North Star IntelligenceThe layer underneath EchoCody

Conversations are cheap. Verified outcomes are not.

EchoCody is the conversation engine — the voice that answers, qualifies and books. North Star Intelligence is the governed layer beneath it: the pipeline that moves an opportunity, the store that remembers why, and the brief that tells an owner what it was worth. One system, two engines, one standard applied to both.

Request an operating review (801) 980-0308

01Conversation-to-Outcome Pipeline

Every step has an end result and a failure route.

A conversation that cannot be written back as a verified outcome is an exception, not a success. Nothing in this pipeline is allowed to end in silence.

Signal

An inbound call, reply, form or scheduled cadence step arrives with a source tag.

If it fails
Unroutable signals are quarantined with the raw payload retained for replay.

Context load

Current CRM state is reloaded before anything is said. Local memory is never authoritative.

If it fails
If CRM state cannot be read, the agent holds and an owner task is created.

Guardrails

Terminal states, DND, consent, channel and local calling hours are evaluated first.

If it fails
Any guardrail hit suppresses engagement and records the reason.

Conversation

Short acknowledgment, one accurate statement, exactly one question — until the next step is clear.

If it fails
Low confidence or hostility hands the conversation to a person.

Disposition

The turn is classified against the rubric, not against sentiment.

If it fails
An unclassifiable turn becomes an exception with a named owner.

Stage write

One authorized controller moves the opportunity using compare-and-set plus an idempotency token.

If it fails
A failed compare-and-set retries once, then raises a conflict exception.

Next action

Every non-terminal opportunity leaves with exactly one owner, one action and one due time.

If it fails
A missing owner or due time surfaces in the command center as a leak.

Receipt

The event, evidence and actor are written to an audit trail with redacted logs.

If it fails
A missing receipt blocks the outcome from counting in the value brief.

02Intelligence store

What the platform remembers — and who owns the truth.

Agent-local memory is never authoritative. Each domain names the system of record it must defer to when the two disagree.

Conversations

Transcripts, timing, dispositions and escalation reasons.

Source of truth · Voice and messaging platform events

Opportunities

Stage history, owner, next action and due time.

Source of truth · GoHighLevel

Consent

Recording notice, opt-outs, DND and channel restrictions.

Source of truth · GoHighLevel

Artifacts

Generated reports, versions, QA verdicts and delivery receipts.

Source of truth · Artifact store

Experiments

Hypotheses, canary assignments, measured results and approvals.

Source of truth · Change log

Economics

Recognized revenue and actual campaign cost, kept apart from projections.

Source of truth · Closed-won records and spend exports

03Executive Value Brief

Six questions. No number without a source.

The brief an owner actually reads. A section with no verified evidence renders as missing — never as a zero, never as an estimate dressed up as a result.

A
What did the system do this week?
Requires: Signed conversation events · Disposition records
B
What did it produce?
Requires: Booked appointment records · Opportunity stage history
C
What did it prevent?
Requires: Unanswered-call baseline · Recovered-contact records
D
Where is money leaking?
Requires: Stage entry and exit counts · Owner and next-action fields
E
What changed since last week?
Requires: Prior period brief · Approved change log
F
What should we do next?
Requires: Ranked leakage stage · Named owner

04Supervised learning loop

It improves on your calls. A person still signs the change.

Eleven steps from a captured conversation to a promoted behavior change. Automatic promotion is not available — a measured win with no human approver stays in the canary.

01 · Capture
Every conversation is stored with its transcript, disposition and timing.
02 · Label
Outcomes are labeled against the disposition rubric, not against sentiment.
03 · Detect
Recurring failure patterns are clustered by stage, objection and vertical.
04 · Hypothesize
Each pattern becomes one written hypothesis with a predicted direction.
05 · Draft
A behavior change is drafted as a diff against the approved script.
06 · Review
A human reads the diff for compliance, accuracy and brand boundaries.
Human gate
07 · Canary
The change runs on a small slice of traffic behind a kill switch.
08 · Measure
The canary is measured against the control on the predicted metric.
09 · Decide
Promote, iterate or discard, recorded with the evidence that decided it.
Human gate
10 · Promote
An approved change is versioned and rolled out with a rollback point.
Human gate
11 · Audit
The change log, evidence and approver are retained as a receipt.

05Compounding growth loop

Five phases, each with a gate you can check.

No phase advances on enthusiasm. Each one exits on a condition an operator can verify in the command center.

01 · Deploy
One agent answering real traffic in one location.
Exit gate: Fifty governed conversations with clean receipts.
02 · Instrument
Every outcome traceable to an event and an owner.
Exit gate: All six brief sections rendering from verified sources.
03 · Diagnose
Find the single stage losing the most opportunities.
Exit gate: One ranked leak with a named owner.
04 · Intervene
One approved behavior change against that leak.
Exit gate: A canary measured against a control.
05 · Expand
Apply what held up to the next workflow or location.
Exit gate: The same standard running in two places.

06Client Launch System

From paid order to governed launch, in one deterministic chain.

Nine nodes. Each one names the owner, the guard that must hold, the end result, the evidence that proves it, and the route taken when it fails. Nothing advances on goodwill.

Verified paid order

Owner: North Star revenue operations
Guard: Payment settled and the signing entity matches the legal data owner on the agreement.
End result: One order record with the client legal entity, scope, enabled bots and billing terms.
Evidence: Settled payment reference, countersigned agreement, order ID.
If it fails: No provisioning starts. The order is held and an owner task is raised to reconcile entity or payment.

Isolated sub-account

Owner: Implementation lead
Guard: Exactly one sub-account per client legal entity. No shared tenancy, no reuse of a prior client's location.
End result: A dedicated CRM sub-account created under the client's legal data ownership, with least-privilege access.
Evidence: Sub-account ID mapped to the order ID, access roster with roles.
If it fails: Provisioning aborts if an existing tenant collides; the collision is resolved before any data is written.

Pinned master snapshot, installed disabled

Owner: Platform engineering
Guard: Only a protected, version-pinned master snapshot may be installed, and it installs with every automation disabled.
End result: The tenant holds a known-good clone at a recorded version, inert until certification passes.
Evidence: Snapshot version hash, install receipt, disabled-state check on every workflow.
If it fails: Any enabled automation found post-install is disabled immediately and the install is re-run from the pinned version.

Brand, vertical, cost and access configuration

Owner: Implementation lead with client approver
Guard: Vertical boundaries and refusal rules are set before any copy is written; the cost register is signed.
End result: Configured brand voice, approved claims, hard boundaries, owner routing and a funded cost register.
Evidence: Signed configuration record, boundary list, cost register with client-funded lines.
If it fails: Missing approval or unfunded cost line blocks the next node and creates a dated client task.

Channels connected

Owner: Implementation lead
Guard: Phone, email, calendar, voice and webhooks each pass a live round-trip test with signature verification.
End result: Every channel verified end to end in the client's own accounts, with signed webhook delivery.
Evidence: Round-trip test log per channel, webhook signature verification receipt.
If it fails: A failing channel stays disabled; the bot depending on it cannot be certified or enabled.

20-scenario certification per enabled bot

Owner: QA reviewer (not the builder)
Guard: All 20 deterministic scenarios must pass, including STOP, wrong person, duplicate webhook, reply interrupt, manual control, vertical restriction and tool failure.
End result: A signed certification record per bot, tied to the pinned snapshot version.
Evidence: Scenario run log with pass/fail per case, reviewer identity, version stamp.
If it fails: Any failure blocks enablement. The defect is fixed, the version is re-pinned, and the full set is re-run.

Acceptance and funding

Owner: Client approver
Guard: The client accepts the certified configuration and confirms usage funding is in place.
End result: Written acceptance plus a funded balance for software, telecom, AI and approved third-party usage.
Evidence: Acceptance record, funding confirmation, usage alert thresholds acknowledged.
If it fails: Without acceptance or funding the tenant stays disabled — nothing dials, nothing sends.

Capped first-ten-conversation launch

Owner: Implementation lead with QA reviewer
Guard: A hard cap of ten live conversations, each reviewed against the rubric before the cap lifts.
End result: Ten reviewed live conversations with dispositions, stage writes and any corrections applied.
Evidence: Ten transcripts with reviewer notes, stage-write receipts, correction log.
If it fails: A material defect in the first ten pauses outbound, reverts to the prior pinned version and re-opens certification.

Executive Value Brief and governed optimization

Owner: Account operator with human approver
Guard: No behavior change is promoted without sample size, measured lift, a reviewer, an approver and a compliance check.
End result: A recurring brief of verified outcomes plus approved, versioned improvements.
Evidence: Brief with per-section sources, experiment log, promotion approvals.
If it fails: A section without verified evidence renders as missing, and an unapproved win stays in the canary.

07Cost rule

You own the usage. We own the work.

No markup, no resold spend, no surprise line item. Every consumable is funded by you and visible to you; our fee covers the people and the process around it.

Client owns or prepays

Software and platform subscriptions
Phone numbers and call minutes
SMS and messaging fees
Email sending and deliverability services
AI model usage
Domains and hosting
Media, assets and licensed content
Approved third-party tools

Advertising spend runs on the client's own card, inside the client's own ad accounts. North Star never fronts, marks up or resells ad spend.

North Star fees cover

Implementation and configuration
Day-to-day management
QA and certification
Reporting and the Executive Value Brief
Governed optimization
Expansion into new bots and verticals

50%
Notify the client's billing owner with the current burn rate and the projected depletion date.
75%
Second notice plus a recommended top-up amount sized to the next 30 days.
90%
Escalate to the account operator; a dated funding task is created and non-critical batch sends are throttled.
100%
Pause nonessential outbound only. Inbound answering, compliance handling, safety escalation and human routing keep running.

Still running at depletion

Inbound calls and replies are still answered
STOP, DND and consent handling stay active
Safety and emergency escalation stays active
Handoff to a human owner stays active

08Capability matrix

Why this is more than white-label GHL.

The platform layer is genuinely strong, and we use it as intended. What we add is the cement: pinning, isolation, certification, evidence and failure routes.

Platform market capabilities compared with the North Star operating standard
Capability	Platform market strength	North Star standard
Packaging and checkout	Native SaaS mode checkout, plans and client billing are built in and work well.	We use it as-is, and attach a signed order record that gates provisioning until payment and legal entity match.
Snapshots and templates	Snapshots and niche templates make a new build fast to stand up.	Every install comes from one protected master snapshot at a pinned version, cloned disabled by default.
Tenancy	Sub-accounts separate clients cleanly.	One sub-account per client legal entity, no reuse, least-privilege access roster, and clean offboarding with an export.
Usage rebilling	Usage rebilling for calls, SMS, email and AI is supported natively.	A client-funded cost register with 50/75/90/100% alerts, and depletion pauses nonessential outbound only.
Pipeline movement	Pipelines, stages and automations move records.	A single authorized stage controller with compare-and-set writes, idempotency tokens and an evidence trail per move.
Reporting	Dashboards and attribution reporting are available out of the box.	Actual and modeled are never mixed: verified outcomes carry sources, and anything unverified renders as missing.
Failure handling	Workflows can branch and retry.	Every node declares an explicit failure route, an owner and a dated next action — nothing ends in silence.
Conversation boundaries	Prompts and knowledge bases shape agent behavior.	Hard vertical boundaries and refusal rules that bypass sales engagement, enforced outside the prompt.
Release control	Builds can be updated at any time.	20-scenario certification per bot, a reviewer who is not the builder, and version-pinned rollback.
Improvement	Teams iterate on prompts and workflows continuously.	A supervised learning loop: promotion requires sample size, measured lift, reviewer, approver and compliance sign-off.
Support and enablement	Strong support, knowledge base and demo resources across the ecosystem.	We lean on those resources and add an operating standard on top of them, not in place of them.

This compares our own operating standard to capabilities the platform market openly documents. We do not audit private builds and make no claim about how any specific agency configures theirs.

HighLevel HighLevel support docs HighLevel marketplace

09Report registry

One category. One name per vertical.

Every report we produce is an Intelligence Brief. These are configurable templates — no customer data, no sample figures, nothing generated until the listed sources are connected.

11 report profiles12 activation profilesIndy Clover runs qualification and re-engagement

Indy Clover

Indy Clover Market Intelligence Brief

Whether a candidate market and operator profile clear qualification.

Requires: Candidate qualification record, territory data, re-engagement history.

Template · not live data

Business Development

Franchise Development Intelligence Brief

Whether a development conversation should proceed to a discovery day.

Requires: Franchise development intake, financial disclosure receipt.

Template · not live data

Legal

Legal Client Intake Intelligence Brief

Whether the matter is in scope and conflict-free before an attorney call.

Requires: Structured intake fields and a conflict-check result. Narrative detail is collected in a secure conversation, never on an ad form.

Template · not live data

Solar

Solar Project Intelligence Brief

Whether the roof, ownership and utility profile support a site visit.

Requires: Homeowner qualification fields, utility profile, appointment record.

Template · not live data

Med spa

Med Spa Patient Growth Intelligence Brief

Which treatment interest justifies a consultation slot this week.

Requires: Consultation bookings and outcome labels. No clinical narrative is requested on an ad form.

Template · not live data

HVAC

HVAC Revenue Recovery Intelligence Brief

Which missed and unbooked demand to recover first.

Requires: Call events, job bookings and closed-won records.

Template · not live data

Roofing

Roofing Project Intelligence Brief

Which inspections and claims to route to a crew first.

Requires: Inspection requests, claim status and job records.

Template · not live data

Plumbing

Plumbing Service Intelligence Brief

Where dispatch capacity is losing containable emergencies.

Requires: Triage events, dispatch records and completed jobs.

Template · not live data

Dental

Dental Patient Access Intelligence Brief

Where access constraints turn demand into no-shows.

Requires: Appointment, held and recall records. No clinical detail on an ad form.

Template · not live data

Real estate

Real Estate Conversion Intelligence Brief

Which buyer and seller conversations deserve agent time this week.

Requires: Separation labels, appointment records and transaction outcomes.

Template · not live data

Pest control

Pest Control Service Intelligence Brief

Which severity and pest-type mix to schedule against route capacity.

Requires: Triage fields, scheduling records and completed treatments.

Template · not live data

A brief is generated from booking facts, CRM records and approved sources only. It is held on QA failure, never sent as a placeholder, and marked delivered only when a real asset URL, version and delivery receipt exist.

10Meta-to-Intelligence standard

From a paid click to recognized revenue, without losing provenance.

Nine steps. Each one ends in a definite result, and each one has a route for the day it fails.

Meta ad

Campaign, ad set, ad and creative version are named before launch so every downstream record can be traced back to the exact unit that produced it.

End result

A launched ad whose identifiers are registered in the tenant campaign registry.

If it fails

An ad with unregistered identifiers is paused; no spend continues against untraceable traffic.

Form or click-to-message

Native prefill for identity, explicit consent language, and exactly three vertical decision questions. Sensitive legal, medical or financial narrative is never requested here.

End result

A submission carrying identity, consent provenance, campaign identifiers and three decision answers.

If it fails

A submission missing consent provenance is quarantined and never enters an outbound cadence.

CRM capture and dedupe

The lead is written into the CRM under a deterministic key. A repeat submission merges onto the existing contact instead of creating a second record.

End result

One contact, one opportunity, one attribution chain, with the duplicate submission logged.

If it fails

An ambiguous match is held for owner review rather than merged on a guess.

Immediate contextual reply

The first reply references the actual ad context and the answers just given — not a generic greeting. Local outbound hours apply; outside them the reply is held and an owner task is created.

End result

A contextual first message delivered inside the reply target, or a held record with a dated owner task.

If it fails

A send failure raises an exception with the channel error and routes the contact to a human.

Reply interrupt

Any inbound reply cancels the scheduled cadence before the agent responds. Compliance and safety signals outrank everything above them.

End result

Cadence suppressed, the live conversation owned by one responder.

If it fails

If the interrupt cannot be confirmed, all automated outbound for the contact is suspended.

Qualified conversation

Qualification uses the vertical's approved criteria and hard boundaries. Low confidence escalates instead of guessing.

End result

A qualified or disqualified label with the evidence line that produced it.

If it fails

Unclear qualification routes to a human with the transcript excerpt attached.

Appointment

Booking stops acquisition cadence and moves the opportunity into brief preparation with an owner and a time.

End result

A calendar record joined to the opportunity and the originating ad identifiers.

If it fails

A failed calendar write blocks the stage move and creates an owner task within the hour.

Intelligence Brief

The vertical's brief is generated from verified facts, QA-checked by a person, and delivered on the timing rule for the meeting.

End result

A delivered brief with version, QA verdict and delivery receipt.

If it fails

QA failure holds delivery. A placeholder is never sent.

Held, closed and revenue feedback

Held, closed-won and recognized revenue are written back against the originating ad so spend decisions use outcomes, not form fills.

End result

A closed feedback loop from ad identifier to recognized revenue.

If it fails

Missing outcome data renders as missing in every view. It is never inferred and never modeled into an actual.

Form contract — required

Native prefill identity fields (name, phone, email) supplied by the platform
Consent provenance: exact language shown, timestamp and submission ID
Campaign ID, ad set ID, ad ID, form ID and form version
Exactly three vertical decision questions

Never requested on the form

Legal matter narrative, opposing parties or case detail
Medical history, symptoms, diagnoses or treatment records
Financial account numbers, balances, credit detail or documents
Any free-text field inviting sensitive disclosure

Sensitive detail is collected only in a secure CRM conversation or a human intake call, after consent, and is never requested on a Meta form.

Feedback events written back to the ad

Lead received
A deduped submission with consent provenance and campaign identifiers.
Source: CRM contact create with attribution payload.
Meaningfully engaged
A two-way exchange where the prospect answered a substantive question.
Source: Conversation events with inbound and outbound turns.
Qualified lead
The vertical's approved criteria were met, with the evidence line recorded.
Source: Qualification label plus transcript reference.
Appointment booked
A calendar record joined to the opportunity.
Source: Calendar write receipt.
Appointment held
The meeting occurred and was dispositioned by the owner.
Source: Owner disposition on the calendar record.
Closed won
A terminal won stage write against the opportunity.
Source: Stage history with idempotency key.
Recognized revenue
Revenue recognized by the client's own accounting rule, not an estimate.
Source: Closed-won revenue records supplied by the client.

Operating target — not a guaranteed result

First substantive reply within 15 seconds. Hard maximum 60 seconds, after which an exception is raised and an owner is alerted.

11Revenue Intelligence Pro

Seven numbers, four phases, one loop.

The commercial tier for operators who want the whole chain measured. Actual and modeled values are always visually distinct, and missing evidence renders as missing.

The seven-number baseline

Leads

Deduped contacts created in the period with consent provenance.

Source · CRM contact records.

Lead cost

Actual spend in the period divided by deduped leads.

Source · Ad platform spend export.

Contacted

Leads that received a delivered first message or answered call.

Source · Conversation delivery receipts.

Booked

Appointments created and joined to an opportunity.

Source · Calendar write receipts.

Held

Appointments dispositioned as held by the owner.

Source · Owner disposition records.

Closed won

Terminal won stage writes in the period.

Source · Stage history with idempotency keys.

Contribution profit per sale

Client-supplied contribution margin per closed sale. Never assumed.

Source · Client finance input.

Four conversion phases

01
New Lead to Booked
How much demand reaches a calendar, and how fast.
Owner: Conversation operations · End result: A booked appointment or a dispositioned reason the lead did not book.
02
Booked to Held
How much of the calendar survives to a real conversation.
Owner: Appointment owner · End result: A held meeting or a reschedule with a new dated slot.
03
Held to Decision
How many conversations reach a yes or an explicit no.
Owner: Closing owner · End result: A decision recorded as closed won, closed lost or a dated next step.
04
Not Now to Re-Engaged
How much of the deferred pipeline comes back on its own terms.
Owner: Nurture owner · End result: A re-engaged conversation or a suppressed contact honoring consent state.

Operating loop

SignalDecisionOwned ActionVerified OutcomeGoverned Learning

Modules

COMPASS

Ranks the single stage losing the most opportunities right now.

End result: One ranked leak with an owner and a dated correction.

WISDOM GRAPH

Holds what the platform has learned about objections, timing and language per vertical.

End result: An approved change candidate with its evidence attached.

ATLAS

Maps campaigns, sources and territories to downstream quality rather than lead volume.

End result: A spend recommendation with a named approver.

HORIZON

Internal operations module for forward capacity and scheduling pressure. Never a report title and never client-facing.

End result: A staffing or capacity action assigned to an operations owner.

Internal operations only

SPHERE

The conversation surface: voice and messaging behavior under one governed standard.

End result: A governed conversation with a written-back disposition.

OUTCOME ENGINE

Turns verified outcomes into the period's Intelligence Brief and the next governed experiment.

End result: A delivered brief and one approved experiment in canary.

Actual and modeled values are visually distinct, always, in every view.
Missing evidence renders as missing — never as zero, never as an estimate.
No sample index score, sample organization count or sample revenue is displayed as fact.
No patent claim and no benchmark-superiority claim is made anywhere.
Every module completes the loop: Signal, Decision, Owned Action, Verified Outcome, Governed Learning.

One system. Two engines. One standard.

EchoCody handles the conversation. North Star Intelligence decides what it meant, writes it where the business already works, and proves it happened. Neither ships without the other.

See the system Book a review