Question 1

Can this start a campaign or send a message, or is that only convention?

Accepted Answer

It cannot, and the restriction is enforced in code rather than promised in prose. Campaigns are always created in DRAFT, and the client that reads HeyReach refuses any endpoint outside a read-only allowlist. One setup module owns the only three writes permitted — create a lead list, add leads to a list, create a campaign — and everything else is blocked. `hr setup --dry-run` prints the full plan and writes nothing, which is the command to run before the real one.

Question 2

Where do the scores come from — the model or hardcoded weights?

Accepted Answer

The model produces the judgements and deterministic code turns them into numbers. Each lead gets one request and each message variant another, with the questions inside a request running in parallel. scoring.py combines role, seniority and company fit with signal relevance weighted by signal type and decayed by age, applies disqualifiers, and emits a 0–100 lead score, a message score, a reply probability and a best-variant pick. Signal age, weights and thresholds all live in that one file, readable and editable.

Question 3

What happens if my account has almost no reply history?

Accepted Answer

It falls back to a logistic prior around your base reply rate and tells you so rather than pretending it has learned something. Calibration turns every conversation you started into one example, scored the same way a new lead is scored, with the label being whether the person ever wrote back. Below 150 conversations only the lead score and message score are used; from 150 it tries 19 features and keeps whichever predicts better on held-out conversations. Under five replies or five non-replies it refuses and exits.

Question 4

How much does running the whole thing actually cost?

Accepted Answer

Less than most people expect, and the README prices both workloads. The demo — 12 example leads and 14 first messages, 26 requests — takes a few seconds and costs about $0.0013. Calibrating on 400 past conversations is roughly 800 requests, under a minute and about $0.038. HeyReach's own subscription is the larger cost, since the tool reads and writes a real workspace. Start with the demo: it costs a fraction of a cent and touches nothing in your account.

Question 5

What does the built-in demo run do?

Accepted Answer

It scores 12 example leads and 14 first messages against a made-up offer in a few seconds for about a tenth of a cent, and touches nothing in your HeyReach account.

Question 6

What does the reply probability model need?

Accepted Answer

At least 5 conversations that got a reply and 5 that did not; under 150 it learns from the lead and message scores, and from 150 it also tries 19 features and keeps whichever predicts better on held-out conversations.

Question 7

What does it write back into HeyReach?

Accepted Answer

Per tier and per chosen variant, one lead list and one single-message DRAFT campaign, with each lead carrying custom fields like gtm_tier, gtm_lead_score, gtm_reply_p, gtm_best_variant, and gtm_why.

Heyreach Jev Bot

About this project

Highlights

Questions about this entry

More Jev resources like this

From the Jev guides

From the Jev community