Choosing health indicators from observed renewal behaviour

Choose health indicators from what could be observed before earlier renewal decisions. Check which observations helped distinguish later outcomes. A plausible signal is only a candidate. A signal recorded after the customer decided is too late to serve as an early indicator.

Define the decision you will compare against

Build a renewal record before analysing signals. Identify the account or contract line, the decision date and the outcome: continued at the same scope, reduced scope, ended or still undecided. Keep a partial renewal distinct from a full loss. If contracts within one account have different dates, decide which decision each snapshot belongs to.

Choose a point before the decision when an account owner could reasonably act. Use the same rule for comparable accounts, such as a defined interval before the relevant decision date. That interval is a design choice, not an industry benchmark.

Reconstruct each candidate indicator as it stood at that point. A later sales note saying the customer would not renew cannot be placed in an earlier snapshot.

Assemble candidate observations

Start with the customer’s intended work and the records available at the chosen snapshot date. Candidates might include completion of the relevant workflow, a customer-confirmed outcome, a material service issue or loss of contact with the person who owns the goal. Define each one by population, period, source, update date and what would count as missing.

For a hypothetical approvals product, the relevant observation might be whether eligible managers completed approvals when requests arose, rather than how often anyone logged in. A relationship observation might be whether the process owner had confirmed that the route still met its needs. Neither proves a renewal decision.

The question is whether it helped the team identify a concern early enough to investigate.

Exclude a candidate if its history cannot be reconstructed reliably. Check for changed event definitions, merged accounts, automated activity and notes entered after the fact. Keep customer statements separate from an account manager’s interpretation. Missing information should remain missing; assigning it a favourable value can distort the comparison.

Compare patterns without overclaiming

Group accounts with broadly comparable product use, contract structure and stage of customer work. For each candidate, inspect later outcomes for accounts with and without the signal. Look at the accounts behind any pattern.

Did the indicator flag customers that continued? Did it miss losses caused by budget or organisational change? Would an account owner have known what to check when it changed?

Avoid choosing a measure solely because it fits one memorable loss. A small group of renewals may show a striking pattern by chance. Larger accounts may also be overrepresented if they receive more meetings and more complete notes.

Record coverage and missingness alongside any comparison, and ask whether the observation behaves differently for new and mature accounts.

Use later renewal periods to check a proposed rule chosen from earlier periods. Keep information learned in the later period out of the design step. If too few comparable decisions exist, use the indicator as a labelled prompt for account review and continue collecting evidence.

Key Considerations in Health Indicator Selection

Keep the indicator only if it improves a decision

A selection record should state the indicator definition, why it might matter, the snapshot date, the outcomes inspected, exceptions found and the action its change would prompt. An indicator may be useful without predicting every loss: it might expose a blocked customer task the team can address.

Conversely, a measure associated with losses but unavailable until the final negotiation has little early-warning value.

Review the selected set after later renewal cohorts and changes in product use or data capture. Retire a measure when it no longer describes the customer’s work or repeatedly sends owners to the wrong question.