AI Automation: Project Spotlight
Their team works on one of the hardest problems in public health: improving access to care and services across the U.S.
To do that, they have to constantly answer a simple but painful question:
Who exactly do we need to reach in every state?
That question sounds straightforward until you try to answer it.
State agencies aren’t centralized. They’re scattered across websites, directories, org charts, PDFs, press releases, and outdated contact pages. LinkedIn fills in some gaps. Government sites fill in others. And none of it is structured.
So building a single outreach list wasn’t just time-consuming: it was inconsistent by nature.
In practice, it could take hours (sometimes days) just to produce a usable set of contacts. And even then, quality varied depending on who did the research.
That became the ceiling.
Not because the team lacked skill but because too much of their time was going into finding people instead of actually working with them.
The Problem with Doing It by Hand
Manual contact research breaks down the moment you try to scale it.
Accuracy varies by researcher. Classification is inconsistent. One person’s “key decision-maker” is another person’s “relevant stakeholder.”
Even on a good day, you might surface 15–20 usable contacts.
That pace doesn’t support national expansion.
And the real cost isn’t the hours — it’s what those hours prevent:
time spent building relationships, not digging for names.
What We Built
We built a multi-source AI contact discovery engine: a pipeline that finds, enriches, and classifies contacts automatically so the team never starts from a blank search bar again.
It runs in four stages:
Discover
Brave Search, Firecrawl, and LinkedIn run in parallel: pulling from agency sites, public records, directories, and professional networks.
Extract
Unstructured results are converted into clean data: names, titles, departments, organizations, and context signals.
Classify
An AI layer evaluates each contact based on relevance, authority level, and mission alignment, separating decision-makers from influencers from coordinators.
Deliver
The team gets ranked, structured, ready-to-use lists. No cleanup. No formatting. Just outreach.
The classification layer is where the leverage shows up.
Because it’s not just about finding contacts: it’s about deciding who actually matters at scale, consistently, without human bias or fatigue.
The Results
Before this system, the team manually researched contacts day by day: often inconsistently, and without a standardized way to prioritize outreach.
Now, instead of starting from scratch each time:
- Contacts are surfaced and compiled across multiple sources automatically
- Relevance is scored and organized instead of manually interpreted
- Lists are structured so the team can focus directly on outreach instead of research
What used to be a manual, repetitive process is now an automated workflow that removes the starting friction entirely.
And that shift matters more than speed.
Because the real win isn’t how fast contacts are found, it’s what the team gets back:
time to actually build relationships instead of assembling lists.
Why This Kind of Work Matters to Us
We’re not interested in adding “AI features” to broken workflows.
We’re interested in something simpler:
Finding where talented people are spending time on work that doesn’t require them and building systems that remove it.
In this case, the problem was clear:
find the right contacts, classify them correctly, deliver them fast.
The solution required:
- multiple data sources
- structured extraction
- an AI classification layer that applies consistent judgment at scale
In eight weeks from kickoff to production, the team can expand into new states without adding headcount just to do research. That’s the kind of leverage that compounds.
And it usually starts in the same place: A process that’s quietly become a bottleneck no one has named yet.
If your team is spending hours on work that should be automated, it’s usually not a tooling problem.
It’s a signal.