Build a Custom AI Campaign Automation System

Most agencies start with off-the-shelf tools like AdEspresso or Madgicx. These are great for creative testing and simple rules, like pausing an ad if its CTR drops below 1%. But their logic is confined to platform metrics. They cannot execute a rule like, "If this client's HubSpot MQL count from this campaign drops 20% week-over-week, decrease the budget by 15% and reallocate to the top-performing ad set from last month." The tools see ad data, not business outcomes.

In practice, a 25-person agency managing 100+ accounts runs on spreadsheets. A senior manager spends every Monday morning pulling data from Google Ads, Meta, and LinkedIn into a master Google Sheet. They manually calculate pacing and performance for each client, then write a summary. This takes hours of skilled labor, is prone to copy-paste errors, and the data is already a day old by the time anyone sees it.

Some agencies try to solve this with internal scripts, often a collection of Python files that run on a schedule. These are brittle. When an ad platform updates its API or a refresh token expires at 2 AM on a Saturday, the script fails silently. Thousands of dollars in ad spend can be misallocated before anyone notices on Monday. The script is a temporary fix, not a reliable production system with logging, monitoring, and error handling.

The structural problem is that existing tools are built for a single advertiser, not an agency. They lack the architecture to manage dozens of distinct client accounts, each with its own budget, KPIs, and data connections. An agency needs a central control plane that can apply unique, sophisticated logic to each client account at scale, something no off-the-shelf SaaS provides.

The first step is a workflow audit of your 3-5 largest client accounts. Syntora maps the entire process, from campaign creation to performance reporting. We identify the exact bidding logic your managers use, the data sources they consult, and the format of the reports they deliver to clients. This audit produces a clear technical specification for the automation system.

Syntora's approach is to build a central control plane using FastAPI and Python. This service connects to the Google Ads and Facebook Marketing APIs directly, giving us full control over bidding and reporting logic. Client-specific configurations, API keys, and performance data are stored in a Supabase database. For generating client-facing summaries, we feed structured performance data into the Claude API to create human-readable insights. The entire system runs on AWS Lambda, providing reliable, event-driven execution for tasks like daily bid adjustments.

The delivered system is a simple internal dashboard, hosted on Vercel, where your team can onboard new clients and define campaign parameters. The system runs automatically in the background, adjusting bids and generating reports based on the rules you set. Instead of discovering a 50% CPA spike days later in a report, your managers get an immediate Slack alert, allowing them to act on real-time data.