Case study · Professional services

A whole business on one screen, and a morning that runs itself.

Seven systems and one founder's head, replaced by a connected AI operating system: an automated chief of staff, a self-checking dashboard and a knowledge graph that maintains itself. Built on Claude and HubSpot.

15 to 20 hrs
founder time returned a month
14
scheduled automations, each fail-loud
126
entity pages that maintain themselves
A morning briefing open on a laptop screen
Sector
B2B professional services
Engagement
AI operating system
Built on
Claude and HubSpot
Connected
HubSpot, Gmail, Slack, Asana, accounting
The situation

Seven systems, and the only place they joined up was one founder's head.

A founder-led B2B professional services firm running on HubSpot: a lean senior team, a handful of retainer clients and a steady project pipeline. The problem was not effort.

The business ran across HubSpot, Gmail, calendar, Slack, Asana, call recordings and accounting, and the joins were the problem. Every morning started with 45 minutes of triage. Numbers lived in different tools and quietly disagreed. Commitments made on calls slipped because nothing chased them, and the monthly finance picture arrived weeks late.

A dashboard on dirty data is a faster way to be wrong.

What we did

One connected operating system, in four layers.

Clean data first, then the chief of staff, then the dashboard, then the intelligence.

01 · Chief of staff
A morning brief that acts

Every weekday at 07:45 an agent reconciles inbox, calendar, Slack, Asana, HubSpot and call recordings against yesterday's plan, drafts replies in the founder's voice and posts one Slack brief. The founder replies in Slack; the next run acts on it.

02 · Dashboard
One screen that checks itself

Built around the four jobs a founder does: founder, exec assistant, finance director and sales manager. Every zone stamped with the age of its data, and every shared figure compared across systems on each refresh.

03 · Sales
A self-improving sales engine

Fortnightly agents find deals hiding in email threads, draft follow-ups for cold deals, close what has genuinely died, and mine every win and loss for lessons that update the proposal generator.

04 · Knowledge
A graph that maintains itself

A living note for every client, prospect, service and person, regenerated daily. 75 days of history backfilled on day one: 87 interactions across 21 companies.

What changed.

15 to 20 hrs

of founder time returned every month.

14

scheduled automations, each one fail-loud when a step breaks.

126

entity pages that maintain themselves, wikilinked both ways.

1 screen

replacing seven logins for the daily read of the business.

What it caught, first two weeks

Where the value stops being theoretical.

A proposal reported overdue, already sent

Tracked sends through the CRM sidebar were invisible to inbox search. The system learned to cross-check CRM activity, corrected the record the same morning and encoded the lesson so it can't repeat.

A new retainer won but not yet invoicing

The gap between committed and invoiced recurring revenue is now a standing rule. It fired the day the deal closed and stays on screen until the first invoice goes out.

A quiet client, caught by history

Backfilling 75 days of calls and email revealed one retainer client with no material contact across the whole period, days before a critical on-site session. Silence generates no notification, so now silence itself is data.

Delivery debt made unmissable

Every overdue deliverable, named with days late, sits at the top of the brief until it's done. Not motivational. Effective.

Why this study is anonymised

It runs a whole business, so we keep the name private.

This is a live operating system running across a firm's inbox, calendar, CRM, finances and call history.
Because it touches everything, we don't publish the name or a named quote.
The build and the numbers are real. That's the standard we'd hold for your business too.

Running on seven systems and one founder's head?

This is what it should feel like instead. Talk to Rikki about a custom AI build on your own stack, and we'll tell you where we'd start, including if the answer is somewhere smaller.

Rikki Lear
Rikki LearCommercial Director