There is something almost suspiciously on-brand about a public relations executive writing a business plan on a McDonald’s bag. It is a story with texture, compression and a logo already attached. David Fuscus says that is how Précis AI began in 2022, at a highway rest stop. Yet the memorable prop can obscure the useful idea inside it. Fuscus had spent decades running Xenophon Strategies. He did not look at generative AI and see a machine that knew too little. He saw one that knew too much about everything and not enough about the peculiar work sitting on his own desk.
PR is full of tasks that appear simple from a distance: draft a release, map stakeholders, adapt a message, prepare for a crisis. Up close, every task is booby-trapped with voice, context, confidential information and reputational consequence. A fluent paragraph is not necessarily a publishable paragraph. A plausible fact is not necessarily a fact. The first thing to fail in a general chatbot is often not grammar. It is professional context.
“In Washington, being wrong and being slow are the same thing.”David Fuscus, founder and CEO
01 · The narrow wagerMake the machine learn the office
Précis AI’s answer was not to train one giant foundation model from scratch. It built a secure layer around several existing models, then filled that layer with professional workflows, domain prompts, private documents and tools that resemble the work itself. Précis Public Relations includes campaign and document management, a creation hub, research, content repurposing and a chatbot. Its DataVaults are designed to let a team use internal material without feeding that material back into public model training.
The distinction sounds modest until you watch someone work. A blank chat box makes the user specify the assignment, the audience, the format, the source material, the tone and the guardrails. A specialist product moves much of that setup into the software. It remembers what sort of room it is in.
The company’s reported results are striking, if properly labeled. Users told Précis and its partner, the Public Relations Global Network, that project times fell by an average of 61 percent. Heavy content producers reported getting 8.5 hours back each week. These are not independent laboratory measurements. They are more like odometer readings from working agencies - useful because the product was driven on real roads.
02 · The test benchTwenty-eight countries before the victory lap
Précis had an unusual place to test the thesis. The software was incubated inside Xenophon, where Fuscus could observe communications work from close range. Then PRGN agencies began beta testing it in October 2023. The network stretched the experiment across six continents and 28 countries, exposing the platform to different languages, clients and habits. In March 2025, PRGN and Précis announced a strategic alliance.
This was product development disguised as distribution. Most young software companies must first invent a customer persona and then hunt for people who resemble it. Précis began among practitioners and borrowed their complaints. Its management makes the blend explicit: Bo Hrytsak leads product and technical development; Chris Fuscus leads prompting and domain content; David Fuscus supplies the operating history. Doug Bennett, the CFO, brings a background in technology and private equity.
The market noticed. In 2025, Précis AI and Xenophon received a PRSA Silver Anvil Award of Excellence for Best AI Solution Provider. The honor mattered less as decoration than as a signal from the profession the product was built to serve.
03 · The second professionWillard goes to Washington
A company can claim to be a platform when it has one product. The claim becomes more interesting when the method survives a second industry. In June 2026, Précis launched Willard for lobbyists, government affairs teams and legislative staff. The name sounds like someone who knows which Capitol corridor saves three minutes. The software is trained around millions of federal and Congressional documents and connects answers back to official material.
The daily tools are concrete: member biographies and voting records, bill and amendment comparison, an AI hearing monitor, political analysis and a fact checker. The company says Willard’s agents coordinate 16 AI models, including models from OpenAI, Anthropic, Google and Perplexity. Highlight & Click Fact Checking asks models to challenge one another and ground the result in verified federal data. The system also promises end-to-end encryption, US data residency and no customer-data training.
| Work problem | General chatbot | Précis approach |
|---|---|---|
| Getting started | User builds the prompt | Workflow and domain prompt already exist |
| Private material | Policy varies by tool and plan | Documents sit in a customer DataVault |
| Verification | Usually one generated answer | Multiple models check claims against official records |
| Team use | Conversation-centric | Shared projects, files and repeatable outputs |
That puts Précis between two familiar categories. On one side are general chatbots, flexible and cheap to begin using. On the other are legacy PR and government-affairs databases, structured but often built for search rather than synthesis. Précis sells the connective tissue: professional context, creation tools, collaboration and security. Its commercial model is enterprise SaaS, sold through demos and trials; Willard’s early access is invitation-only.
04 · The copyable lessonDo not begin with the chatbot
The most portable thing here is not a feature. It is the sequence. Précis did not start with “Where can we add AI?” It began with work that experienced people could describe in irritating detail. The company then separated repetition from judgment. Drafting, comparison and repurposing became software tasks. Taste, relationships, accountability and final approval remained human tasks.
- Choose a profession, not a market adjective. “PR teams” is useful. “Knowledge workers” is fog.
- Borrow a real workflow. Build around the document, handoff and deadline people already recognize.
- Make private context a feature. The most valuable prompt is often the customer’s own archive.
- Measure time returned. Hours saved are easier to defend than claims about transformation.
The approach has boundaries. It is strongest where work repeats, domain knowledge matters and mistakes carry a cost. It is less compelling for a casual one-off draft, a user unwilling to supply context, or a field without dependable source material. Multi-model checking can reduce error; it cannot make judgment disappear. Private storage can lower one class of risk; it cannot rescue careless access controls or bad internal data.
This is why the paper bag matters after all. It represents the opposite of the universal interface: a plan written for a specific problem by someone close enough to recognize it. Précis AI may grow into more professions, but its case becomes weaker if it forgets the narrowness that made the first two products legible. The company is not really betting that AI will know everything. It is betting that, at work, knowing which details matter is worth paying for.