Achieving 5x faster client reporting with AI agents

The objective

Twenty-First Digital manages paid media campaigns for 25+ publisher clients across regional magazines and national publications. They needed to eliminate several hours of reporting work each week while improving consistency and quality. The goal was an automated system that could create client-ready performance reports. It could analyze campaigns against past baselines. It could also keep each client’s unique tone and context.

The gap

The manual process was broken at every level. Weekly reports took several minutes per client to generate - manually entering budgets, goals, and campaign context every time. There was no clear benchmarking system. The team relied on memory and scattered spreadsheets to define “good” performance. Client-specific context lived in people's heads, not documentation. Report quality varied depending on who wrote it.

The cost included hours from specialists. It also included ending existing tool subscriptions. It also carried a risk of inconsistent reporting. Consequently, they needed to rebuild the entire reporting stack from scratch.

Knowledge base architecture

The first thing we did was audit what data existed. TFD had 2+ years of campaign performance data in Databox, but no systematic benchmarking. Client profiles and tone preferences existed only in team members' heads or scattered email threads. There was no single source of truth.

We built a knowledge base system using Claude Projects, one dedicated instance per client. Each project contains:
‍
• Campaign Reference
Active campaigns with goals, budgets, flight dates, and searchable identifiers
‍
• Historical Benchmarks
2+ years of performance data extracted from Databox and organized by campaign type
‍
• Client Profile & Tone Guide
Target audience characteristics, brand voice, language preferences, separate guidance for internal vs external communications
‍
• Report Format Template
Exact structure with examples showing expected output
‍
A critical discovery: historical benchmarks varied dramatically by campaign type and publication category. Some campaign types performed 3-4x better than others, and averaging them together created meaningless baselines. We had to segment benchmarks by multiple dimensions - campaign objective, publication type, audience characteristics - to create useful comparison points. This level of nuance was impossible with the solution.
‍
The knowledge base approach solved a core problem: how do you give AI the context it needs without rebuilding it every time? By storing campaign data, benchmarks, and client preferences as markdown documents, we built a system. This system works well for people. It also works well for machines. AI can query it instantly. The team can edit it when things change.

Live data integration & intelligent analysis

We integrated Databox's MCP server to pull real-time campaign metrics — giving the system access to current performance across all standard advertising metrics, historical trends for week-over-week and month-over-month comparisons, and custom calculated metrics specific to TFD's reporting methodology.

From there, the analysis engine does the heavy lifting. It automatically:

• Calculates pacing status based on configurable thresholds (campaigns flagged as on track, at risk, or off track)
• Compares current performance against historical benchmarks
• Provides client-specific context
• Adjusts tone and depth depending on whether the report is internal or client-facing
‍
The key innovation is contextual performance evaluation. Instead of simply reporting "this campaign cost $X," the system answers the more important question: is that good or bad for this campaign type, this client, and this history? That's the question TFD's team was spending hours manually answering every week.

Results so far

Beyond the time and cost savings, the system also solved the institutional knowledge problem. Historical context is now pulled automatically rather than relying on team memory, CPL benchmarks are quantified by campaign type instead of guessed, and every report maintains each client's specific voice regardless of who's running the account.

What's next

• Full Portfolio Rollout
Expanding the reporting system to all 25+ clients. The 2-client pilot validated the approach; now scaling knowledge bases and report templates across the entire portfolio.
‍
• Proactive Monitoring & Alerts
Automated daily checks that flag campaigns requiring attention. Slack notifications for pacing problems, budget overruns, and performance anomalies - shifting from reactive to proactive campaign management.

• SEO Audit Service
We use our competitive intelligence tools to deliver technical SEO audits for TFD’s publisher clients. We use QA flow, our in-house automated website quality assurance platform.

Key learnings