An ops copilot that knows every SOP, every customer, and every internal tool — and never forgets.

An internal AI assistant your team uses every day — answers ops questions, enforces SOPs, drafts invoices, triages tickets, automates scheduling — all over Slack or your internal portal. Your team currently spends 30–40% of their day on internal navigation: 'where's the SOP for X', 'what did we charge that client last time', 'who owns this account'. The…

The problem, precisely

Inside any 8–60 person service business there is a hidden tax that nobody puts on the P&L: the time people spend asking and re-answering internal questions. Where's the SOP for the refund process? What's the price we quoted Acme on the install in March? Did anyone follow up on the inspection report Steve sent last Tuesday? Who's the owner on the Riverside account?

How we build it

The copilot is a Slack-native (or Microsoft Teams) agent built on Anthropic Claude 3.5 Sonnet with a structured tool-call layer to every internal system that matters. We integrate with: Notion (for SOPs / wiki), Google Drive (for shared docs), HubSpot/Salesforce/Pipedrive (for customer history + pricing), Jobber/Housecall Pro/ServiceTitan (for dispatch + job records), QuickBooks/Xero/Stripe (for invoicing), Slack (for thread search), Gmail/Outlook (optional, for email history), and any internal tool with an API. Read-access is the default; write-access (draft an invoice, create a Notion page, send an email) is gated per-tool and per-user role. The RAG layer is built on Pinecone for vector search across documents and Cohere rerank for precision.

What you get

• Slack-native (or Teams) ops copilot installed on @copilot or /ops mention
• Read integrations: Notion, Google Drive, HubSpot/Salesforce/Pipedrive, Jobber/HCP/ServiceTitan, QuickBooks/Xero/Stripe, Gmail/Outlook
• Write integrations (role-gated): draft invoice, create Notion page, send email, schedule appointment, push CRM update
• Custom SOP enforcement: the copilot will refuse to draft an action that violates a documented policy (e.g. refunds over $500 require owner approval)
• Per-user role-based access controls (RBAC)
• Audit log of every query and tool call (compliance-grade)
• Auto-suggested 'we should write this down' backlog — captures the long tail of tribal knowledge
• Quarterly retraining on new SOPs + updated tool docs

Operations team case: 22 minutes/person/day saved on internal lookups × 14 people × $42/hr loaded = $2,150/week recovered.

• Average knowledge-worker time spent on internal information search: 20–30% of workday (McKinsey, IDC)
• Realistic recovery from a good RAG copilot: 60–80% of that time
• Per-person savings: 15–30 minutes/day
• Whole-team cost saved: linear with headcount × loaded labor rate

A worked example

A 14-person ops team at a regional services franchise recovered an average of 22 minutes/person/day. At a $42/hr loaded labor rate that's $2,150/week, or $112K/year. System cost: $899/mo. Payback: 4 weeks.

Cost and timeline

Starts at $7,500 build. Ongoing: $899/mo + API pass-through (~$0.40 per heavy query, ~$0.04 per light query). Ships in 21 days · deeper integrations phased in after launch. Delivered as internal AI Operations Assistant, available across Global.

Will my data go to OpenAI / Anthropic for training?

No. We use Anthropic's API with the no-train flag set (zero-retention enterprise endpoint where available). For high-compliance contexts (healthcare, legal, finance) we route through Anthropic's HIPAA-eligible endpoints or self-hosted Llama-3.1 on Modal. Either way: your data is never in a third-party training set.

How does access control work?

We define roles (admin, manager, IC) and per-role allow-lists for which tools each role can call, and which data segments each role can query. A junior IC asking 'show me the financials for last month' gets a polite no; their manager gets the answer. RBAC is enforced server-side, not just in the prompt.

What if our SOPs are out of date?

The copilot will surface its source for every answer with a 'last updated' timestamp. When you spot stale info, you update the source — Notion, Drive, whatever — and the copilot's vector index syncs within 60 seconds. We also run a quarterly 'SOP freshness audit' that flags policies that haven't been touched in 12+ months.