Brief every cold call before it starts

Cold calling well takes about ten minutes of preparation per account, which is why almost nobody does it. A rep with forty calls to make has four hundred minutes of research in front of them and an afternoon to make the calls in, so the research collapses into a glance at the company website and an opener that could have been used on any of the forty.

The preparation is not hard work, though. It is a handful of lookups that always run in the same order against sources that are already public: who they are hiring, what they run, what the person has said out loud. Every one of those is an API call, and none of them needs judgement until the last step, which makes cold call preparation a good job for an AI agent.

What goes into each cold call brief

The run starts when a new call list is built in HubSpot, so the briefs are waiting before anyone picks up a phone. For every account on the list, the agent works through the same sources in the same order:

• The account record. Past interactions, earlier call notes, who owns the account, and whether anyone there has already been contacted.
• A dated company signal. Current job postings, especially for the function you sell into; technologies the company runs or is hiring for; recent public activity such as a funding round, an office opening, a product launch or a press mention. These come from TheirStack and Apify.
• The lead's own public words. A conference talk, a podcast, a post, a quote in a trade publication, found by searching the person by name and company with Serper.

Signals are ranked by how recent and how specific they are. A posting from last week beats a funding round from last year. A posting is dated, public, and probably approved by the person who answers the phone. Every signal keeps its URL, because the note is worthless if the rep cannot check it when the prospect pushes back.

What makes AI cold call prep worth reading

The note goes back onto the CRM record rather than into a chat message or a document. It sounds like a trivial delivery detail. It decides whether the brief gets read. The rep already has the record open at the moment they dial, and anything living somewhere else has to be hunted down while a phone is ringing.

Reading the CRM before searching externally catches the failure that embarrasses people. An account that a colleague already spoke to six weeks ago does not need an opener about their hiring, it needs a sentence about where that conversation stopped. Doing the external research first and the internal check second produces a beautifully researched call to someone who has already said no. When the record shows a prior conversation, the note says so at the top.

A brief that pads an empty result with a line about industry trends teaches the rep that the notes are decorative, and after the third one they stop reading any of them, including the good ones. Saying "no dated signal found, open on the problem" costs nothing and keeps the other thirty-nine worth reading.

The lead's words and the company's words stay apart. A press release is the company talking, and the note labels it that way. Only a talk, a post or a quote from the person themselves is presented as their own, since getting that wrong on a call is worse than having nothing.

One opener, under twenty-five words

A cold call opener has about eight seconds to earn the next thirty. The brief carries a single line, under twenty-five words, that names the specific signal and connects it to the reason for the call, written so the rep can read it aloud without sounding like they are reading. Underneath it sits the raw signal, so the rep can improvise when the call goes somewhere else. Three variants would be less useful: a rep scanning options on a screen while a phone rings picks none of them.

Each note follows one structure: the signal and its date, the source URL, what the person has said publicly, the opener, and one line on anything the CRM already knew.

Where it applies

The note is an ordinary call note on the account record, and the prior-contact check is an ordinary read, so the process runs the same way on any connected CRM. At the end of a run it reports how many accounts were briefed, how many had a dated signal, how many had nothing and were flagged as such, and how many already had prior contact worth knowing about.