Every resource manager has a mental model of consultant-client chemistry. They just don't usually call it that. They call it "a good fit," or "the client will love her," or "I know this client, they need someone more senior." These judgments are real, they often turn out to be correct, and they are almost entirely absent from formal allocation frameworks.
The question worth asking is whether chemistry can be operationalized — turned into a scoring signal that sits alongside domain expertise and utilization targets in the matching model, rather than living exclusively in a managing director's memory. My position, after spending several years building recommendation systems before we applied this thinking to professional services, is: yes, but only if you're honest about what chemistry actually is and what data can and can't encode it.
The Asymmetry Between Domain Fit and Chemistry
Domain fit is relatively easy to model. You have a skill taxonomy, you have an engagement requirement, you run a matching function. The output is noisy but directionally useful. Most PSA systems support this natively at some level of skill tagging depth.
Chemistry is harder because it's relational, not categorical. Domain fit is a property of the consultant. Chemistry is a property of the pairing — it emerges from the interaction between a specific consultant and a specific client, and it changes over time as that relationship develops. A consultant who had a strained engagement with a client two years ago, when the project scope was poorly defined, might have excellent chemistry with that same client today on a well-scoped advisory engagement. The variable isn't the person or the client — it's the context.
This contextual dependency is why chemistry is hard to model naively: you can't just tag a consultant as "good with financial services clients" and call it a day. The tag conflates dozens of distinct client relationships, each with its own trajectory.
Three Measurable Proxies for Chemistry
Rather than trying to measure chemistry directly, the more tractable approach is identifying measurable proxies — observable outcomes in the PSA record that correlate with good chemistry. These are lagging indicators, not direct measures, but they're stable enough to use as scoring inputs.
Post-Engagement Client Satisfaction Scores, Disaggregated
Most consulting firms collect some form of client satisfaction data at engagement close — an NPS survey, a structured feedback form, a rating on a 1–5 scale. The challenge is that these scores are typically stored at the engagement level in the PSA, not disaggregated by consultant. An engagement involving four consultants gets one client satisfaction score, and it's not attributed to individuals.
When you do have individual-level satisfaction data — even a subset — it becomes one of the strongest predictors of future engagement success. The signal isn't perfect: satisfaction scores are influenced by scope management, pricing, and factors outside any consultant's control. But across a sufficient sample (we've found 8–12 prior data points per consultant tends to be enough for a stable signal), the noise averages out and the relationship between prior satisfaction scores and future engagement performance is meaningful.
Re-Engagement Rate by Client
When a client requests a specific consultant by name for a follow-on engagement, that's a strong chemistry signal — arguably the strongest you can observe in a PSA system. It doesn't require a satisfaction score to be collected. It's visible in the engagement record as a consultant-to-client repeat pattern.
The inverse signal is also informative: when a client who typically requests specific consultants by name doesn't request them again, that pattern is worth tracking. It rarely means the engagement was a failure, but it's a weak negative signal worth weighting appropriately.
Scope Change Patterns by Consultant-Client Pairing
Engagements that expand in scope mid-delivery — additional phases added, scope extended, follow-on work authorized before the original engagement closes — correlate with strong client satisfaction and, implicitly, with good chemistry. A consultant who consistently generates scope expansions with specific client types is delivering enough value that clients want more, which is as close to a behavioral chemistry signal as the PSA record can produce.
This proxy is noisy in the opposite direction: scope expansion can also reflect poor initial scoping. The signal needs to be paired with satisfaction data to distinguish "good chemistry led to expanded work" from "poor scoping led to change orders." When both signals align — scope expansion and strong satisfaction — the chemistry inference is much more reliable.
The People Graph Model
When you formalize these proxies as edges in a directed graph — where nodes are consultants and clients, and edges carry weights based on prior engagement outcomes — you get a structure we call the People Graph. Each pairing has an associated chemistry score that updates each time a new engagement closes between those nodes.
The People Graph has useful properties for allocation scoring. New consultants start with no edge weights, which means they get scored on domain fit and utilization targets only — a reasonable prior given no history. Consultants with a rich history of engagements with a specific client type get scored more confidently, because there's more historical signal to draw on. The model degrades gracefully when data is sparse and improves as the data accumulates.
The more interesting cases are consultants with strong chemistry scores at one client and poor scores at another in the same sector. A human reviewer looking at their CV would see consistent "financial services" experience and might not distinguish between these. The graph model does distinguish, because it's tracking the pairing, not just the domain.
What the Model Doesn't Replace
There are chemistry dimensions that no PSA record will ever capture. Whether a consultant's communication style meshes with a particular client stakeholder. Whether there's a personal connection or a prior relationship that preexists the formal engagement. Whether a client is going through an internal reorganization that will make any new consultant relationship difficult regardless of fit scores.
We're not arguing that a chemistry score should override a resource manager's direct knowledge of a client situation. The value of the model is not in replacing that judgment — it's in making the judgment explicit and consistent across a larger consultant pool than any single manager can hold in their head simultaneously.
A managing director with 10 years of client relationship history can make good chemistry calls on 20 consultants they know well. At 150 consultants, that same director is making chemistry calls on dozens of people they've barely interacted with, based on secondhand knowledge and pattern-matching to prior placements. The model doesn't replace the 20-consultant intuition — it extends it to cover the other 130.
Implementation Without the "Black Box" Problem
The practical resistance to chemistry scoring at most firms is what I'd call the black box problem: resource managers won't trust a score they can't explain to a managing director or a client. If the scoring model says "this consultant has a 62% chemistry fit with this client" and no one can explain why, the score gets ignored.
The solution is factor decomposition — showing the chemistry score not as a single number but as a weighted combination of its component signals: prior satisfaction score (with this client, or with comparable client profiles), re-engagement rate, scope expansion pattern, and tenure in the relevant sector. When a resource manager can see that a score is 62% because there are two prior engagements with this client type, one of which produced a below-average satisfaction score and neither produced a re-engagement request, the score becomes something they can interrogate rather than something they have to accept on faith.
Explainability is not just a nice-to-have in professional services allocation. It's the difference between a scoring model that gets used and one that sits in a product demo and never influences a real decision.
The firms that have moved furthest in this direction — formalizing chemistry as a staffing dimension rather than leaving it as institutional knowledge — tend to share a common characteristic: they started small. One practice area, one client sector, one manager who was willing to test the approach against their own mental model and compare outcomes. The buy-in built from there, because the results were visible in the practice area's client satisfaction trends before they were visible in any aggregate metric.
