The Operational Diagnostic

What it is for

To establish, with evidence rather than impression, where a firm's operating capacity goes and which of its systems hold records nobody has classified.

Most firms can describe their tech stack. Fewer can say which AI capabilities are active inside it, and fewer still can say which of those produce outputs that may be records. The diagnostic exists to turn those into answered questions.

The four parts

Part one · Where the time goes

A structured account of advisor and operations time across a representative two-week period, categorised by activity.

Categories: client-facing conversation, meeting preparation, post-meeting documentation, CRM maintenance, compliance activity, portfolio work, reporting, internal coordination, and administration. The output is a distribution, not a total — the useful finding is usually the ratio between client-facing time and the documentation tail behind it.

The context for this is well established. Schwab's RIA AI adoption study found 63% of RIAs using AI, with roughly one in ten having integrated it into how the business actually runs. Time measurement usually explains that gap: tools were adopted alongside the existing process rather than replacing any part of it.

Part two · Which system holds which record

A map from record category to system of record, with the gaps and duplicates marked.

For each category — client communications, suitability documentation, meeting records, advice rationale, compliance attestations, portfolio decisions — establish where it lives, who can retrieve it, and how long it is kept. Two findings recur: categories held in more than one system with no authoritative copy, and categories held in none.

Part three · The AI inventory

Every AI capability active across the firm's existing tools, whether or not it was adopted deliberately.

This is usually the part that surprises. Firms typically identify one or two AI tools by name and then discover several more switched on inside a CRM, a planning tool, a custodian portal and a meeting platform. Kitces' research on notetaker adoption describes exactly this pattern of AI arriving inside tools adopted for other reasons.

For each: the vendor, the capability, the data it can reach, whether outputs reach clients, whether a human reviews before they do, and whether model version is exposed.

Part four · Controls, written versus practised

For each governance control, whether it exists in writing, whether it is practised, and whether the practice leaves a record.

Three columns, because they diverge. The survey benchmark is useful here: against 80% AI adoption, 48% of firms had human-in-the-loop oversight policies, 37% had output validation procedures, 30% had third-party AI policies and 14% had AI-updated incident response, per the 2026 compliance testing survey. A firm can locate itself against those numbers.

What comes out

A distribution of time, a record map with its gaps named, an AI inventory with governance status per entry, and a control matrix against the market benchmark.

The finding that recurs most often is not that a firm is doing something wrong. It is that the firm has more AI in production than it has reviewed, and the review question was never assigned to anyone.