How to build an AI chief of staff that connects to all your tools
The first 45 minutes of a founder’s day is often the same.
Open the CRM. Check what moved. Open the PM tool. See what is overdue. Scan the inbox. Look for anything urgent. Open the financials. Check the cash position.
By the time the first meeting starts, the founder has a picture of the business. That picture took 45 minutes to assemble from four separate tools. It required no judgment whatsoever.
An AI Chief of Staff assembles it automatically. The founder arrives to a brief. The 45 minutes go back.
What an AI Chief of Staff actually does: the full capability picture
The AI Chief of Staff is a connected information synthesis system.
It pulls structured data from the company’s operational tools, applies AI analysis to identify what matters, and surfaces the relevant signal to the founder; either as a scheduled brief or in response to a specific question.
Capability 1: Morning brief (the core)
Every morning before the founder’s first meeting, a structured brief covering:
- Pipeline: total pipeline value, deals that moved in the last 24 hours, stalled deals requiring attention
- Projects: overdue tasks, client deliverables due in the next 48 hours, any team member blocked
- Finance: current cash position, invoices overdue by more than 7 days, week-over-week revenue change
- Communications: emails that genuinely require the founder’s attention; filtered from the noise
- Priority actions: two to three specific actions recommended based on the brief’s data
Capability 2: Conversational data queries
The founder asks a natural language question; the AI queries the connected data sources:
- “Which active deals have not been updated in 14 days?” → CRM query
- “What is our cash position today versus last month?” → Accounting query
- “Which team members have the most overdue tasks this week?” → PM tool query
- “Which clients have had zero contact from us in the last 30 days?” → CRM activity query
Capability 3: Drafting on demand
With the data context and company context pack loaded:
- “Draft a follow-up to the Meridian deal based on where we are in the pipeline”
- “Write a client update for the Henderson project based on this week’s PM data”
- “Draft a response to this email [paste email]“
What it does not do
- Make decisions. The AI surfaces information and recommendations; the founder makes every decision.
- Replace a human Chief of Staff for relationship-intensive executive work.
- Initiate external communications without human approval.
The five data connections: what to connect and in what order
Connection 1: CRM (highest priority)
What it enables: pipeline visibility, deal health monitoring, client contact tracking, stalled deal detection.
How to connect:
| CRM | Integration | Setup time |
|---|---|---|
| HubSpot | Native Make/Zapier | Under 30 minutes |
| Salesforce | Native Make/Zapier | Under 30 minutes |
| Close | API via Make/Zapier | Under 30 minutes |
| Pipedrive | Native Make/Zapier | Under 30 minutes |
Data pulled: all active deals with stage, value, last activity date, and owner. Pipeline summary aggregated by stage.
Frequency: pulled daily at 5:30am for the morning brief.
Connection 2: Project management tool (second priority)
What it enables: project health monitoring, overdue task detection, team workload visibility.
| PM tool | Integration |
|---|---|
| Asana, Monday | Native Make/Zapier |
| Linear | API via Make/Zapier |
| Notion | Notion API via Make/Zapier (1–2 extra hours) |
Data pulled: active projects with status and due date; tasks overdue by more than 24 hours; tasks due in the next 48 hours; blocked tasks.
Connection 3: Accounting tool (third priority)
What it enables: cash position visibility, overdue invoice monitoring, revenue tracking.
QuickBooks and Xero both have native Make/Zapier integrations.
Data pulled: current cash position; overdue invoices (30+ days); revenue this week versus last week; outstanding AP over 30 days.
Connection 4: Email (fourth priority)
What it enables: important email flagging; surfacing the three to five emails that genuinely require the founder’s attention.
The filtering prompt:
Review the following emails received in the last 24 hours. Identify which require
the founder's attention today, in priority order.
Flag as urgent if:
- Email from a client mentioning a problem or delay
- Email mentions a time-sensitive decision
- Email from a team member reporting a blocker
Ignore: newsletters, automated notifications, cold outreach, routine status updates.
[Paste email subjects and sender names, or full content]Connection 5: Calendar (fifth priority; completes the brief)
What it enables: daily meeting context; what the founder is doing today and what preparation is needed.
What the AI adds: for each external meeting, a one-sentence context note drawn from the CRM or PM tool.
Example: “Miller Group call; last CRM note: three weeks since last contact, proposal in stage 3.”
The morning brief architecture: the exact format and build
The morning brief format:
AI CHIEF OF STAFF; DAILY BRIEF
[Day, Date]
TODAY'S SCHEDULE
[Time] ; [Meeting name] ; [One-line context from CRM/PM]
[Time] ; [Meeting name] ; [One-line context from CRM/PM]
PIPELINE STATUS
Total active pipeline: $[X] across [N] deals
Moved forward this week: [deal name, stage change]
Stalled (14+ days no activity): [deal names and days since last activity]
Closing this month: [deals in final stage with close date]
PROJECT HEALTH
All green: [N] projects on track
Attention needed: [Project name] ; [specific issue]
Due in 48 hours: [Task or deliverable name, project, owner]
FINANCIAL POSITION
Cash position: $[X] (vs $[X] last week)
Overdue receivables: [Client name, amount, days overdue]
Invoices going out this week: $[X]
FLAGGED EMAILS
[Sender] ; [Subject] ; [One-line summary of why this is flagged]
[Sender] ; [Subject] ; [One-line summary]
PRIORITY ACTIONS
1. [Specific action recommended based on brief data]
2. [Specific action]
3. [Specific action, if needed]The build: one focused day
| Time block | Work | Duration |
|---|---|---|
| Morning | Connect CRM and PM tool to Make; configure 5:30am data pull to Google Sheet; test | 3 hours |
| Afternoon | Write AI analysis prompt; configure Make to pass data to Claude API, receive brief, send via email at 6am; test end-to-end | 3 hours |
| End of day | Connect accounting, email, and calendar integrations; run first full brief on live data | 1 hour |
Total build time: 7–8 hours. Total monthly cost to run: $3–$8 in API inference.
The conversational mode: how to add on-demand queries
The morning brief gives the daily picture. The conversational mode gives the ability to go deeper on any question the brief surfaces; or any question that occurs between briefs.
The simplest implementation: a Slack command
A Slack bot accepts natural language questions from the founder and returns answers from the connected data sources.
1. Founder types: /costaff which deals have stalled?
2. Slack bot triggers a Make/Zapier scenario
3. Scenario queries the CRM data (from Google Sheet or live API)
4. AI receives the query and current data; returns a specific answer
5. Answer appears in the Slack thread within 30 secondsBuild time: 2–3 hours once the data connections from the morning brief are already in place.
Query types that work well:
- Lookups: “What is the status of the Henderson deal?”
- Aggregations: “How many deals are in proposal stage right now?”
- Comparisons: “Is our pipeline bigger or smaller than this time last month?”
- Health checks: “Which team members have the most overdue tasks?”
- Financial: “What is outstanding AR over 30 days?”
Query types that work less well:
- Questions requiring knowledge not in the connected data sources (“why did this deal stall?”; the CRM has the activity log, not the reason)
- Real-time data queries when only batch-exported data is connected
- Strategic questions without operational data backing
What makes it a Chief of Staff rather than a dashboard: the judgment layer
A dashboard shows data. A Chief of Staff interprets it.
Three judgment elements the AI adds:
Element 1: Prioritization across 47 items
The Henderson deal has not been touched in 18 days and closes this month. That is both urgent and impactful. The Whitmore deal is stalled but closes in 90 days; a note for next week, not today.
The AI surfaces which two or three items require attention today; not a full list of everything that moved.
Element 2: Pattern recognition across tools
The dashboard shows data within each tool. The AI Chief of Staff sees across tools.
The client with a stalled deal also has a late invoice and an overdue project task. That combination; revenue risk plus relationship risk plus operational risk; is a pattern the dashboard does not surface because it exists across three separate systems.
Element 3: Action recommendation
The dashboard shows what happened. The AI recommends what to do about it; drawing on the decision rules in the business context.
“The Henderson deal has not been touched in 18 days and closes this month. Based on your standard follow-up protocol, this requires a personal call before the end of day.”
These three elements require the context pack.
Without business context loaded, the AI produces generic observations. With it, the brief surfaces the specific signals the founder would have prioritized manually.
Common questions on the AI Chief of Staff
”Does this work for a company not using HubSpot or Salesforce?”
Yes. The architecture works with any CRM that has a Make or Zapier integration: Close, Pipedrive, Zoho, Copper, and most other mid-market CRMs.
For companies running pipeline in a Notion database or Google Sheet rather than a dedicated CRM: the Notion API or Google Sheets connection replaces the CRM integration. The brief structure is the same; the data source is different.
”What if my tools don’t have Make/Zapier integrations?”
Most mid-market operational tools do. For those that do not: weekly manual CSV export into the aggregation Google Sheet is the fallback. Adds 15–20 minutes of manual work but keeps the brief running.
If a core tool consistently lacks automation access, that is a signal to evaluate whether the tool should be replaced with one that has better API access.
”How do I handle sensitive financial data going through an AI system?”
Use the Claude API (not the consumer subscription) under the Data Processing Addendum terms. The API’s data processing agreement covers that submitted data is not used for training.
For particularly sensitive financial data: design the aggregation layer to pass summary figures rather than raw transaction data. The AI receives “cash position $487,000, up $23,000 from last week” rather than individual transaction records.
”Can more than one person receive the brief?”
Yes. The Make email module sends to a list of recipients. Typically: the founder and the operations lead.
For larger teams, a Slack post to a private channel makes the brief accessible to a wider group without requiring a distribution list update every time someone joins or leaves.
”How much does it cost to run this daily?”
At current API pricing (2026):
- The daily brief processes approximately 2,000–5,000 tokens of input
- AI output: 500–800 tokens
- Cost per brief: $0.10–$0.30 using Claude Sonnet
- Monthly total: $3–$9
The cost of the brief is effectively zero relative to the value of the morning time it returns.
”What happens when a data connection breaks?”
Make and Zapier have error notification systems; a failed scenario sends an error email.
The most common failures:
- API credential expirations (typically every 60–90 days for some integrations)
- Data format changes when a tool updates its export structure
Build a simple error handling step in the Make scenario: if the data pull fails, send the founder a brief that says “data connection issue; check Make error log.”
This prevents confusion about why the brief did not arrive.
Want the AI Chief of Staff designed and built; connected to all your operational tools and delivering the morning brief before your first meeting?
The AI Chief of Staff is a connected system built on top of the tools the company already runs on; CRM, PM tool, accounting, email, calendar.
An AI layer synthesises the data, applies judgment from the business context, and surfaces what matters before the founder’s first meeting.
The morning brief is the foundation. The conversational mode extends it.
The build takes one focused day for the core brief and an additional half day for conversational queries. The return is the 45–60 minutes of morning information assembly that currently consumes the most expensive hour of the founder’s day.
Path one: start with the CRM connection this week. Connect your CRM to Make, pull your active pipeline data to a Google Sheet, and run the pipeline section of the brief prompt above manually. The manual version takes 10 minutes and shows you immediately whether the automated version is worth building.
Path two: bring in a partner. If you want the full AI Chief of Staff; all five data connections, the morning brief, the conversational Slack interface, and the business context layer; that is the connected operational environment Phos AI Labs builds in Phase 3 and Phase 4. Across 400+ business engagements, the pattern is consistent. Thirty minutes, no deck. Start here.