What Should My AI Assistant Log on Every Send or Call?

The minimum fields every touch needs

A usable record of outreach activity isn't a vague note that "an email went out." It needs enough detail that a different person — or the same assistant a month later — can reconstruct exactly what happened without asking you. At minimum that means:

• Who was contacted (name, email or number, and company)
• Which channel (email, call, or another touchpoint)
• Which message or script version was used
• Which step of the sequence this was — first touch, second follow-up, and so on
• When it happened
• The outcome: sent, delivered, bounced, opened, replied, answered, voicemail, no answer, or opted out

Miss any one of these and a record stops being useful for the decisions that depend on it. A log that says someone was "emailed" with no timestamp can't tell you whether it's safe to follow up today or whether you'd be sending two touches in the same week.

Why what was sent matters as much as that it was sent

It's not enough to know a touch happened — you need to know which version of the message it was. If a sequence has more than one subject line or opening being tried, and the record doesn't capture which one went to which contact, you lose the ability to tell later which approach a given reply is actually responding to. The same applies to call scripts: if a call used one opening line versus another, that detail belongs in the record, not just "called, no answer."

Calls need outcome codes too, not just "done"

Email outcomes are fairly standard — delivered, bounced, opened, replied. Calls need their own vocabulary: answered and interested, answered and declined, voicemail left, no answer, wrong number, gatekeeper only. Lumping all of these into "call made" throws away the information that determines what happens next. A wrong number should trigger a correction to the contact record; a voicemail should trigger a specific, limited number of follow-ups; a flat decline should move the contact to suppression, not another attempt next week.

How records turn into decisions, not just history

The reason this record-keeping matters isn't archival. It's that every pacing and suppression decision your assistant makes depends on reading this history correctly:

• Deciding whether a contact has already been touched this week depends on timestamps being accurate and complete
• Deciding whether to try a different angle on a follow-up depends on knowing what was already said
• Deciding whether a lead is worth continued effort depends on the accumulated outcome history, not just the most recent touch
• Flagging a contact for suppression depends on catching an opt-out or hard bounce the moment it's recorded, not after a delay

An assistant without access to this record isn't being careless when it double-sends or ignores a decline — it has no way to know those things happened. The record is what makes its judgment trustworthy rather than a guess.

Why this can't live only in the assistant's memory

A conversation with an assistant doesn't persist the way a durable record needs to. If the history of every send and call exists only as something the assistant "remembers" from earlier in a session, it disappears the moment that session ends, and a fresh conversation starts with no knowledge of what's already happened. The record needs to live somewhere durable that any session, or any person picking up the account, can read from — which is a different thing from the assistant simply being good at recalling recent context.

Keeping that record accurate and current, for every contact, every channel, every day a campaign runs, is the unglamorous part of running outreach well. It is also the part that makes the difference between an assistant giving advice based on what actually happened and an assistant giving advice based on what it assumes happened.

What to take away

• Log who, which channel, which message version, which sequence step, when, and the outcome — for every single touch.
• Call outcomes need their own categories; "called" isn't enough information to act on.
• Suppression and pacing decisions are only as good as the record they're read from.
• The record has to live outside the assistant's conversational memory, somewhere durable that persists across sessions and people.