What this station holds
Operations is the kitchen’s nervous system. Every other station depends on it. The rep needs the lead enrichment pipeline to run on time. The service team needs the customer context pulled from three places and synthesized into one. The finance team needs the recurring reports so the board has numbers to decide with. The research team needs the prior work surfaced fast when there’s no time to dig.
Seventy percent of the operations team’s week is structured, repeatable work. Files come in and need filing. Invoices arrive and need reading. Meetings happen and need notes. Reports are due on a cadence and need building. Contracts need classifying. Handoffs cross departments and need translating. The team is doing the same work the same way, on the same schedule, every week.
The Operations Station is built to own that rhythm. It reads inbound documents and sorts them by type. It pulls the numbers out of unstructured text and drops them into the right cell. It builds the board report from the quarter’s numbers. It generates the meeting brief from last week’s notes and the open items list. It writes the handoff doc so the next team doesn’t have to rebuild context. It watches the resource calendar and flags where the pipeline is over or under-loaded.
This is not the station replacing judgment. This is the station doing the filing so judgment stays sharp.
The dependency property within the station
Pull the Chef and the station generates correct output but the business runs on yesterday’s data. Speed without schema is noise. Pull the Sous Chef and documents stack in the wrong folders. The weekly report shows up late because the station was never told what day the finance team needs it. Pull any single early dish (classification, extraction, recurring reports) and the station fragments. Classification fails without a schema. Extraction fails without guardrails on confidence thresholds. Reports fail without measurement of accuracy.
The difference from other stations is the weight on schema definition. You cannot give the station a job without first writing down what done looks like. A classification schema. A confidence threshold. A template. The operating cadence. Most operators try to deploy first and define the schema afterward. The station fails. The operator concludes AI can’t do this work. The truth is simpler. The schema was never written.
Operations dishes 1-3 run on what sits waiting. Document types. Invoices. Forms. Standard templates. The training set is already sitting in your folders. Dishes 4-6 cross a threshold. Context plumbing gets hard. The meeting prep needs last week’s notes plus the project tracker plus the Slack thread. The handoff doc needs data from two different systems. Measurement starts to matter. Dishes 7-10 stay taste-heavy. Resource allocation needs review. Cadence design needs the Chef. These don’t live at the pass because there’s a shortage of time. They live at the pass because the recipe gets sharper when the Chef is tasting.
The dish order
The first three dishes run on what sits waiting in your business. Inbound documents. The filing system you already use. Past extract jobs. The guardrails are already encoded in your rules. You just haven’t told the station what they are.
The middle four cross a threshold. Context plumbing matters. Meeting prep needs the Slack thread. Handoff translation needs data flow between systems. Resource allocation needs the capacity view and the project tracker connected. Guardrails have to be written before deployment, not after.
The last three are where most operators stall. Complaint synthesis needs human eyes. Resource allocation needs authorization before anything moves. Strategic cadence design needs the Chef’s hands on it. These don’t live at the pass because the Chef is busy. They live at the pass because the recipe gets sharper when the Chef is tasting.
- Document Classification. Inbound files, contracts, invoices, forms. Sorted, tagged, and routed without human handling. See document-classification.md.
- Data Extraction. Pull the structured fields out of unstructured documents. Into the system of record. See data-extraction.md.
- Recurring Report Generation. The weekly dashboard. The monthly board pack. The quarterly review. Drafted from source data, formatted to your standard. See recurring-report-generation.md.
- Meeting Prep Packets. For every recurring meeting, generate the brief. Agenda, prior decisions, open items, relevant data. Before the meeting starts. See meeting-prep-packets.md.
- Process Documentation Drafting. When someone runs a workflow, the station produces the SOP from the actual run. Observed, not invented. See process-documentation-drafting.md.
- Standard Internal Communication. Status updates, weekly notes, leadership memos. Drafted from the actual week, not from a template. See standard-internal-communication.md.
- Vendor and Partner Triage. Inbound vendor pitches, partner inquiries, RFPs. Sorted, summarized, and queued by relevance. See vendor-and-partner-triage.md.
- Workflow Handoff Translation. When a job crosses departments, the station writes the handoff doc. What’s done, what’s pending, what the next team needs. See workflow-handoff-translation.md.
- Resource Allocation Analysis. Read the project tracker, the calendar, and the team’s capacity. Surface where the pipeline is over or under-loaded. See resource-allocation-analysis.md.
- Strategic Operating Cadence Design. Read the last quarter. Meetings, deliverables, decisions. Propose where the cadence is broken. See strategic-operating-cadence-design.md.
What breaks this station
Schema never gets written. The operator wants to deploy classification without defining the categories. The station defaults to a generic taxonomy and misses the business-specific splits. By month two, the team is quiet-reclassifying documents and the station never learns.
Context windows are starved. The meeting prep needs the transcript plus the notes from last time plus the open items list. The station sees only the current week’s transcript. It generates a generic agenda. The meeting prep that reads like the Chief Operations Officer wrote it would use all three sources.
Guardrails aren’t defined upfront. The station extracts data from invoices. The confidence is 82 percent. The operator never said “below ninety percent, escalate to a human.” So the station is pushing 82-percent-confident numbers into the system of record. Bad data compounds.
Measurement lives nowhere. The recurring report goes live. It’s accurate on day one. Month two, the source changed and the report is silently wrong. The team is quiet-fixing it. Nobody updated the recipe. The station never knows it broke.
Feedback loop dies early. The team is reordering handoff documents the station drafted because they’re missing the context the next team actually needs. Nobody tells the station what the next team said. Three months later, the handoff docs are still incomplete because the miss never made it back into training.
The Monday Move for this station
Pick one document type. Invoices. Contracts. Intake forms. Write down the categories that matter. Those categories are the schema. Pull ten examples from last month. Show the station what done looks like. Have it classify the next ten. Compare. Write the first three by hand. Show them to the station. Then let it classify twenty more while you watch. The station is now running. Operations is at the pass. This is what week four looks like.
Original framework. Distilled from client work. Companion to The Station Plan and The Professional Recipe.