Consultant Preference and Interest Matcher

Hard-skill fit and calendar space decide who can take the role. Preference matching only ranks among people who already can. Treat it as a second signal, not a replacement for a skills-to-project matching engine or a utilization and bench risk predictor.

If you fold interest into one blended "fit" score, you will staff the enthusiastic weak skill match, or skip the available person who has been quietly lobbying off the work. Keep three columns in front of the staffing lead: skill evidence, availability, interest. Argue with each column separately. The assignment stays a human decision.

The quality problem is simple. Two people with the same skill tags will not deliver the same way if one has been repeating a sector they are trying to leave. The matcher does not prove motivation. It surfaces a conversation you should have before the kickoff email goes out.

Run the matcher on a shortlist, not the whole bench. A firm-wide interest ranking invites gaming and typecasting. Ten people who already pass skill and calendar is a conversation. Two hundred ranked "culture fits" is a liability.

The allowed corpus is narrow on purpose.

• Bios and practice profiles the consultant can see and edit. Industry, methods, languages, travel posture, stated development goals.
• Engagement history as staffing already records it. Role, duration, sector, offer type (diagnostic, implementation, PMO). Not the client confidential work product.
• Explicit interest statements from mobility forms, marketplace applications, or a short survey the person signed.

Do not scrape Slack DMs. A private "I never want to work with that partner again" is not a staffing feature. Do not ingest performance reviews, 360s, or compensation notes. Those files were collected for a different purpose. Using them for resourcing without a separate, stated basis breaks trust and can violate employment rules.

Client names in history can be sensitive. Prefer sector and offer type over named accounts unless the consultant public bio already lists them. Client context for a new joiner belongs in a stakeholder onboarding brief synthesizer after the person is staffed, not in the decision of who gets the seat.

Tell people the matcher exists, what it reads, and how to correct a wrong inference. An interest file they cannot see will be treated as a secret score. If a short survey and the model disagree, believe the person. The model is reading prose. The person is stating a preference.

This is an illustrative staffing conversation, not a case study.

A staffing lead needs a senior for a fourteen-week public-sector ERP diagnostic. Skills matching returns two people with comparable SAP FI evidence and similar years. Utilization says both have calendar space in the window. Neither sits on a client the firm cannot pull them from.

Consultant A has a three-sentence bio: "results-oriented consultant who enjoys solving complex problems and working with stakeholders." Their last three engagements are manufacturing plants. The file does not say they dislike government work. It also does not say they want it.

Consultant B bio names public-sector ERP, points to a practice note, and lists three consecutive government engagements. They filled the mobility form with "remain in public sector this year."

What the matcher should return:

• B: documented interest in this sector and offer type, with quotes or form fields you can show them.
• A: insufficient interest evidence. Do not infer "does not want public sector" from a manufacturing trail. Do not infer "collaborative" or "low energy" from a thin bio.
• Open question: is A being typecast into plants because that is all they have been given?

The useful output is a prompt for a ten-minute conversation, not an assignment. The lead asks A whether they want a stretch into public sector, and asks B whether "remain in public sector" still holds or was a way to avoid a travel-heavy plant job. Then the lead staffs.

If you are hiring into the role rather than staffing from the bench, a bio is not a substitute for a structured interview. Use a role-specific interview question generator for competencies. Do not treat a LinkedIn-style summary as proof of motivation.

Two failure modes show up in the same week if you do not look for them.

A short bio plus a few project titles is not a psychograph. Models still emit "strong client presence," "prefers independent work," "culture fit for a demanding sponsor." That language has no source. Ban trait labels unless they quote a sentence the person wrote or a field they filled. If the bio is thin, the interest column is "unknown," not "neutral," and not a guessed temperament.

Once people learn that "low interest" keeps them off a role, the interest file becomes a way to refuse hard staffing: the turnaround nobody wants, the client with a reputation, the bench-filling PMO, the travel week. Partners learn the same trick on behalf of their favorites. The matcher will corroborate a polished bio that only lists desirable sectors.

Guardrails staffing leads can actually run:

• Interest never auto-rejects a qualified, available person. It flags a conversation.
• A pattern of "low interest" on every difficult seat is a career conversation, not a model output. Utilization still shows who is idle. Someone who is available and skilled does not get a permanent exemption from unglamorous work because their bio is carefully written.
• Do not feed inferred preference into agentic re-staffing on scope change as an automatic swap criterion. Scope-driven moves already disturb clients and careers. Adding "they would be happier elsewhere" is how you churn a team under the banner of fit.
• Audit for concentration: the same names on every interesting sector, the same names on every grind. If the interest column explains that split, check whether the file is accurate or protective.

Stated interest decays. A form from two years ago is not a current preference. Require a refresh when the person updates their bio, or on a fixed cadence, and believe a live correction over an old inference.

The weekly loop is short.

1. Skill shortlist from profiles and delivery history, not from last year tags alone.
2. Availability filter from committed and pipeline work, including bench risk.
3. Interest pass on that shortlist only, from opted-in unstructured text and history.
4. Conversation with evidence: quoted bio lines, form fields, engagement list. The consultant can strike a wrong inference before the role is locked.
5. Human assignment. No silent ranking that becomes the staffing sheet.

Do not merge the three scores. A single number hides whether you are compromising on skill, on calendar, or on willingness. Compromises happen. Name them. Do not write the interest inference back into the PSA resource record as a sortable field. Partners will sort it. Keep it on the shortlist pack for this role.

Tooling sits in classes, not a ranked list. Skills graphs inside professional services automation, including Kantata and Mosaic, are the hard-skill backbone. They tell you who has done this kind of work. They do not, on their own, tell you who wants the next one. Talent marketplaces, internal or vendor-hosted, collect explicit applications to open roles, which is a cleaner interest signal than a parsed paragraph. A general LLM is useful when the bio and history are unstructured and the consultant has opted in. Point the model at those texts. Do not give it Slack, review packets, or private CRM notes.

The system is working when two qualified, available seniors produce a one-page interest note the staffing lead can argue with, the person can correct the file, and the firm still fills the hard seats. It is failing when a three-line bio becomes a personality sketch, or when a practice has staffed itself into only the work it already likes.