Loading page…
The tool links ad-platform activity to pipeline outcomes, but its design puts calculations in code and campaign decisions behind a human approval step.
Listen to this story
LangChain has open-sourced its Slack-based Paid Media Agent, a template for connecting ad-platform metrics with leads, opportunities and pipeline data. The system uses deterministic code for calculations, date windows, account matching and safeguards, leaving the model to interpret results and suggest campaign changes. Humans must approve those changes. LangChain reports a reporting run fell from 1,112 seconds to 85, with roughly 40× lower cost.
The agent can recommend keywords, targeting, ad copy and search campaigns, but cannot apply changes without human approval.
An early report used 3.9 million input tokens, cost just over $3 and took 1,112 seconds before redesign.
LangChain reports paid media grew from 0% to 20% of marketing pipeline in six months; CPL fell 30% as spend rose 60%.
LangChain says it reduced an early paid-media reporting workflow from 18 minutes to 85 seconds. It has now open-sourced the Slack-based Paid Media Agent behind that work, giving teams a starting point for analyzing ad performance and proposing changes without letting the agent apply them on its own.
The agent combines advertising-platform data with lead and pipeline data from LangChain’s warehouse. Every Monday, it posts a summary and branded PDF for each platform; the marketing team can then ask it follow-up questions in the corresponding Slack thread.
It can propose keywords, targeting changes, ad copy and new search campaigns. LangChain routes those campaign changes through human approval, making the agent a recommendation system rather than an independent buyer of ads.
LangChain’s initial workflow asked a model to read raw campaign, keyword, pipeline and landing-page data, perform calculations and write the report. On a frozen test set, the company says one report consumed about 3.9 million input tokens, took 1,112 seconds and cost just over $3.
The revised system assigns repeatable work to deterministic code: calculations, date windows, account matching and safeguards. The model instead evaluates campaign performance, explains likely causes and recommends next steps using the results it receives.
The release includes a pattern, not a ready-made marketing operation. LangChain’s own workspace relies on company-specific campaign knowledge and operating rules. Its reported performance gains are from that internal deployment: paid media rose from 0% to 20% of its marketing pipeline in six months, while cost per qualified lead fell 30% from June to August as monthly spend increased about 60%.
LangChain also reports saving about $5,000 a month by bringing analysis and reporting in-house rather than using an agency. The more transferable lesson is its division of labor: systems of record own the metrics, code enforces fixed rules, and people approve changes that affect a live campaign.
Loading discussion...
Join the conversation
Explain where you would keep or remove the approval step.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.