Building an Always-On Prospecting Engine with Agentic AI

Challenge

Freese Project Solutions, a Twin Cities professional services business providing Owner’s Representative services to manage corporate tenants’ office-space buildouts from architect selection through move-in, had a go-to-market strategy driven by referral relationships. When a company signs a lease on new office or light-industrial space, the buildout that follows is Freese’s service window, and the broker who closed that lease is the path to a new customer. But the firm had no dedicated business-development capacity yet, so deal announcements surfaced and expired unseen, the brokers behind them went unmapped, and the short window between a lease signing and the assembly of a buildout team closed without a conversation. The firm needed the market watched, understood, and turned into warm, well-timed introductions, without adding headcount.

Approach

Trexin’s first move was strategic, not technical: embracing the notion that the broker, not the tenant, is really the ongoing prospect, and recognizing that a broker’s lease announcements are the triggering event to reconnect with that broker. After all, a tenant lead is consumed once, but a broker relationship produces referrals for years. That framing shaped Trexin’s whole solution approach: every detected deal now enriches a permanent broker intelligence graph first and generates outreach second, and lease announcements (when buildouts are usually not yet staffed) outrank construction permits (when a buildout team has typically already been hired) as sales-triggering signals.

Trexin built the solution on an agentic AI architecture using Anthropic’s Claude Managed Agents: four cooperating agents (a daily signal scout, a lead curator, a weekly brief builder, and an on-demand deal-brief generator) sharing persistent Claude memory stores for the deal log, the broker graph, and a read-only playbook holding every business rule, including a scoring rubric, a source whitelist, outreach rules in the Client’s voice, and hard AI guardrails.

Intentionally, broker outreach is never sent automatically; every communication is presented as a fully written draft for Freese’s team to review and personally send. And the solution’s user interface is deliberately just email. A weekly brief and same-day hot-lead alerts arrive in the user’s regular inbox, and the user replies to any brief or alert in their own natural language, which the agentic system parses back into its memory, adjusting status, tactics, and strategy accordingly. No new app to learn, no CRM to feed.

Outcome

The system went from kickoff to unattended production in the Client’s own Claude environment in under a month, validated against a written verification checklist. It scans the Twin Cities market every business day and delivers a weekly prospect brief every Monday morning, and the Client steers it by replying to those briefs from his inbox.

Measurable outcomes Freese Project Solutions achieved in its first weeks of production:

• Cost reduction: ~$50,000 a year. Freese avoided a planned $52,000 annual spend on a new employee or outside service to scan the market for contacts and projects. The agentic AI engine now does that work for under $720 a year in compute, a reduction of 98%.
• Time saved: 5 hours a week of founder time. Before the agentic AI engine, the founder spent at least five hours every week scanning for contacts and project signals himself. In a young firm the founder’s hours are the scarcest resource in the company; those five now go to client delivery, broker relationships, and the growth work that only he can do.
• Sales channel expansion: 1 new referral partner. Within a week of the agentic AI engine going live, Freese identified a previously unknown brokerage specializing in tenant representation and investment sales, and opened the relationship.
• Time to market: new deals seen within 1 business day. Freese now learns of lease and expansion announcements across the Twin Cities the morning after they appear, with the broker identified and an introduction drafted, instead of weeks later or not at all.
• Extended market intelligence: 0 to 77 broker dossiers in 2 weeks. Freese went from tracking no broker relationships to holding backgrounders on more than six dozen brokers, 18 of them built on day one, each maintained automatically as new deals surface.
• Enhanced lead management: 33 active deals. Freese went from opportunistic, one-off outreach to a continuously scored pipeline, 10 deals qualified on day one, each ranked against Freese’s own rules and re-ranked as those rules change.
• Continuous improvement: 2 scoring rules changed in 2 weeks, by email. Freese refined how the agentic AI engine ranks deals simply by replying to its briefs, no meeting or change request required.

Trexin continues to operate the system. We monitor the daily runs, resolve issues, and adjust the rubric and features as Freese’s market focus evolves, so the Client gets the results without owning the upkeep.

“I went from hoping the right deal crossed my radar to opening a Monday email that already knows my market: which deals are live, which brokers are behind them, and what I should say to them. It’s business-development capacity we simply didn’t have before.”

Ben Freese, Founder, Freese Project Solutions

Why Trexin

Organizations whose demand signals appear in public announcements, from real estate services and construction trades to advisory, staffing, and commercial insurance, share this shape: the deal is public, the window is short, and the relationship is the durable asset. Trexin brought the strategic reframing that shaped the system, the production disciplines that made it trustworthy (human-in-the-loop gates, audited guardrails, provenance on every fact), and the engineering to take an agentic AI system from kickoff to unattended operation in weeks. We get enterprise AI from pilot to production.