02 · Account Managers
Stop Exporting CSVs: How Account Managers Use Affise MCP to Run QBRs in Minutes
Ask your account data a question and get an answer. Say no to dashboards and pivot tables.
Account managers know the drill. A client calls on Friday afternoon asking how their top three offers performed last quarter. You pull up Affise, set the date range, pick your slices, wait for the export, open Excel, build a pivot table, and thirty minutes later you have a number. The client has moved on.
QBR season is even worse. You spend two days pulling data from different partners and advertisers, dropping it into slides, and writing up the story. By the time you finish the deck, the numbers are already a week old. This isn't a reporting problem, it's a workflow problem. And more dashboards won't fix it.
The Affise MCP closes the gap between asking and knowing. You type a question in plain English, and your AI assistant handles the rest: it calls the right Affise tools, gathers the data, and gives you an answer you can use right away. No exports. No fiddling with filters. No waiting around.
Meet the Affise MCP
The Affise MCP (Model Context Protocol) is a hosted server that connects Affise to AI assistants — Claude, Cursor, Continue.dev, and others. When you type a question, the assistant picks the right Affise tool, runs the query, and returns structured results in your chat window. You can follow up, slice differently, or ask for a written summary in one conversation. It lives at https://mcp.affise.com and connects via OAuth in about two minutes.
What You Can Do
QBR Prep in Minutes
Instead of building a deck from scratch, open a conversation and ask for exactly what your client review needs.
›"Revenue, conversions and CR by month for the last quarter."
›"Top 10 partners by revenue last quarter."
In two prompts you have the core numbers for a quarterly review. Follow up with a country breakdown or a partner-level drill-down in the same thread. What used to take hours of exports and formatting now takes minutes, and you can regenerate it any time the client asks for a different cut.
›[screenshot: chat showing a monthly revenue table followed by a top-partners ranking, no dashboards open]
Period-Over-Period Account Health
Clients want to know whether things are improving. Comparing two periods used to mean two separate exports and a manual delta column.
›"Compare this month vs last month: clicks, conversions, revenue."
The assistant runs both ranges and presents the comparison. You see movement at a glance and can immediately drill into whichever number shifted.
Early Churn and Decline Signals
The partners and offers that matter most to an account can deteriorate slowly: a gradual drop in CR, rising trafficback, shrinking volume. These signals often go unnoticed until they are already a problem.
›"Show trafficback for the last 30 days, top reasons."
›"Show retention rate for last 30 days by offer."
Running these checks takes seconds. If a partner's traffic is being rejected at a higher rate than last month, or retention on a key offer is declining, you see it before the client does. That gives you time to diagnose and respond rather than explain.
Turning Numbers into a Narrative
Raw tables are not a client deliverable. The analysis prompts bridge the gap between data and recommendations.
After pulling stats for a period, invoke the analysis prompt directly:
›"/analyze_stats on the last result with analysis_type: performance, format: actionable."
The assistant reads the data it just retrieved and returns a structured interpretation of where performance is strong, where it is weak, and what to do about it. The actionable format is designed for account management: it leads with findings that have a next step attached, not just observations.
For a deeper cut across an entire account, use the orchestration prompt:
›"/auto_analysis for partner 3554 last 30 days."
auto_analysis runs multiple queries — stats, trafficback, conversions — and synthesizes them into a single account picture. It is the equivalent of building a multi-tab report, condensed into one prompt and one response.
Across 5 active offers the account confirmed 11,908 conversions for ~$465 in the last 30 days. Volume is concentrated in CashApp_MX_AOS_CPA (10,910 confirmed) but at a 0.01% CR, while QuickBite_MY_AOS_CPA brings the best confirmed earning (~$181) at a 2.23% CR. The MegaShop_CL_CPI offers convert steadily; one variant is currently paused.
36.6M clicks bounced as trafficback — 98.8% of it overcap (traffic hitting capped offers), then mistargeting-country (308k) and mistargeting-os (146k). This is a caps + targeting problem, not partner quality.
- Raise caps or rebalance volume on
CashApp_MX_AOS_CPA— it takes the traffic but bounces 36.5M clicks as overcap at a 0.01% CR. Even a small cap increase is large upside. - Audit
QuickBite_MY_AOS_CPAbefore scaling: 486 declined vs 439 confirmed (~53% decline). Check the goal and targeting rules driving the rejects. - Fix geo/OS targeting (mistargeting-country 308k, OS 146k) or re-point that traffic to eligible offers — and decide whether to resume or drop the paused
MegaShop_CL_CPIvariant.
auto_analysis answer as it renders in the chat — one prompt pulls stats, trafficback, and offers and synthesizes a summary, a trafficback note, and prioritized recommendations. Real figures from the account; offer names are anonymized.Answering Live on a Client Call
A client asks mid-call how their volume compares to the same period last month. Previously the answer was "let me check and get back to you." With the MCP open in another window, you can answer in the time it takes them to finish the question.
›"Show me revenue by country last 30 days."
›"Time-to-action breakdown for last week."
The assistant returns results in seconds. You can read from the output directly or paste a summary into the call notes. Clients notice when their account manager has the numbers ready without a follow-up email.
Analysis Prompts: Your Account Management Superpower
The difference between an account manager who sends a table and one who sends a recommendation is interpretation. The Affise MCP analysis prompts are built for that second type.
analyze_stats with format: actionable does not describe the data. Instead, it tells you what the data means and what to do next. auto_analysis goes further, pulling multiple data types together so you are not just analyzing clicks in isolation; you are seeing clicks alongside trafficback and conversion patterns in one view.
These prompts accept a format parameter: summary for a quick read, detailed for a full breakdown, and actionable for the version that leads with recommendations. For client-facing work, actionable is usually the right call. It gives you draft language you can adapt for a slide or an email without starting from a blank page.
QBR Tables, Rendered in the Chat
In clients that support MCP UI apps, the numbers behind your QBR don't come back as plain text. Instead, they render as an interactive, sortable stats table right inside the conversation, ready to read out on a call or screenshot straight into the deck. Ask for the quarter-over-quarter breakdown and you get a grid you can scan, not a paragraph you have to parse. Where the client has no UI support, the same figures return as clean text.
Getting Connected
The Affise MCP connects in about two minutes. Add it as a custom connector in your AI client, point it at https://mcp.affise.com, and authenticate with one-click OAuth using your Affise credentials. Full step-by-step instructions are in the Client Quickstart. Once it is connected, every conversation you have with your AI assistant has direct access to your Affise account data.
Start With Your Next Client Review
Select an account that has a review coming up. Open your AI client, connect the MCP if you have not already done this, and ask for the last quarter's performance by month. See how long it takes. Then ask for the top partners, a period-over-period comparison, and an actionable analysis — all in the same thread.
The goal isn't to take over your decisions. It's to get you the numbers quickly, so clients value your insight, not your dashboard skills.