Every Company That Sells to Government Will Run on NationGraph

What does it actually take to sell to the public sector?

Since founding NationGraph, Eden and I have sat through over a thousand conversations with companies that sell to the public sector. The types of companies change. Their work all looks the same.

Take a rep selling to transit agencies. There are roughly 1,800 public transit systems in America, inside a market of 110,000+ agencies that also includes ~19,000 cities and towns, ~3,000 counties, and ~13,000 school districts. Whatever the vertical, their week starts with the same questions:

• Which of these are worth my time this year?
• Who has the contract today, and when does it end?
• What did the board say three weeks ago, and is it on page 47 of the minutes or buried in a two-hour video?
• What happened overnight (a grant awarded, a budget amended, an official resigned, an audit failed) that should reshuffle who I spend today on?
• What is the agency actually paying the incumbent I'm trying to displace?

Every answer is public. And every answer costs hours of a skilled person's day, because it lives across thousands of websites, portals, and archives that were never designed to talk to each other.

A thousand conversations, one pattern: teams spend more time researching than selling. Some tell me it's most of the week. Some put it at 90%.

So teams pull people off quota to build and babysit homegrown systems: data feeds, alert subscriptions, and spreadsheets stitched into something that's supposed to do the work automatically, and never quite delivers. We've met sellers so burned by bad contact data that they write their own scrapers at night. Nobody takes a sales job to build software. But here they are.

Leaders who do the math find each qualified lead costs four to five hours of research. A full workday of finding, every day, before a single selling conversation happens.

None of this is relationship building in the field and all of it is the laborious work of finding information. Selling to the public sector is expensive because knowing is expensive.

Where agents are good, and where they're not

Agents are built for a particular shape of work. They can read ten thousand pages without getting bored, watch every board meeting in the country at once, cross-reference a budget line against a board minute against a contract record, and remember every expiration date, and they will write all of it into your systems cleanly and do it again tomorrow without being asked. What that work has in common is public sources, clear goals, and a definition of done. Look back at the rep's week, and the finding half is exactly that shape.

What agents are not good at, and shouldn't pretend to be, is the other half of the work. The public sector runs on relationships, trust is the currency, and the best teams in this market spend their weeks in the field, shaking hands and discussing agency problems over lunches, earning the kind of trust that closes deals years later. No agent walks into a county building or earns a handshake.

Veterans of this market will tell you most of what they win was won before the RFP dropped. They were in the room early, understood the problem, and earned the trust that shaped what the eventual requirements said. Show up early and the relationship decides the deal. Show up late and all that's left to compete on is price.

So the job splits cleanly in two, one half machine-shaped and the other half human-shaped, and for fifty years both halves have been done by the same expensive and skilled person.

A great agent takes two ingredients, deep domain expertise and data verified over years, and that is what NationGraph is: the graph of all public sector data. The companies that sell to the public sector will run the unglamorous, laborious half of this job on NationGraph agents.

Most agents are shallow, and this is hard to get right

If agents are so obviously suited to this work, why isn't everyone already selling this way? Because most agents are shallow, and an agent is only as good as the data underneath it. Ask one a question and you get your own data back, rearranged. Ask which accounts you should call and it checks your CRM, ask where a deal stands and it reads your notes, which makes it a search bar pointed at your own data.

Ask a general-purpose model (Claude, GPT, Gemini) which accounts are ready to buy and it might come back with Washington. There are 88 places in America called Washington, so which one held the budget hearing, and which one's fire department controls its own money? Knowing exactly which Washington said the thing, in which meeting and under whose budget, is the unglamorous problem of entity resolution, and it decides whether an agent is useful or just confident.

A real agent has to clear a higher bar. It owns its domain completely, works however you want to work, and gets better every time it runs. What separates the agents that clear that bar is the depth of what sits underneath them.

The signal that decides a deal might be two hours into a city council's YouTube recording, where the auto-transcript spelled the company's name wrong. Search for "Cintas" and you will find nothing, because the machine heard "sin tus," so a keyword tool misses it forever while an agent with real depth catches it, because it understands what was said in that room and why it matters to you.

Few domains are deeper or messier than public sector data, which is why nobody built this until now. Mastering it takes careers spent walking the halls, and then years of engineering to turn that mastery into a map.

What we built

We spent two years building the thing underneath the agent, a living map of all 110,000+ state, local, and education institutions in America, with budgets, board meetings, expiring contracts, buying signals, public records, verified contacts, and news all indexed into one knowledge graph. What sits on top is Compass, our agent.

Why a graph, and not just a database? Because government agencies are nested and the relationships are not always clear. Sell to fire departments and you'll learn fast that many don't control their own budgets, and that the city or the county does. Ask a general-purpose AI what's happening at the fire station and you'll get nothing, because the answer isn't at the fire station, it's in the county budget hearing two levels up. The graph holds those relationships, which agency answers to which, whose budget funds whom, and which board approves what, so the agent can navigate government the way government actually works.

"Living" matters as much as "graph," because government data moves constantly as budgets get re-passed, officials rotate, and contracts expire. Scrape it once and it starts rotting the same day, which is why the contacts in generic databases bounce, so something has to never stop reading the record. Ours doesn't. The longer Compass works this terrain, the sharper its understanding of how government operates, and the map gets truer while every static alternative goes stale.

What running on an agent looks like

You don't use Compass like a search engine, you orchestrate it like a team. You describe the goal. "I need to understand which of my accounts in my territory are pre-RFP based on the last time they bought what I sell." Or: "we're launching a new inspection product for mid-size counties, build the plan."

The agent comes back with a better plan than you would have written, because it already knows which counties have the problem, which ones control their own budgets, and which ones are funded. Then it does the work, reading the budgets, the board minutes, the meeting videos, and the contract records across every source, including the ones where the transcript spelled the name wrong.

It figures out who holds the incumbent contract, what they charge, and when it expires. It finds the actual people, the ones who aren't on LinkedIn, and verifies them. It writes accounts, contacts, and context into your CRM cleanly and with no duplicates, and it drafts outreach in the language of each agency's own documents. The follow-up plan comes with dates attached: the renewal window, the budget hearing, and the next board meeting where the topic resurfaces.

And then it runs the plan, so when the county administrator says the sentence that matters in a meeting nobody had time to attend, it moves. It reads, reasons, writes, and acts, and every unglamorous step that used to eat the week is inside that loop. This is what we see at every launch: teams respond to opportunities they'd never have found on their first day live, and demos get booked in the first 48 hours.

Why this is inevitable

For fifty years two constraints defined this market. The public record was opaque, navigating it took expensive human labor, and building the map underneath it would have cost too much to attempt. Both constraints broke at the same time, when the record became machine-readable and AI became capable of reading it, reasoning over it, and acting on it.

Just as every company needed a website to be the front door of its digital store, every company that sells to government is going to run agents that deeply understand how government works, so their people stop hunting and start showing up early, where deals are decided. The companies adopting first are simply going to open a gap, earlier to every conversation and deeper in every room, compounding quarter after quarter until it gets very hard to close from behind.

The reclaimed half of the week is for exactly that, and the relationship gets deeper as a result. An agent that understands your product against an agency's current and future reality, its budget pressure, its board politics, and the legislation about to hit it, means you walk into the room speaking to their needs instead of delivering a pitch. The agent buys back the time to build the relationship, and makes sure you're never late to it.

The work is worth it

Selling to government is some of the toughest selling there is. The cycles run years, the record is buried, the buyers are hard to find, and the opportunity you just discovered was probably decided twelve months ago by somebody who showed up before you did.

But the companies that do this work matter more than most of the economy, because they keep the fire trucks running, the schools online, and the water clean. For decades they've carried every portal, every board meeting, and every records request on manual effort and relationships, because knowing was expensive. It isn't anymore.