Security record
- Read scopes only: deals, contacts, companies, owners, and schemas
- One encrypted token per client
- Every access is logged
- DPA offered
Operator runbook · read-only first
How it works: the nightly engine, owned by a human.
Start with read-only access. We run the nightly method and hand over the outcome, so there is no new tool for your team to babysit.
HubSpot objection
Why not just use HubSpot's own forecasting?
HubSpot can show you a number. It cannot tell you which deals are lying, chase the reps who left the fields blank, or stand behind the forecast in your Monday meeting. Breeze surfaces the data. We operate the method and hand you a reviewed answer. The half you are buying is the human who owns it.
01 / Access model
Your account stays read-only.
You create a HubSpot Private App token, tick the approved read scopes, and share it in a five-minute setup supported by a Loom. OAuth can come later. Write-backs are considered only after trust is earned and explicit approval is given.
02 / Nightly pipeline
A frozen snapshot before any judgment.
Approved HubSpot fields are read once a night into a snapshot. Rules produce the evidence first; constrained AI writes only the judgment layer and never invents a number.
- 01 / nightly pathHubSpotRead-only Private App token
- 02 / nightly pathIngestNightly read of approved scopes
- 03 / nightly pathSnapshotFrozen copy in Postgres
- 04 / nightly pathFour enginesHygiene · Deal-risk · Forecast · Segmentation
- 05 / nightly pathAI judgmentConstrained verdict and narrative
- 06 / nightly pathBrief + SlackHuman-owned handoff every Monday
03 / Evidence engines
Four engines, clear boundaries.
Deterministic checks and reconciliation stay explainable. AI is limited to a deal verdict and the brief narrative; it never invents a number.
01 / Deterministic checks
Hygiene
Flags missing amount, close date, owner, next step, or stage; past close dates; stalled deals; duplicate candidates; and won or lost deals left open.
02 / Rules → constrained AI
Deal-risk
Deterministic signals catch 2+ close-date pushes, going dark, single-threading, and age. AI adds a one-line verdict for human review and never invents a number.
03 / Explainable reconciliation
Forecast
Rep commit, stage-weighted view, category roll-up, and the gap are shown side by side, with no black-box ML.
04 / Deterministic benchmarks
Segmentation
Benchmarks are separated by pipeline, product, type, and segment so unlike cohorts are not blended together.
04 / Human handoff
The weekly operating loop.
01 / cadence
Nightly sweep
Read-only ingestion, snapshot, hygiene checks, and risk signals.
02 / cadence
Friday snapshot
A frozen view preserves what changed and where the number stands.
03 / cadence
Monday brief
Reconciliation, top risky deals, per-rep follow-up, and week-over-week movement arrive in Slack before the call.
Why a service, not a tool
The engines run nightly. A human owns what reaches your call.
Software can surface another dashboard. We operate the method, review the constrained AI judgment, and hand over the brief and follow-up outcome with a named human accountable for it.