If you look at how SaaS customer success tools define account health, the model is almost always the same: build a composite score from usage metrics, support ticket volume, NPS, and contract value, then weight them according to whatever the CS team's intuition suggests matters most. The score goes green, amber, or red. You work the red accounts.
For SaaS companies, this approach has genuine merit. Login frequency tells you something real about whether customers are getting value from the product. Feature adoption tells you whether they have successfully integrated your tool into their workflows. These are legitimate proxies for retention risk.
For agencies and consultancies, the same model largely fails. And the reason it fails is not that agencies are doing it wrong. It is that the underlying value delivery model is fundamentally different, and the metrics need to reflect that difference.
What SaaS health scores actually measure
SaaS health scores are fundamentally usage-based. They measure how much of the product the customer is using and whether that usage is increasing or decreasing. This works because in SaaS, value delivery is mediated by the product itself. If customers are not using it, they are not getting value. If usage drops, risk increases.
The key assumption embedded in this model is that you can observe value delivery through product telemetry. Login, feature activation, workflow integration: these generate data. They are measurable inputs to a health model.
In an agency or consultancy context, value delivery is primarily relational and judgmental. The client hires you for your expertise, your perspective, your ability to think about their problem. The deliverables are outputs of that relationship, but the deliverables alone are not the relationship. A perfectly executed campaign that the client feels disconnected from is not a healthy account. A messy project that generated genuine strategic partnership often is.
There is no telemetry for this. The data points are qualitative: how the client talks about the work, what questions they ask, whether they are bringing you into conversations that matter to them.
Why composite scores flatten the signal
The standard approach to building an agency health score is to take the metrics that are available (satisfaction surveys, email response rate, deliverable approval time, contract renewal history) and create a weighted composite. You adjust the weights based on which metrics seemed to predict churn in past non-renewals. The score becomes a number, and that number gets red-amber-greened.
There are a few practical problems with this. The first is that the metrics available to agencies are largely lagging indicators. Satisfaction survey scores drop after something has already gone wrong. Approval time increases when the relationship is already transactional. These metrics confirm a problem; they do not predict one.
The second problem is that weighting schemes are not stable across client types. A strategy consultancy client who routinely takes two weeks to approve a deliverable because their internal processes are slow is not a risk signal. A brand agency client who used to approve in 24 hours and now takes two weeks is. The same number means different things on different accounts. A single composite score cannot hold that context.
The third problem is that the most important signal, which is the relational and conversational quality of the engagement, is not being measured at all. The composite score is measuring shadows of the relationship: the residue of interactions in the form of response times and satisfaction scores. Not the interactions themselves.
What works better: signal tracking over scoring
Rather than a single composite health score, the approach that tends to work better for agencies is tracking a set of distinct signals that can move independently of each other, and surfacing changes in those signals rather than changes in a composite.
This distinction matters operationally. A single composite score hides what is moving. If your health score drops from 82 to 71, what drove it? Was it the satisfaction survey? Was it email response time? Was it something in the meeting notes? A composite does not tell you. You have to go back to the raw data to diagnose.
Separate signal tracking means you know immediately: "Verbatim quality on this account has been declining for three weeks, but the satisfaction score is still green and response time is normal." That tells you something specific. The client may be attending the motions without being fully present. That is an investigable hypothesis. You know what to bring to the next conversation.
The role of qualitative data in a signals model
The limitation of separate signal tracking for most agencies is that the most informative signals are qualitative, and qualitative data does not score itself. Survey verbatims, meeting notes, and email threads require interpretation. At a portfolio scale of 40 or 50 accounts, manually reading and annotating all of this is not realistic.
This is the specific problem Avara addresses. Not replacing account manager judgment, but making it possible to extract structured signals from qualitative data at portfolio scale. The meeting notes your team already writes, the survey verbatims your clients already submit: the information is there. The challenge is reading it consistently across every account, every week, without that becoming a full-time job for someone.
We would not argue that a qualitative signal model is complete or that it eliminates false positives. It reduces the accounts you need to look at closely. Instead of reviewing 40 accounts in detail every week, you review the 6 where the signals have shifted meaningfully. That is a practical improvement even if the model is imperfect.
Building a health model that fits agency value delivery
A health model that actually works for agencies needs to be built around the nature of the value being delivered, not borrowed from SaaS. That means taking qualitative sources seriously as data inputs, not just supplementary context. It means tracking trends over time within each account, not comparing accounts to an absolute threshold. It means giving account managers signals to investigate, not verdicts to act on.
The analogy is useful: a good health model for agencies should work like a good editor, not like a calculator. It should tell you what is worth looking at and give you enough context to decide what to do. It should not pretend that a single number captures the state of a complex human relationship.
If you are thinking about what this looks like in practice, the product walkthrough explains how Avara structures signals from your qualitative sources and surfaces what has changed this week.