Reference
Agentic CRM glossary
Plain-language definitions of the terms used in agentic CRM and AI-assisted sales.
Agentic CRM
A CRM with AI agents inside it that read context, make a judgement and propose an action, instead of only storing what people type.
AI agent
A program that sends a language model a task and a set of tools, runs the tools the model asks for, and repeats until the task is done. Its path is not drawn in advance, which is why its reach must be bounded.
Human in the loop
A design where a person approves an AI system's output before it has any effect on real data or real people.
Approval gate
The one step through which an agent's proposal can change records. In Tolva, nothing an agent writes reaches a contact, deal or inbox until a person approves.
Lead research brief
A short summary of a contact's company, written by an agent from web sources, with a fit score and a suggested next step.
Fit score
A 0 to 100 estimate of how well a contact matches the customer you want, proposed by the research agent and approved by a person.
Engagement score
A score built from what a contact did with your emails. In Tolva a click counts for more than any number of opens.
Email sequence
A series of emails and delays that a contact moves through automatically after enrolment, until they finish, reply, bounce or unsubscribe.
Reply detection
Recognising that a person answered an email so that every automated sequence for them stops. Email services report opens and clicks, but not replies, so the reply itself has to be received.
Suppression list
The addresses that must never be emailed again, built from unsubscribes, hard bounces and spam reports.
Stale deal
An open deal with no email, call or meeting for 14 days.
Agent spend cap
A monthly budget for AI runs that is checked before every run. When it is reached the agents stop and say so.