Agulha Digital — Learning from Churn

Overview

When a customer leaves, the usual question is why did they leave? In this project I changed the question to when did the value stop being evident? — and looked at customers who left or did not move forward as a source of learning, not just a loss.

I looked at two moments in the customer lifecycle: a deal breaker, when a customer tries the product and decides not to continue, and churn, when a customer has used the product and then stops.

This is a public, anonymised version of the case. Accounts are not named, customer statements are paraphrased, and no internal figures are shown.

The Problem

Lost deals and cancellations were seen as separate, one-off events. Each one had its own story, but there was no shared reading of what they had in common — and no way to notice risk while there was still time to act.

When did the value stop being evident — and what could we have noticed earlier?

My Role

I designed and ran the analysis: I collected and triangulated the evidence, analysed usage data, wrote the reading of each case and turned the patterns into recommendations for the team.

Approach

The sample was small on purpose: two deal breakers, one churned customer and one account under observation. Because of that, every finding is treated as a signal or hypothesis to validate, not as a proven cause.

I combined three sources:

• Voice of the customer. Direct feedback and follow-up contacts, mostly via WhatsApp.
• Relationship history. The context of each pilot, onboarding, negotiation and cancellation.
• Usage data. Documents generated and templates used over time.

Limitation: few customers answered after their decision, so the analysis relies on triangulation rather than in-depth interviews.

Evidence & Insights

• Learning after the decision is hard. Once a customer has left, there is little reason for them to keep talking about the product. The moment of churn may be too late to start learning.
• Account A (deal breaker): the new process did not feel faster. The pilot started with a complex document the team already produced by hand. Filling in the new form took about as long as typing it, so the effort of changing was not matched by a perceived gain in efficiency.
• Account B (deal breaker): a solution with little chance to be used. Complex templates needed several rounds of alignment, for a use case that turned out to be rare. The effort of the pilot was high compared with how often the case actually occurred.
• Account C (churn): churn started before the cancellation. Usage declined gradually — fewer documents and fewer active templates — long before the formal cancellation.
• Account D (under observation): the same pattern, visible in advance. Falling volume and shrinking template use were already visible, without any cancellation. Usage data can point to risk before a customer says anything.

• Combined signals matter more than single metrics. A drop in volume together with fewer active templates says more than either indicator alone. Template activity — created, changed and actually used — works as a sign of how alive an account is.

Decisions & Experiments

Pilots: reduce the cost of discovering value Proposed

1. Start with a use case that recurs often enough in the customer’s routine.
2. Start with low investment, using existing capabilities wherever possible.
3. Define success before building: usage, volume, perceived efficiency and the people involved.
4. Scale customisation only as evidence of demand and value appears.
5. Define the move to a paid plan from the start: duration, validation criteria, price and conditions to continue.

Retention: notice change earlier Proposed

• Track usage (documents generated, frequency, volume trend) and adoption (active users, concentration of use, templates created, changed and used).
• Treat a persistent drop in volume, fewer templates in use, no changes for a long time — and above all several indicators falling together — as signals for a conversation.
• Compare each customer mainly with its own history, building a baseline per account to separate normal variation from risk.

Product opportunity: which configuration steps could customers do on their own, so pilots depend less on custom work? Proposed

Outcomes & Learnings

The analysis gave the team a shared reading of four cases, three moments where it can act — during the pilot, at the move to a paid plan, and after closing — and a set of signals to watch.

• The best moment to learn from churn is before the customer cancels. A churn survey helps to understand an exit; usage signals create the chance to act before it.
• Small samples call for humility. Being explicit that these are signals, not causes, made the findings more useful, not less.
• There is no universal risk threshold. Accounts differ in size and rhythm, so each one is best compared with itself.

Next Steps

• Check the signals against more accounts, including customers who stayed.
• Set up simple monitoring of combined usage signals per account.
• Apply the pilot criteria to the next pilots and compare the results.