How Many AI Tools Does Your Firm Actually Run?

The counting problem

Firms count the AI they bought. The obligations attach to the AI they run.

Those are different numbers because most AI in an advisory practice was never a purchase decision. It shipped inside something already in place — a CRM that added meeting summaries, a planning tool that added narrative generation, a video platform that added transcription and switched it on by default.

Kitces' research on notetaker adoption describes this pattern directly: an advisor working across several common platforms can have multiple notetaking capabilities available without having selected any of them.

Why the market has no word for this

Because it is not a purchasing problem, and the vocabulary in this market is organised around purchasing.

There is a well-developed language for choosing a tool — comparisons, alternatives, pricing pages, feature tables. There is almost no language for the state of having accumulated several without choosing. So the condition goes unnamed, and unnamed conditions do not get budget, owners or review cycles.

The surrounding data describes the same thing from different angles. Schwab's study found 63% of RIAs using AI with roughly one in ten having integrated it into how the business runs. Advisor360's Connected Wealth Report found 74% of firms not getting full value from the tools they already have. Orion's 2026 survey analysis found the top pain point to be disconnected systems that do not communicate, with 61% naming technology integration a priority.

Those are three measurements of accumulation without integration.

What the accumulation costs

Duplicated capability, fragmented records, and a governance surface larger than the one anyone is watching.

• Duplicated capability. Several systems producing overlapping summaries, none authoritative.
• Fragmented records. The account of a single client relationship split across four vendors with different retention periods.
• Unreviewed vendors. AI features adopted without the diligence a new AI vendor would have triggered.
• Unwatched surface. Policies written for the tools the firm chose, while outputs reach clients from tools it did not.

The compliance consequence is measurable. Against 80% of firms having formally adopted AI tools, 30% had policies addressing third-party AI use, per the 2026 compliance testing survey. AI arriving inside an existing vendor is third-party AI use, and it is the third-party AI least likely to have been reviewed.

How to count properly

1. Start from the vendor list, not from memory. Accounts payable is a better source than a staff survey.
2. For each vendor, ask what AI capability exists and whether it is enabled.
3. Record what data it can reach and whether its output can reach a client.
4. Note whether a human reviews before it does, and whether that review leaves a record.
5. Note whether model version is exposed. Where it is not, record that rather than leaving it blank.
6. Review quarterly. Vendor AI features ship faster than annual policy cycles.

The output is the inventory that makes an AI use policy enforceable. Without it, a policy applies to an undefined set.