When the customer base grows, adding people is an obvious way to protect the experience. Sometimes it is exactly the right answer. It is also expensive, and it does not fix a weak operating model.
At IBM I worked on a Digital Customer Success and product-led growth model across 34 software businesses and more than 30,000 customers. At that scale we used high-touch, low-touch and tech-touch engagement so coverage could remain economically viable.
The interesting part was not creating the tiers. It was deciding which customers needed which kind of help, when they needed it and what evidence should trigger a change.
That is a telemetry problem before it is a staffing problem.
Human attention is valuable because judgement is valuable
Good Customer Success people interpret ambiguity. They understand the politics around a customer decision, notice when the stated problem and the real problem differ, challenge assumptions and reshape a plan when the customer's context changes.
Using humans for that work makes sense.
Using them to repeatedly discover information the business already has, send routine reminders or manually stitch together product activity from several systems makes less sense. The goal is not to remove people from the relationship. It is to reserve human attention for the points where judgement, context and trust can change the outcome.
Commercial segmentation is only one dimension
A common starting point is to segment by account value: larger accounts get high-touch coverage and smaller accounts get a lighter or digital model.
That can be sensible, but ARR alone does not tell you how much help a customer needs.
A large customer with a clear outcome, strong adoption and capable internal owners may require less intervention than expected. A smaller but strategically important customer may be struggling through a complicated transition and need more. The same customer may also need different modes of engagement at different points in the journey.
This is where evidence matters. What stage is the customer in? Which outcome are they working towards? What behaviour would indicate progress? Where have they stalled? Is the issue technical, operational, organisational or commercial? Which intervention has a reasonable chance of helping?
Without those states, tiering becomes a static label rather than a dynamic operating model.
Digital Customer Success needs a reason to act
Tech-touch can easily become a collection of email sequences, in-product prompts and automated check-ins. Those can be useful, but they are delivery mechanisms.
A digital motion becomes more valuable when it knows why it is acting.
A customer has not reached a capability that matters to their stated objective. An implementation milestone has passed but expected usage has not started. The account is using the product but has weak evidence of business value. A key stakeholder has changed. An outcome has been achieved and a different journey should begin.
Those are different states, and they justify different actions. The automation is downstream of the understanding.
Test the plays against behaviour
One of the disciplines I valued in the IBM work was testing engagement plays against actual adoption behaviour.
A campaign could produce clicks, responses or attendance and still fail to change the behaviour we cared about. When that happened, the play needed to change or disappear.
That matters because scaled Customer Success can generate a lot of activity. Activity is easy to see. The better question is whether the intervention moved the customer towards a meaningful state.
Four areas tend to connect
In the work I am developing through Routeability, four areas keep appearing together: the customer journey, a common taxonomy, adoption, and automation or AI.
I do not treat them as a maturity checklist. If the taxonomy is already strong, leave it alone. If the journey is clear but people are manually executing predictable work, automation may be the priority. If nobody agrees what value means, automation is premature.
The sequence should follow the evidence.
That changes the capacity question too. Instead of starting with "How many customers can one CSM cover?", I would start with:
Which customer states genuinely require human judgement, and which can be supported reliably through a repeatable digital motion?
Once that is clearer, capacity becomes a design decision rather than a headcount reflex.


