The SMB hub for AI operations
Most AI projects stall on the same three gaps: undocumented workflows, missing SOPs, and data locked in silos. I find them, fix them, then automate what's left, so AI gives you back time, margin, and room to scale. From an operator who scaled a company from $0 to $265M.
Live queue
Running“What time does the fuel dock close today?”
Hoursconf 0.97Draft ready“Can I move my reservation to next weekend?”
Reservation changeconf 0.88Staff review“I think I was charged twice last month.”
Billing disputeconf 0.62Escalated
Simulated run of the Email Triage blueprint · nothing sends without approval
The operator behind the hub
AI is new. Running operations isn't.
I spent twenty years building and running operating companies — from a $0 to $265M healthcare network to a $1B real estate platform. Now I help owner-led businesses put AI to work the same way I ran operations: with clear owners, measured results, and controls that keep it from breaking.
The AI Hub
Everything an SMB needs to put AI to work
Three pillars. Each one collects blueprints, guides, and the latest updates.
Your tool stack, connected
AI that works inside the tools you already pay for
Pick a tool to see what we automate with it and which blueprints use it. No rip-and-replace — the AI Ops layer sits between your systems and your team.
Make
Automation
What we automate
- ✓Watches inboxes, forms, and apps for new work
- ✓Routes AI output by confidence and risk
- ✓Retries and deduplication so nothing runs twice
Used in blueprints
Connect your systems
Inbox, CRM, accounting, and the sheets your team already keeps — wired into one flow so every task starts with the right facts.
Eval gate
Passed94%pass rate · floor 85%
- CASES
- 150
- INTENT
- 96%
- ESCALATED
- 100%
Example report · Email Triage blueprint
Measure before you trust
Every agent is tested against a labeled set of your real work. If a change lowers the pass rate, it doesn't ship.
Inbox
AI Ops agent
Drafted a reply to the balance question · conf 0.93
Try it: click to approve
Let AI act — with approval
AI drafts and prepares; your team approves with one click. Low-risk tasks earn autonomy once the data says so.
The Earned Autonomy method
AI starts supervised. It earns autonomy with data.
Most AI projects stall because nobody owns them, nobody measured the before, and nobody planned for mistakes. Scroll through how every workflow moves from Red to Green.
Autonomy level
Red: Human decides
Autonomy is earned with data. It's never assumed.
- Stage 1 · Red
Human decides
AI researches and drafts. A person makes every decision. Every new workflow starts here, and so does anything touching money, legal exposure, or an upset customer.
Exit when: a labeled test set exists and accuracy is known.
- Stage 2 · Yellow
Human approves
AI acts, but a person reviews before anything reaches a customer. Low-confidence work goes to its own queue.
Track: accuracy, share approved without edits, escalation rate.
- Stage 3 · Green
AI runs, people audit
Promoted one task type at a time, only after it holds its pass rate. A fixed sample is still reviewed every month.
Demotion is allowed: if accuracy drops, it goes back to Yellow.
Companies I've worked with
Blueprints
See the exact systems, not just the promises
Working reference builds with the architecture, stack, controls, and what each one measures. Clearly labeled — no invented client results.
Drag to explore →
Hospitality & booking
Inbound Email Triage Agent with Eval Gating
An agent reads every inbound email, classifies it, pulls facts from your systems, and writes a draft reply for staff to approve. Nothing sends automatically until it's measured.
Make.com · Relevance AI · Gmail · QuickBooks Online · Google Sheets · promptfoo
See the architecture →Cross-industry
Ops Knowledge Brain (Cited Answers from Your SOPs)
Staff ask questions in plain English and get answers drawn only from your policies and SOPs — with the source document and page cited under every answer.
n8n · Supabase pgvector · Claude · Google Drive
See the architecture →Property management & finance
Document Intake Extractor (Invoices, Leases, Contracts)
Uploaded invoices, leases, and contracts are parsed into structured fields with a confidence score on each one, then written to your tracker — low-confidence fields flagged for a person.
n8n · Claude · Google Sheets · OCR
See the architecture →Professional services
60-Second Lead Intake Pipeline
A web inquiry is captured, researched, scored, logged in HubSpot, answered with a personalized first reply, and announced to your team in Slack — in under a minute.
n8n · Claude · HubSpot · Gmail · Slack · Tally
See the architecture →Cross-industry
AI Ops KPI Dashboard with Weekly Narrative
A live dashboard of your core operating numbers, plus a plain-English weekly brief that explains what moved, flags risks, and recommends three actions.
n8n · Google Sheets · Claude · Recharts
See the architecture →Cross-industry
Multi-Agent Operations Advisory Team
Four specialized agents — research, strategy, marketing, and operations — work in sequence on a business challenge and hand back an executive brief with a 30-day plan.
n8n · Claude
See the architecture →Professional services
Sales Pipeline and Forecast Board
Every open deal on one board by stage, with live totals and a forecast weighted by stage win rates, so you see what is likely to close without maintaining a spreadsheet.
HubSpot or Pipedrive · n8n · Claude · Google Sheets
See the architecture →Professional services
Staged Client Onboarding Checklist
A step-by-step onboarding flow where finishing one stage unlocks the next, every step has one owner, and progress is visible to your team and the client.
Notion or Airtable · n8n · Gmail · DocuSign or PandaDoc
See the architecture →Cross-industry
SOP-Driven Task Drafts
A task board that shows the written procedure next to each task and drafts a first pass from the two, so your team starts from a draft that follows your own process.
Notion or Airtable · Claude · n8n · Google Drive
See the architecture →Try them yourself
Eight interactive reference builds
Click through working tools for lead follow-up, sales, onboarding, delivery, knowledge, billing, reporting, and support. They run on sample data and are reference builds, not client results.
- Reference buildLeadPilotMarketing & Lead GenSortable lead table — pick a lead and get a data-driven follow-up draft.See the problem it solves →
- Reference buildPipelineIQSalesDrag-and-drop deal board with live totals and weighted forecast.See the problem it solves →
- Reference buildOnboardOnboardingStaged checklist — finishing a stage unlocks the next and updates progress.See the problem it solves →
- Reference buildFlowDeskDeliveryTask board with an SOP panel that drafts a first pass from task and SOP data.See the problem it solves →
- Reference buildKnowledgeBaseCommunication & KnowledgeLive search and tag filters across an SOP article set.See the problem it solves →
- Reference buildLedgerFlowFinance & BillingInvoice table — mark one paid and AR aging and DSO recalculate.See the problem it solves →
- Reference buildPulseBoardReporting & AnalyticsDate and segment filters redraw a chart and a plain-English trend summary.See the problem it solves →
- Reference buildRetainAIInbound messagesInbound message queue by priority and account health, with a tailored draft reply.See the problem it solves →
How an install runs
Five phases. Governance before the build.
Owners, thresholds, and a failure plan are set before anything is automated — so AI lands on clean processes, not broken ones.
- 1
Align
Weeks 1–2Leadership interview, readiness score, shadow-AI inventory, top-10 workflow list.
- 2
Map
Weeks 2–3Document the three highest-value workflows and capture baselines.
- 3
Govern
Weeks 3–4Assign owners, set risk tiers, build the approval matrix and failure playbook — before building.
- 4
Deploy
Weeks 4–8Build against the rules. Run in parallel with the human process until quality thresholds are met.
- 5
Optimize
OngoingMonthly review of KPIs, exceptions, and cost. Promote workflows from Yellow to Green as they earn it.
SMB AI Daily
What changed in AI operations — and what it means for you
Short, sourced briefings written for owners and operators, not engineers.
Work with Daniel
Diagnose. Build with controls. Oversee.
Every engagement follows the same three steps. Start with the Diagnostic — most clients know within two weeks exactly where AI pays off.
Step 1
AI Ops Diagnostic
From $2,500
2 weeks · fixed fee
Find the three workflows worth automating first, with baselines and a 90-day roadmap.
- ✓Workflow inventory & baselines
- ✓Shadow-AI check
- ✓Readiness scorecard
- ✓Ranked roadmap
Step 2
Governed Build
From $6,000
per workflow · 3–6 weeks
One workflow built on your tools with controls, an eval set, and a failure playbook.
- ✓Make / n8n + AI agent build
- ✓Approval matrix & quality gates
- ✓Labeled test set
- ✓Team handoff & runbook
Step 3
AI Ops Oversight
From $2,000/mo
monthly retainer
Keep it accurate, keep costs in line, and promote workflows to autonomy as they earn it.
- ✓Monthly accuracy & cost review
- ✓Eval re-runs after changes
- ✓Promotion to autonomy
- ✓New workflow each quarter
FAQ
Straight answers
We already pay for ChatGPT or Copilot. Why do we need this?
Generic tools help individuals. Operations change when AI is wired into your workflows — your inbox, CRM, and accounting — with owners, controls, and measurement. That's the part we install.
Will AI send things to our customers without us seeing them?
Not unless you decide it should. Every workflow starts supervised: AI drafts, your team approves. Specific low-risk tasks are promoted to autonomous only after they hold an agreed accuracy rate.
Do we need developers or new servers?
No. Builds run on licensed tools you log into — Make, n8n cloud, your CRM, and an AI provider — and come with a runbook your team can maintain.
Are the blueprints real client projects?
They're working reference builds, and we label them that way. They show exactly what we'd install and how it's governed. Client results are published only with measured before-and-after numbers and permission.
How do you measure whether it worked?
We capture a baseline before building — time per task, cost per unit, error rate — and report against it monthly. Efficiency claims are always per workflow, against your own numbers.
Find the three workflows worth automating first.
A 20-minute operations call. No pitch deck — we look at where your team's hours go and whether AI is the right fix.