Post-Investment Metrics That Compute Themselves

Whether a post-investment metric survives comes down to one test: do both the numerator and the denominator come from business objects that already exist in the system. If they do, the metric updates as a by-product of work being done. If they do not, somebody must key the numbers in on a schedule, and manual entry always loses to a reporting deadline. The usual failure is not a badly designed metric. It is a metric designed without asking which action produces the value. When the honest answer is “the portfolio manager remembers to fill it in,” the metric becomes a spreadsheet inside a year.

Only four families compute themselves

The first is capital actions: paid-in progress, unfunded commitments, ownership changes, dilution ratio. Their source objects are closing records and registry filings; the action completing is the data landing. The second is accounting consistency: cost basis, carrying value, distributions received, exit proceeds — sourced from payment and receipt records. The third is process timeliness: on-time collection of periodic reports, completeness of resolution files, appraisal reports on file — all timestamps on workflow steps. The fourth is disclosure: LP report send dates and query response times, taken from portal logs.

Metric family	Source object	Automation verdict	Typical failure
Capital actions	Closing record, registry filing, agreement terms	Fully automatic	Committed and paid-in share one field; progress looks inflated
Accounting consistency	Payment and receipt records	Fully automatic	Silent manual journal adjustments; valuation will not tie
Process timeliness	Approval and archive timestamps	Fully automatic	Back-dated entry presented as completion
Operating health	Portfolio company reporting or interface	Semi-automatic; depends on external reporting	Collection gaps hidden behind averages
Strategic contribution	No system object	Inherently manual	Forced into a score, then never reviewed

The half that depends on outside reporting

Operating-health metrics — revenue, gross margin, order backlog, R&D intensity — only exist because portfolio companies send data. Their real metric is therefore collection coverage. Two rules hold here: fix the reporting definitions before defining any metric, and treat the gap itself as a metric. Inconsistent statement formats are the norm across portfolio companies. Catalpa lists financial statement recognition as its first production AI scenario in published material; the value sits not in recognition accuracy but in converging many formats onto one definition table. Until account mapping is finished, no operating metric is comparable (see AI agents and intelligent risk).

Run one test before going live with an externally sourced metric: over three consecutive periods, count how many entities require a manual fill for the same field. If the fill rate settles into a normal band, the metric is valid. If it does not, the metric measures collection capability, not portfolio quality.

Metrics that will always be manual: what to do instead

Some figures have no system source and should not pretend otherwise: strategic synergy value, local investment attracted, employment and tax contribution, technology transfer items, board agenda influence. Their correct role is a disclosure field, not a performance metric. Collect them because an agreement or a supervisory template asks for them, keep the narrative and supporting file attached, and exclude them from ranking and alerting.

Turning them into metrics costs something specific. Once a figure is ranked, portfolio managers optimise the number rather than the fact. Once it is alerted, the threshold has no calibratable denominator, so the first false trigger produces a request to switch it off.

Four attributes every metric must carry

Before a metric enters the system, send it back if any of these are missing:

• Formula: numerator, denominator, exclusions, currency and unit, written as one repeatable line.
• Source object: name the action that produces the value, not the department that “provides” it.
• Owner: who approves a definition change, who handles the exception. A metric without an owner has nobody accountable for its accuracy.
• Closing action: what happens after a trigger, who confirms closure, within how long. A metric with no closing action produces noise that still leaves an audit trail.

The ceiling on metric count is set by review capacity

Coverage is not the same as quantity. Catalpa publishes a risk monitoring design built on 18 alerting indicators; what matters in that design is not the number but the fact that each indicator carries a trigger definition and a downstream action. The number of exceptions a team can genuinely process each week depends on available review hours, not on how many thresholds the software allows.

Take this list into the next metric review:

• Sort every current metric into the five families above, and demote anything “inherently manual” that is currently alerting.
• Write the source object for each numerator and denominator. Delete any metric where you cannot.
• Count last quarter’s triggers and the share confirmed as real risk by the business. Retune or retire indicators with many triggers and low confirmation.
• Assign an owner and a closure deadline to every surviving metric, recorded in the role description rather than a system note.
• Define an annual definition review, with version numbers and effective dates retained.

When we advise against building a metric system

Three cases. First, the organisation holds minority stakes without board seats or contractual reporting rights; every operating metric then rests on the counterparty’s goodwill, so the collection agreement comes first. Second, a post-investment team under five people that also carries fundraising and LP reporting; a full metric system converts them into data-entry clerks, so start with capital actions and accounting consistency only. Third, a fund already in liquidation, where the only metric worth keeping is recovery progress against distribution dates.

Maturity is not demonstrated by count. Three to fifty metrics that update themselves, each with an owner and a closing action, and an exception volume matched to review capacity, come closer to what “portfolio visibility” is supposed to mean than two hundred metrics that require maintenance.