The Reporting Process at Vye [Updated for AI]

Contents

1. Why we're doing this

2. High-level overview

3. Reporting cadence
4. Who does what?
5. How do I pull the report?
6. When things go wrong
7. How Claude works (the non-technical version) 

Why we're doing this

Reporting has always been one of the most time-intensive parts of client delivery at Vye. Writing insights, pulling channel data, formatting deliverables, and tailoring each report to each client's goals took hours — and quality varied depending on who was doing it and how much time they had. We built an AI-powered reporting workflow to solve that:

• Speed without sacrificing quality. Claude drafts structured, insight-driven reports in a fraction of the time, freeing up DMS and AM bandwidth for higher-value work.

• Consistency across clients. Every report follows the same logic and standards, regardless of who runs it.

• Smarter insights. The AI synthesizes data across channels and surfaces patterns that might get missed in a manual pull.

• Scalability. As our client roster grows, we can't afford to grow reporting hours at the same rate. This is how we scale without scaling headcount.

The goal is for every client to feel like they have a dedicated analyst on their account — because now they do.

High-Level Overview

The AI-powered reporting process runs inside Claude. Each client has a dedicated Claude skill that holds their reporting configuration, historical context, and any client-specific quirks.

When it's time to run a report, the team member triggers the process inside the client's project. Claude then:

1. Pulls the client's reporting config automatically (data sources, deliverable format, required comparisons, methodology caveats)
2. Asks for the data it needs — the team member pastes or uploads it
3. Drafts the full report: channel-by-channel performance, insights, and recommendations
4. Prompts the team member to add context Claude can't know (call notes, campaign context, what the client said recently)
5. Produces the final deliverable in the right format for that client

The team member is not replaced — they are the expert in the room taking on an a more strategic editorial role while Claude handles the structure, drafting, and synthesis.

Reporting Cadence

We're moving away from treating monthly and quarterly reporting as separate, parallel tracks. Going forward, we operate with one philosophy: We are always working toward a quarterly view. Monthly and mid-quarter updates are checkpoints along the way — not standalone deliverables with their own separate logic.

This does not mean month-over-month insights are lost or omitted. What this means in practice:

• Monthly updates feed into and contextualize the quarter — they're not just snapshots of last month.

• Every month, we're updating the client's quarter-to-date KPI progress, not just reporting on what happened in the prior period.

• By the time a quarter closes, the executive report should be largely written — not starting from scratch.

• This eliminates the crunch of "monthly reporting season" followed immediately by "quarterly reporting season."

The cadence varies by client. What's consistent is the orientation: every update points toward the quarter.

Who does what?

Reporting responsibilities are split by client tier.

How do I pull the report?

Pulling the report lives inside your client's Claude Project on claude.ai. Here's the basic flow:

1. Open the client's Claude Project
2. Start a new conversation and ask it to pull the report
3. Claude will load the client's reporting config and tell you exactly what data it needs and confirm any details
   
4. Paste or upload the requested data — Claude will walk you through it source by source
   
5. Add context when prompted: call notes, campaign details, anything from recent client conversations
6. Review Claude's draft, make edits, and flag anything that needs additional input
7. AM reviews and signs off before it goes to the client

When things go wrong

The AI reporting process is new, and things will occasionally break or behave unexpectedly. Here's how to handle common issues. 

Here's how to handle common issues.

How to submit a ticket: Submit all reporting process issues through this request form. You will receive confirmation that your request has been confirmed and as progress is made. 

How Claude works (the non-technical version) 

You don't need to be an AI expert to use this workflow. But understanding a few basics will help you get better output and troubleshoot when something feels off.

Claude is context-dependent. Claude doesn't have memory between sessions. Every time you open a report, it starts fresh — which is why the client's Claude Project exists. The Project stores the reporting config and client context so Claude always knows what kind of client it's working with. Always work inside the right client's Project. Running a report outside the Project, or in the wrong one, means Claude won't have the config and will produce generic output.

Claude is a collaborator, not an oracle. Claude is very good at structure, synthesis, and drafting. It is not good at knowing what happened on your last status call, what the client's CEO cares about this quarter, or why a campaign underperformed. That's your job. The process is designed around this: Claude asks you for context at the right moments. Give it real context — not just data — and the output will be significantly stronger.

The reporting config is the source of truth. Every client project has a reporting config file. This tells Claude how to report on that client: which data sources to ask for, what format the deliverable takes, what comparisons to make, and any known methodology quirks. If the config is wrong or missing, Claude will make its best guess — and that guess will probably miss the mark. If you notice the config is out of date, flag it and submit a ticket.

Tips for better output:

• Paste data cleanly — exported tables work well; messy copy-paste from dashboards can confuse Claude
• Be specific when adding context ("the client mentioned they're expecting a board meeting next month" is more useful than "the client is important")
• If a section comes back weak, ask Claude to redo it: "This insight doesn't explain why it happened — add the context I gave you about the LinkedIn campaign"
• Don't edit Claude's draft in the chat and then re-ask Claude about it — Claude won't see your edits; give new instructions instead