACT → SETTLE → LEARN.
• The Decision Optimisation Engine tests whether the receipt affected a later decision.
• A Software Factory repeats a proven decision route while preserving human authority.
• Compounding flow exists only when retained evidence demonstrably improves a later cycle.

The engine proves one decision route. The factory repeats a route only after the rule is stable enough to encode. Neither should take human authority over values, exceptions, irreversible commitments, or consequences.

Copyable experiment

Choose one recurring, consequential SME decision. A weekly opportunity, staffing, purchasing, scheduling, or exception decision is enough.

INTELLIGENCE ARBITRAGE EXPERIMENT

Decision:
Owner:
Beneficiary:
Valuable outcome:
Baseline:
Human-owned standards:
Hard gates:
Kill signal:

Current route:
People:
Data:
Tools:
Approvals:
Known handoff losses:

Comparable decisions: 3–5
Measures:
- decision lead time
- rework
- beneficiary outcome
- human agency
- proof retained

Later-decision test:
Which recalled precedent changed, confirmed, or did not apply to the route?

Rule supported by evidence:
Rule that must remain a heuristic:
Next review:

Run it:

1. Choose one recurring, consequential SME decision.
2. Freeze its beneficiary, standard, baseline, and kill signal.
3. Map the current route across people, data, tools, and approvals.
4. Run the governed route for three to five comparable decisions.
5. Compare lead time, rework, outcome, agency, and proof retention.
6. Require a later decision to state how recalled precedent changed—or deliberately did not change—the route.
7. Promote only the rule supported by observed evidence.

Stop if throughput or margin improves while beneficiary value, agency, trust, safety, or quality declines. These are gates, not weights that efficiency can overpower.

Experiment proof

Three to five decisions can show whether the route is usable, attributable, and worth testing further. They cannot establish broad compounding or guarantee growth.

Evidence	Honest conclusion
The route is faster but later decisions ignore its receipts	Local efficiency, not compounding
A later decision cites precedent but the beneficiary outcome declines	Memory exists; the route is not virtuous
A retained rule changes a comparable later route and improves the frozen gauge	Early evidence of compounding in this decision class
The same rule survives repeated comparable cycles	Candidate for careful workflow encoding

The strongest counter-case is that additional routing controls create more administration than value. If decision lead time, agency, or beneficiary value worsens without better proof, simplify or stop the route.

Next route

• Use the Decision Optimisation Engine to prove that one receipt changes a later choice.
• Use the Software Factory only after a repeated route has earned the right to be encoded.
• Begin an SME operations Journey by bringing one recurring decision, its current route, and the evidence you already trust. The first useful outcome is a bounded experiment, not a platform commitment.

Changes my mind: repeated comparable decisions show that the governed route adds cost or constraint without improving beneficiary value, agency, decision quality, or retained proof.

Context

• Regenerative Standards Economy — route a visible intelligence gap into a purpose-qualified experiment and return mature learning without treating selection or delivery as proof.
• Tight Five Loops — keep purpose, performance, platform, process, and people visible while testing the route.
• Telco Network Value Engine — inspect the lived routing precedent and its evidence boundary.
• Decision Optimisation Engine — test whether a receipt changes a later choice.
• Software Factory — encode a route only after repeated evidence supports the rule.

Questions

Next question: Which recurring decision loses the most useful context between the person who notices and the system that acts?

• Can you map one decision through intent, truth, capability, standards, route, settlement, and learning?
• Which standard must remain a hard gate even if throughput or margin rises?
• What later-decision evidence would prove that this route compounds rather than merely records?