System of record

What it is

Every organisation has systems whose entries are "real": if the appointment is not in the practice-management system it does not exist; if the payment is not in the billing platform it was not made. Those are the systems of record. A voice agent may have its own database of calls, transcripts and "bookings", but that database is a log, not a record. Staff do not look at it; downstream processes do not read it.

Reading from the system of record (availability, balance, order status) is the easier half. Writing to it, correctly and with the right permissions, is the half that makes the agent useful.

Why it matters when buying

Two reasons. First, outcomes: a booking that exists only in the vendor's dashboard has to be re-keyed by a person, which is a message-taking service at agent prices. Second, measurement: containment and accuracy KPIs should be computed against the system of record, not against the agent's own claims about what it did. If a vendor cannot show you the record it wrote, the KPI is unverifiable.

What to ask

Ask which of your systems the agent reads from and which it writes to, through which interface (official API, partner integration, screen automation), and with what account and permissions. Ask for a demo booking or update that you can see in a sandbox of your own system during the call. Ask how discrepancies between the agent's log and your system are detected and reported.