When I was doing partnership work at an enterprise software company, the account health model was straightforward: product usage, license utilization, support ticket trends, and NPS. Each input got a weight, the weights produced a score, the score determined who the customer success team called first. It was imperfect but directionally useful because in that context, how much the customer uses the product is a genuine proxy for whether they are getting value.
When I moved into work that was closer to agency and consultancy operations, I carried that model forward by instinct. It turned out to be largely the wrong model for a different kind of value delivery, and the reasons why matter for anyone trying to build account health monitoring for a professional services firm.
The fundamental difference in value delivery
SaaS delivers value through a product. The customer gets access to software, they use it, and the extent to which they use it and integrate it into their workflows determines whether they are getting the value they paid for. Usage is a reasonable proxy for value realization because the delivery mechanism and the value mechanism are both the product.
Professional services deliver value through people, judgment, and relationships. The client hires an agency or consultancy for expertise, perspective, and execution that they cannot or do not want to provide internally. The deliverables are tangible outputs of this, but the deliverables are not the full value. The value also includes the quality of the thinking behind the deliverables, the client's confidence in the team's judgment, the degree to which they feel the team understands their business, and whether the relationship creates better outcomes than the client could achieve with a different provider.
None of this is measured by usage data. There is no product telemetry for "client confidence in team judgment." It lives in conversations, in how the client talks about the work in meetings, in whether they bring the consultancy into discussions that go beyond the contracted scope, in the texture of the relationship over time.
Three SaaS health metrics and why they fail in professional services
Feature adoption / product usage. Not applicable in the obvious sense, but there are agency equivalents that people try to use: deliverable review rates, portal logins if the agency has a client portal, email response times. These are operational engagement metrics, not relationship health metrics. A client who reviews every deliverable promptly and logs into the portal weekly can still be internally disengaged from the strategic partnership while going through the operational motions. The metrics tell you the process is running; they do not tell you about the relationship.
Support ticket volume. In SaaS, high ticket volume can indicate poor product fit or onboarding issues, and low ticket volume in a well-onboarded account often signals healthy self-sufficiency. In professional services, there is no direct equivalent, and attempting to use complaint rates or revision request frequency as proxies for health runs into significant problems. A client who requests many revisions might be highly engaged and invested in quality. A client who never requests revisions might be disengaged. The direction is not interpretable without context.
NPS and satisfaction scores. These are more useful than the above but still have structural limitations in professional services. In SaaS, NPS primarily measures whether customers are satisfied enough with the product to recommend it. In professional services, NPS and satisfaction scores measure something more complex: the client's retrospective assessment of the relationship and delivery quality. This assessment can stay stable or even improve while the client has already internally decided not to renew. The score reflects a momentary satisfaction assessment, not a forward commitment to continue the engagement. It is a lagging indicator by design.
What professional services health actually looks like
If the SaaS metrics do not transfer, what should a professional services health model track instead? The answer is: the qualitative layer of the relationship, captured from the sources where it actually lives.
Relationship quality in professional services is primarily observable in three places. Meeting content: what topics does the client raise, what questions do they ask, how specific and forward-looking is their engagement with the work. Survey open-text responses: how detailed, specific, and genuinely invested do verbatim responses look over time, and how has that quality changed. Communication patterns: is the client initiating conversations that go beyond operational necessity, and is the tone of those communications partnership-oriented or transaction-oriented.
These are qualitative sources. That is the challenge. Structured quantitative data is easy to aggregate and display in a dashboard. Qualitative text requires interpretation, and manual interpretation at portfolio scale is not feasible for most boutique or growing firms.
The NLP layer that makes this tractable
The reason this problem has historically been difficult is that reading and interpreting qualitative text at scale requires a level of analytical resource that most professional services firms do not have. A 15-person strategy consultancy is not going to hire a data analyst to read meeting notes and surface relationship signals.
NLP-based extraction changes that. The same NLP approaches that power document classification, sentiment analysis, and entity extraction in other contexts can be applied to the specific extraction task of reading meeting notes and survey verbatims for the signal types that matter for professional services account health: specificity levels, question presence, forward-looking language, topic engagement quality, tone register.
This is the problem Avara is built to solve. Not applying generic sentiment analysis to meeting notes, but running structured extraction for the specific signal types that are predictive of relationship health in agency and consultancy contexts. The output is not a score that tells you whether an account is healthy. It is a pattern surface that tells you what has changed in the last four to six weeks in the accounts you manage.
What this model does not replace
Qualitative signal monitoring can tell you which accounts to pay attention to and give you a hypothesis about what might be happening. It cannot tell you why a relationship is cooling or what to do about it. Those questions require account management experience, relationship knowledge, and direct conversation that no system will substitute.
We are not proposing that agencies automate account management. We are proposing that they stop relying on metrics that were designed for a different business model, and start building the qualitative signal monitoring that their business model actually requires. The account manager still does the work. The tool makes it possible to focus that work on the accounts and signals that most need it.
If you want to understand what this looks like in practice, the product page explains how Avara extracts from your existing notes and surfaces what has changed, without requiring you to change how your team works.