Explain why it surfaced
Make supporting factors inspectable instead of asking the broker to trust a number.
Connected Logistics / Full use case
Connected Logistics overview Hard constraints → credible candidates → explanation → broker judgment.
Redwood Logistics / Carrier Matching
Turning Broker Judgment Into Explainable Recommendations
A carrier could look perfect mathematically and still be the wrong choice operationally.
IMPLEMENTED / EVOLVING WORK · PORTFOLIO RECONSTRUCTION
Summary
Carrier selection involved route, equipment, availability, accessorials, customer restrictions, preferences, pricing, margin, insurance, certifications, intermodal needs and relationship knowledge.
I helped turn that decision space into ranked guidance brokers could inspect and challenge—not an unquestionable answer.
Explore in Connected LogisticsEligibility ≠ suitability.
Can this carrier take the load?
Compatible equipment, a usable route, valid required credentials and known capacity make consideration possible.
Eligibility establishes the candidate set.Should the broker look here first?
Preferences, service history, timing, pricing and relationship context shape which eligible option is credible now.
Suitability helps prioritize attention.Generalized decision model, not a production screenshot or scoring-formula disclosure.
Matching reduced the search space. It did not remove the decision.
The tension
Brokers knew things the system did not.
Manual search relied on experience and memory. A bare algorithmic score risked false confidence.
Knowledge the system could represent
Formal requirements, recorded preferences, prior activity, pricing rules and known availability.
These could eliminate poor options and help rank the rest.
Knowledge carried by the broker
A recent conversation, a relationship concern, a changing market or newly available capacity.
An eligible lower-ranked carrier could still be the better first call.
Make expert judgment faster without pretending to replace it.
Research + system
Fit came from different kinds of evidence.
Some factors were mandatory gates. Others shaped preference, commercial viability or the broker’s next call.
Research → broker heuristics
Hard constraints to establish
Can this carrier serve the load?
Route, equipment, required accessorials, customer restrictions, insurance / certifications and intermodal requirements.
Comparative fit / broker context
Is this a credible first option?
Preferences, availability, pricing, margin, recorded history and relationship knowledge.
The boundary depends on the load.
A factor may be mandatory in one context and comparative in another. Hard gates do not become optional because a fit signal is strong.
How the problem changed
- 01Search the carrier universeA large field depends on broker knowledge.
- 02Eligibility removes impossible optionsHard requirements narrow the set.
- 03Ranking prioritizes credible optionsKnown fit makes comparison faster.
- 04Explanation exposes evidenceThe broker can inspect why an option surfaced.
- 05Broker decides / overridesContext outside the system still matters.
The pattern
Filter first. Rank second. Decide last.
Use rules to remove impossible choices and recommendations to prioritize plausible ones.
Early design / decision model
- 01Filter · hard constraintsRemove invalid choices before ranking.
- 02Rank · softer fit signalsPrioritize the credible set using known evidence.
- 03Decide · broker contextInspect, compare and apply knowledge the system may lack.
RECONSTRUCTION
Do not rank an invalid choice.
Wrong equipment, missing required credentials or a formal customer exclusion should prevent consideration—not merely subtract a few points.
Result: a candidate set that meets the load’s mandatory requirements. Unverified requirements still need review.
RECONSTRUCTION
Prioritize the credible set.
Compare eligible carriers using known lane fit, preferences, history, current availability and commercial context.
Result: a smaller set worth inspecting. A match score expresses relative fit, not guaranteed service or a probability of success.
RECONSTRUCTION
The broker retains the decision.
Inspect the reasons, compare alternatives and apply knowledge outside structured data. Choose, skip or search further.
Boundary: override the ranking—not mandatory compliance or customer requirements.
Design decisions
Explain the recommendation. Preserve the expert.
The score helps sort. The evidence helps decide what to do.
Separate gates from preferences
Missing required insurance is not equivalent to a weaker lane preference.
Reduce noise before review
A ranked list of hundreds still leaves the search problem with the broker.
Preserve ranking override
Recent conversations and relationship context can justify choosing another eligible carrier.
Expose uncertainty
Strong evidence of fit is different from simply finding no negative evidence.
Detailed matching workflow
- 01 · Entry point
Open load requirements
Schematic view / evidenceRoute · equipment · dates · customer · accessorials · commercial context.
- User action
- Review the known load requirements.
- System / shared response
- The carrier search starts from a specific decision context.
- 02 · Workflow state
Check hard eligibility gates
Schematic view / evidenceMandatory compatibility, customer and credential requirements.
- User action
- Verify the non-negotiable constraints.
- System / shared response
- Incompatible choices are removed; missing evidence requires review.
- 03 · Workflow state
Rank credible candidates
Schematic view / evidenceKnown lane fit · preferences · availability · history · pricing / margin.
- User action
- Review the smaller candidate set.
- System / shared response
- Ranking prioritizes attention; it does not guarantee service.
- 04 · Workflow state
Inspect recommendation rationale
Schematic view / evidenceSupporting factors and limited or stale context.
- User action
- Compare the evidence with current conversations and relationship knowledge.
- System / shared response
- The broker can judge what the ranking knows and what it may miss.
DecisionDoes the recommendation fit the current operational context?
Supported → selectChoose a credible eligible candidate
The broker owns the choice and the next booking or contact action.
Incomplete / conflicting context → override or searchChoose another eligible option or investigate further
Override the ranking, never mandatory requirements. Unverified eligibility remains a blocker.
Resulting state / return to evidenceBroker-owned choice, or continued investigation
The system narrows the search. It does not automatically award freight or treat a fit score as truth.
Directional · learn from disagreement
Repeatedly skipping highly ranked carriers could reveal stale preferences, missing customer rules, weak evidence or the wrong weighting.
Investigate the pattern with brokers. Do not treat every override as user error—or automatically train an opaque model on every click.
Expert override is evidence, not necessarily failure.
Fit guidance beside shipment requirements
Evidence → Decision → Tradeoff
- 01EvidenceBrokers juggle hard constraints and softer context.
- 02DecisionFilter, then rank credible options.
- 03TradeoffA score can create false confidence.
- 04ResponseInspectable rationale and broker override.
Impact in reality
What was real, what is reconstructed, and what remains directional.
No independent matching-throughput or acceptance result is claimed. The +15% booking result belongs to the separate one-click pilot.
Matching + broker review
Multiple fit factors, eligibility constraints, graded ranking, thresholds and retained broker judgment.
A director authored the algorithm; my work shaped interpretation and the broker-facing decision experience. This summary does not establish an exact shipped version or historical threshold boundary.
More explicit reasoning
Visible reason codes, clearer uncertainty, a stronger gates/preferences distinction and connected bidding/award context.
These explanatory behaviors are not all claimed in their current form historically.
Investigate disagreement
Use repeated ranking overrides to question missing evidence and assumptions.
Not autonomous selection, automatic learning or a validated predictive model.
Reality check · the evidence the system did and did not have
Reasonably representable
Equipment, route, formal customer restrictions, credentials, recorded preferences, historical activity, known availability and commercial rules.
Still difficult
Changing capacity, stale data, incomplete history, informal relationship knowledge, market conditions and recent conversations.
A recommendation is only as certain as the evidence it actually has.
From expert search to explainable decision support
- Identify hard constraints.
- Remove invalid choices.
- Identify softer signals.
- Rank the credible set.
- Explain why an option surfaced.
- Preserve expert override.
- Investigate repeated disagreement.
Useful anywhere software can narrow a complex choice but should not pretend to own the final judgment.
Connected Logistics: interpret what is happening → automate what is certain → support judgment where uncertainty remains.