Platform Dashboard

The problem

In 2014, running a targeted direct-response campaign required stitching together three separate tools — a geo/demo selector, a forecaster, and a creative-management system — with manual handoffs, spreadsheet math, and no in-flight estimate of what the campaign would produce before you launched it. Small-and-medium advertisers could not touch this workflow; it was agency-only. The patent bundles the whole loop into a single GUI that a non-specialist can drive.

The mechanism

The dashboard lets a user draw a geography on a map, pick demographic parameters, and immediately see an estimated statistic (reach, impressions, expected cost) computed from the underlying user-identification and habit-index systems (Families 1 and 2). The user sets campaign parameters, associates creative, and launches in the same UI. The novelty is the tight-loop coupling between the inventory estimator and the campaign builder — the estimate isn't a separate forecasting tool, it's a live function of the very audience definition the user is editing.

What it proved

Self-service audience building works if the estimator is real-time and honest. The pattern — define → estimate → parameterize → activate, all in one surface — became the template for every modern ad platform (Meta Ads Manager, Google Ads, TikTok Ads Manager, Amazon DSP) and every modern data-ops tool (dbt Cloud, Census, Hightouch audience builders).

If it were built today

The human-driven dashboard as the primary control surface is already a legacy assumption. In 2026, the "user" of a Platform Dashboard is increasingly an AI agent — a campaign agent run by a marketer, a procurement agent run by a buyer, a benefits agent run by an HR system. The modern reinterpretation is a declarative intent interface ("reach pregnant households in Nashville metro with ≤$18 CAC") backed by an agentic planner that explores the parameter space, queries the estimator, runs counterfactual simulations, and returns a ranked set of campaign configurations — with the human only approving the plan. Under the hood the estimator becomes an LLM-callable tool (MCP server, OpenAI function, Anthropic tool-use), the creative-association step becomes generative (the agent drafts variants), and the dashboard GUI shrinks into an audit and override surface over an agent log. The cleverness migrates from "real-time estimate as the user drags a slider" to "real-time estimate as the agent reasons," and the hardest problem becomes provenance — every plan step must be explainable, reversible, and subject to human veto.