private / NDA2026

AI Executive Dashboard

One interface for a whole company's operations.

Role

Architecture, agent infrastructure, implementation

departments

visibility

real-time

reporting

auto

n8nGoHighLevelSlack APILLM APIsNode.js

Asking how last week went

Export from the CRM
Pull calling-platform stats separately
A third person reconciles the two
Answer arrives Wednesday, about Monday
Four tools, each holding a quarter of the truth

Now

One pipeline every view reads from
A metric means the same thing everywhere
Department heads stop pinging each other
Current when the question is asked
Ask it in Slack and get an answer

What it looks like

CEO dashboard: KPI tiles across the top, then per-department metric tables with this week against target, an owner and a source for every number — The meta tracker. Two to three KPIs per department, this week against target — and every number carries an owner and the tracker it came from, so a red cell has somewhere to go.

Client management: account cards showing retainer, cost per lead, show rate, health status and open requests — Every account in one place — what it pays, how delivery is going, and the last 30 days of numbers.

Fulfillment overview: pipeline by status, workload by owner, and busiest accounts — Where delivery stands right now. Pipeline by status on the left, workload per person in the middle — the view that stops work quietly piling on one owner.

Fulfillment board: a kanban of client requests across intake, scoping, in progress, client review and blocked — The same requests as a board. Every move between columns is written to the request's history.

How it fits together

runs itselfneeds a person

01
Four source systems
CRM · calling · Slack · internal tools
02
One pipeline
cross-platform APIs, normalised
03
LLM layer
reads the pipeline, explains it
04
Four department views
and reporting that writes itself
05human
Ask in Slack
operators state intent

Unify first, visualise second. Everything downstream reads the pipeline rather than the source tools, which is what stops the same metric meaning two different things in two different views.

The data underneath

Marketing tracker: spend, leads, cost per lead, qualified, booked and show rate per client — Cost per lead and cost per booking, per client, per day. What you zoom into when the CEO view turns red.

Sales tracker: setter and closer metrics per person per week, dials through to cash collected — Per person, per week — dials through to cash. Where a show-rate problem gets a name.

Retention tracker: churn, upsells, referrals and reviews with a reason attached to each event — Every churn needs a reason attached. Retention is cheaper than acquisition.

Quarter goals: objectives with key results, each with a start, a target, an owner and progress measured from the start — Progress measured from where the number started, not from zero.

The agent layer underneath

runs itselfneeds a person

01human
Intent
stated in plain language
02
Claude Code
routes it to the right agent
03
Hermes Agent SDK
executes autonomously
04
Result
back in the channel it started in

An operator states an intent; the routing and execution happen without anyone shepherding each step.

What it changed

4 departments, one interface: Call Center, Marketing, Sales, Fulfillment — each sees its own slice of the same pipeline.
Real-time, not assembled: Replaces a weekly export-and-reconcile cycle that always described last week.
Reporting writes itself: The recurring manual assembly is gone, not faster.
Slack is the interface: Agents and custom skills run where the team already is, rather than behind another login.

Code and screenshots are under NDA. The architecture and the outcomes are what I can show.