Case Study

Lenfest AI Collaborative and Fellowship Program: AI-Powered Prospecting Agent

An overview of how The Seattle Times created an AI tool to streamline ad sales + open source code

September 19, 2025

Local advertising sales teams face a common challenge: limited capacity to identify and qualify new prospects. Researching potential advertisers is slow, inconsistent, and resource-intensive—often leaving high-value opportunities untapped while also taking away from time in the field actually contacting prospects and closing sales.

Despite a two-year transformation project resulting in sharper KPIs, “rules of engagement” to cut non-selling time, and a tighter new-business focus, manual prospect research still soaked up hours and missed opportunities for The Seattle Times Media Solutions group. The team runs both owned-and-operated advertising and a full-service digital agency. The product set is deep. The processes are complex.

The Seattle Times set out to test whether a generative AI “prospecting agent” could help automate this process. Its goal was not to replace human sales expertise, but to free up capacity by streamlining lead research and preparation. If successful, the tool could accelerate revenue growth, make sales teams more efficient, and provide a model for other news organizations navigating sustainability challenges.

Their bet: A conversational AI agent could compress prospecting research from hours to minutes, surface better-fit leads, and prep account executives (AEs) for smarter discovery – fueling new revenue without adding headcount.

Early results are promising, including new nontraditional sales, efficiency improvements, and enthusiasm among the staff. According to Amber Aldrich, vice president of advertising, in the first week “one AE used our prospecting agent, found a nontraditional lead we would’ve never called on, and closed an immediate sale.”

Product design & development

Approach: The Times designed an AI-powered agent that automates prospecting tasks by researching potential advertisers, categorizing leads, and preparing background intelligence for reps.
What it does:
- Generates targeted lead lists by category + geography
- Checks conflicts (“is this already being worked?”) via CRM
- Scores leads against a curated list of Seattle Times specific criteria (≈ 40+ = jump on it)
- Assembles Lead Evaluation and Business Research reports with sources/citations
- Drafts discovery questions so reps walk in prepared
How it works (plain-English stack):
- Interface: Custom GPT inside ChatGPT (conversational, familiar UI)
- Agent brain: Heavily tuned system prompts per task (score, report)
- Systems bridge: Azure-hosted API server (“window to our world”)
- Data: Adpoint (specialized ad CRM) for history/contacts/status; public web for citations
Process: Early iterations focused on quickly proving value: could the agent save meaningful time and deliver usable leads? Frequent account executives and sales management input and testing made the process iterative, allowing for faster decisions. The team quickly pivoted to a ChatGPT interface to improve effectiveness, blend-in the agent functionality with their default lead research tool (ChatGPT), and make adoption seamless.
Team:
- AI Fellow (technical lead) developed and refined prototypes.
- Sales leadership defined workflows and validated outputs.
- Engineers and product staff supported design choices and integration.
- Cross-functional feedback loops ensured the solution was grounded in day-to-day sales needs.
Partners: Development was led in-house, with insights from Microsoft, OpenAI, and other members of the Lenfest AI Collaborative community.

Key learnings

Successes

Time saved: Sales teams reduced prospect research time from hours to minutes.
New leads identified: Soon after rollout, a sales rep used the tool to spot a nontraditional prospect, leading to a direct sale.
Improved call prep: “I’m walking into calls more organized and professional,” one sales rep said.
Adoption: Initial skepticism gave way to enthusiasm once teams saw tangible results.
Scalability: The agent demonstrated capacity to process and structure hundreds of prospects at once with scores and decision-maker intel.

Challenges

Data reliability: Early prototypes surfaced accuracy issues, requiring validation steps to filter and confirm results. Add validation and citations to counter early accuracy anxiety.
Solution pivot: Shifted from a Copilot-style interface to a custom ChatGPT solution, embedding agent capabilities into the manual prospecting tasks users were already performing in ChatGPT.
Change management: Building sales staff confidence in AI required tailored training and guardrails. Teach people to “fish” with conversational queries (as prompts that are too short or too long will both fail).
Integration complexity: Connecting prototypes with CRM systems and ensuring compliance added layers of difficulty beyond early testing.

What’s next

Short-term roadmap: Develop separate AI lead-qualification pipelines for select target client segments (education, nonprofits, etc.), with their outputs (qualified, enriched, and prioritized leads) made available in the Prospecting Agent for users’ prospecting efforts; expand from prospecting to pitch drafting and account intelligence.
Key questions:
- How can automation scale without flooding reps with low-quality leads?
- What balance between speed and accuracy creates the most sales value?
- How can AI be applied to retention and upselling strategies?
Looking ahead: A proof-of-concept for full-cycle AI-assisted, human-driven sales—from lead identification to tailored pitch support.
Related product in pilot: The Seattle Times is developing an InfoHub that puts specs, deadlines, case studies, training docs, and business rules into a single conversational search for the whole ad sales department—speeding onboarding and reducing manager lift. How it works:
- Multi-agent Copilots built on Microsoft Copilot Studio over SharePoint grounded data (no heavy lift)
- Light curation to prune duplicates; guardrails via prompt pattern

Why this matters

The Seattle Times’ project demonstrates that AI can support the core business model of local journalism by looking at solutions outside the newsroom. By supercharging sales teams with faster prospecting and richer intelligence, AI can directly strengthen newsroom sustainability—and offer a replicable model for other local publishers.