Crewpath's scoring model was designed with one constraint above all others: every score must be explainable. Not just in aggregate — dimension by dimension, input by input. If a partner is going to trust a ranked output, they need to understand why Consultant A scored 91 on domain expertise and Consultant B scored 67. A black box that produces a match is not a tool that gets adopted in a professional services partnership. An explainable engine that shows its reasoning earns a place in the workflow.
This post walks through the four dimensions in the default model, what data feeds each one, and how the composite score is constructed.
Dimension 1: Domain Expertise (default weight: 30%)
Domain expertise answers one question: how well does this consultant's project history match the engagement type on the table? It is the most structurally important dimension and the one most firms want to be confident in before going live.
The model draws from the last 36 months of project records. Engagements are classified against a standardized taxonomy of consulting types — M&A integration, operational restructuring, cost optimization, change management, financial modeling, technology implementation, and around two dozen others covering the professional services landscape. The taxonomy is firm-configurable; the default set covers 28 engagement types that map well to most generalist consulting and advisory practices.
Recency is weighted within that 36-month window. Work completed in the last 12 months carries full weight. Work in the 12-to-24-month band carries roughly two-thirds weight. Work in the 24-to-36-month band carries one-third. The decay logic reflects real knowledge currency: an M&A integration project completed last quarter is more predictive of performance on the next M&A engagement than work from two and a half years ago.
Certifications and formal qualifications contribute a secondary signal on top of project history. A CFA designation on a finance-heavy engagement, or a certified change management practitioner credential on a restructuring project, adds a fixed boost within the relevant taxonomy category. The certification signal does not override sparse project history — it supplements it. A consultant with strong recent project history and no certification outscores a consultant with a credential and no recent relevant projects.
Dimension 2: Client Chemistry (default weight: 25%)
Client chemistry is the model's most subjective-seeming dimension and the most important to instrument correctly. It draws from three sources: post-engagement ratings, relationship flags, and sector proximity signals.
Post-engagement ratings are the feedback scores your firm collects from clients after each project closes. If a consultant delivered a strong engagement for a client and received a high rating, that signal persists as a positive chemistry indicator for subsequent potential engagements with that client. The decay function for relationship signals is deliberately slower than for domain expertise — relationships age less quickly than technical currency, and a strong track record with a client from 30 months ago is still meaningful information.
Relationship flags are manual inputs: partners can mark a consultant-client pairing as preferred or flagged, with a required reason code. Flags adjust the quantitative rating signal; they don't hard-block. The reason code is always visible to the decision-maker reviewing the ranked output, which preserves transparency even when a manual input is overriding the quantitative baseline.
Sector proximity is the fallback for new pairings — a consultant who has never worked with a given client but has strong ratings from clients in the same industry sector receives a modest positive chemistry indicator. It's a weak signal, but it's better than treating the pairing as entirely unknown.
Dimension 3: Utilization Pressure (default weight: 25%)
Utilization pressure answers the firm-side economic question: how much does this consultant need billable work right now, and how well does their availability window align with the engagement start date?
The calculation combines current billable load as a percentage of target utilization with the timing and length of the next open bench window. A consultant at 60% billable load with an engagement ending in two weeks has high utilization pressure — they are approaching bench and the timing is urgent. A consultant at 92% billable load with a 16-week engagement confirmed has effectively zero utilization pressure — deploying them on another engagement would push them into overextension, which is a separate problem the model accounts for.
Timing alignment matters as much as the utilization rate itself. A consultant who is currently on bench but whose schedule is committed starting in three weeks scores lower on this dimension than one whose bench window opens on the same day the engagement needs to start. The dimension measures productive deployment opportunity, not just idle time.
Dimension 4: Bench Cost Delta (default weight: 20%)
Bench cost delta is the most explicitly economic dimension. It compares the loaded daily cost of the consultant sitting idle against the contribution margin they generate if deployed on the engagement in question.
A consultant whose fully-loaded daily cost (salary, benefits, overhead, employer contributions) approaches or exceeds the billing rate for the engagement represents a poor economic match, independent of how well they fit on the other three dimensions. A consultant whose cost is substantially below the engagement billing rate creates strong margin — and when they are also available and qualified, deploying them reduces bench drag while improving the profitability of the engagement.
We want to be direct about where this dimension can be misused: bench cost delta is not a mechanism for systematically excluding senior consultants from high-fit engagements. A 20% weight means economics are one consideration among four, not a veto. When domain expertise and client chemistry clearly point to a specific consultant, the bench cost signal does not override that recommendation — it makes the tradeoff visible so the confirming partner is working with full information rather than partial information.
The Composite Score and Weight Tuning
The four dimension scores combine through a weighted sum. In the default model: Domain Expertise (30%) + Client Chemistry (25%) + Utilization Pressure (25%) + Bench Cost Delta (20%) = Composite Score, expressed out of 100. The ranked list generated for every engagement brief shows the composite score plus each dimension's individual contribution, so a decision-maker can immediately see which dimension is driving a ranking.
Growth and Enterprise customers can adjust these weights through the admin panel. A firm that consistently prioritizes client relationships above all other factors can shift client chemistry to 35% and reduce domain expertise to 25%. A firm running a high-volume, lower-margin practice may weight bench cost delta at 30%. The model recalculates all active scoring passes immediately when weights change. Historical runs retain their original weight configuration — that auditability is non-negotiable, because changing weights retroactively would undermine the model's usefulness as a calibration tool.
What the model does not do
The scoring model does not predict engagement success and does not claim to. It identifies the consultant most likely to be appropriate based on the structured signals available — project history, relationship data, availability, and economics. The factors that don't appear in any database — interpersonal style, leadership presence, micro-specializations within a taxonomy category, context known only to the originating partner — remain in the partner's judgment when they review the ranked output and decide whether to confirm the top match or override it.
That is the design intent. The model narrows the field and surfaces credible candidates with explicit reasoning. The partner confirms or overrides with a documented reason. Overrides are tracked, and the patterns in those overrides are what calibrate the model's weights over time toward what each firm actually values — not what the default model assumed they would value. The two should converge over 6 to 12 months of active use.