From Backlog to Breathing Room
At USPS, the challenge was straightforward but massive. Millions of customers rely on the agency daily, and even a small percentage of missed calls can quickly become a public-facing problem.
Rather than hiring their way out of it, Colón’s team focused on improving how agents work. By deploying Salesforce’s Agentforce platform, they aimed to make it faster and easier for employees to find accurate information within a sprawling knowledge base.
The early payoff has been increased efficiency — particularly as AI begins to handle routine inquiries and assist agents in real time.
The idea isn’t to replace workers, Colón emphasized. It’s to give them leverage.
“This isn’t an agent replacer. It’s about making our people more effective,” he said.
READ MORE: Identity security is paramount for federal AI adoption.
Rethinking What “Basic” Service Means
At the Department of Labor, Tonya Slater Lowe is tackling a similar challenge from a different angle.
As director of the National Contact Center, she oversees support for 26 programs, each with its own requirements, audiences and performance metrics. Historically, that has meant a traditional tiered service model: basic questions at the front lines, complex issues escalated upward.
AI is changing that structure.
“Anything that can be documented and written and placed in an article, it’s tier 0,” Slater Lowe said.
In practice, that means an AI assistant can now handle not just simple FAQs but also some escalations and referrals that once required human intervention. The result is a shift in how work is distributed across teams.
“It doesn’t take jobs. It complements the work,” she said.
For employees, that creates space to focus on more nuanced, high-value interactions — the kinds of conversations that benefit most from a human touch.
The Unseen Work: Getting Data Ready
If there was one point both leaders returned to again and again, it was this: AI is only as good as the data behind it.
At USPS, that meant rethinking how knowledge articles were written and structured. Content designed for humans — who might skim or search using keywords — doesn’t always translate cleanly to AI systems.
“A lot of the work comes in the preparation,” Colón said.
That preparation includes everything from reorganizing content to refining prompts and gathering feedback from front-line agents. USPS even brought agents into early testing environments to validate whether AI-generated answers matched what they would provide themselves.
At the Department of Labor, Slater Lowe described a similar foundation — one built over years, not months.
Her team maintains a centralized knowledge base supporting thousands of contacts across multiple programs. That structure, she said, has been critical to preparing for AI.
“If you do that work up front, it will be so much easier,” she said. “Get your house in order.”
LEARN MORE: Federal agencies can prepare data for AI applications.
No Shortcuts to Scale
Despite the promise of AI, neither agency is treating implementation as a quick win.
The Department of Labor is still scaling its capabilities after completing an initial proof of concept. Testing involves evaluating a wide range of user inputs and tracking performance across dozens of metrics, from containment rates to cost per transaction.
USPS has taken a phased approach, starting with internal use cases before expanding into more advanced capabilities, such as incorporating operational data and eventually analyzing live customer interactions.
Along the way, there have been bumps.
“There’s no perfect IT implementation,” Colón said with a laugh.
That willingness to iterate — and to accept imperfection — is part of what makes these efforts sustainable.
Bringing People Along
Technology alone won’t make such transformations stick. Both leaders stressed the importance of communication, inclusion and trust.
At USPS, Colón’s team worked closely with employees and union representatives to explain what AI would — and would not — do. Agents were invited to test the system early and provide feedback, helping to shape the final product.
At the Department of Labor, Slater Lowe has focused on building cross-functional teams and identifying internal champions who can advocate for the technology.
“You will get perspectives that you never thought you would get,” she said.
In both cases, the goal is the same: Make employees part of the process, not subjects of it.
DIVE DEEPER: Change management plays a role in AI adoption.
A Practical Path Forward
For federal leaders watching these efforts, the takeaway isn’t that AI will solve everything overnight. It’s that meaningful progress is possible with the right groundwork.
Colón offered a simple starting point: Don’t be afraid to try.
“We’re already using AI in our daily lives,” he said.
Slater Lowe’s advice was just as direct, if a bit more pointed.
“Don’t talk about the house,” she said. “Get your house in order.”
Between those two perspectives lies a pragmatic roadmap — one where AI isn’t a silver bullet, but a tool agencies can use to steadily improve how they serve the public.
