How a Financial Services Firm Got 180 Employees Building With AI in a Single Sprint

Most enterprise AI programs stall in the same place. Licenses get bought, a pilot team gets access, an all-hands explains the roadmap, and then very little happens. The tools sit unused because nobody outside the pilot team has a reason to open them, and the people closest to the day-to-day problems never get a chance to point AI at the work they actually do.

Stifel took a different route. Working with BeMyApp, the firm ran an internal program that paired structured AI training with a hands-on build phase, opened it to employees across experience levels and locations, and ended with 41 working prototypes aimed at real operational problems.

Here is how it was built and what came out of it.

The problem: training without application doesn't stick

Stifel's teams spanned a wide range of AI familiarity. Some engineers were already comfortable with agent frameworks. Others had never used an AI tool in a work context. A single training track would have bored the first group and lost the second.

There was also a stack problem. Generic AI training teaches generic AI. It doesn't teach people to build inside the specific environment their employer runs on, which means the skills don't transfer to Monday morning.

The program had to solve both: meet people at different starting points, and teach them on the tools they would actually use.

The approach: learn first, then build

The program ran in two connected phases:

Phase 1: Enablement

Expert-led trainings, live webinars and guided labs, all aligned to Stifel's own technology stack: AWS, Microsoft Copilot Studio, LangGraph and Salesforce Agentforce. Sessions were pitched at different experience levels so that a first-time user and an experienced engineer could each find a useful entry point. The labs mattered more than the lectures; participants worked inside realistic development scenarios rather than watching demos.

Phase 2: The Build

With the training complete, teams moved into a hackathon phase and applied what they had just learned to operational challenges they picked themselves. The challenge framing was deliberately open: build an AI-driven solution using Stifel's core technology stack that improves internal workflows, increases operational efficiency, or drives product innovation.

The sequencing is the part worth copying. A hackathon without prior enablement produces a lot of half-finished projects from the handful of people who already knew how to build. Training without a build phase produces certificates nobody acts on. Running them back to back means the training has an immediate destination, and the build phase starts from a shared baseline.

Delivery was hybrid, which let employees in different offices and time zones take part without travel budgets deciding who got to participate.

The results

• 180 participants took part across the enablement and build phases.
• 41 completed project submissions were delivered by the deadline, a completion rate that reflects the enablement phase doing its job.
• Mentors were embedded throughout, giving teams somewhere to go when they hit a wall rather than quietly dropping out.

The spread of what people chose to build is revealing:

Roughly a third of teams went after product capability, while the remaining two-thirds distributed across the agent and assistant tooling in Stifel's stack. For a firm trying to work out where its AI investment should concentrate, that distribution is a useful signal on its own; it shows where employees instinctively see the highest-value applications.

What teams actually built

1. MCP Gateway (Tracker): a secure gateway letting internal teams interact with MCP servers safely, improving access control and standardization across internal systems.
2. AcceleratePR: an AI assistant for pull request workflows that generates summaries and supports security and compliance documentation, cutting manual effort out of the development cycle.
3. Aletheia: a Copilot-driven approach to reducing false positives in application security findings, helping teams prioritize genuine threats instead of working through noise.
4. ASK IT Self-Service Support Agent: an agent that answers common IT questions and automates ticket creation, reducing support workload while improving response times.
5. Stifel Pension Builder with AI Agents: a tool that analyzes dividend safety across securities and surfaces transparent insights to support portfolio decisions.

Four of the five target internal friction: security triage, IT support, developer workflow, system access. That is what happens when you let the people doing the work choose the problem. Nobody assigned "reduce false positives in application security findings" as a challenge track. Someone who spends their week clearing those alerts picked it.

What made it work

• Training was tied to the real stack. Participants learned on AWS, Copilot Studio, LangGraph and Agentforce because that is what Stifel runs. Skills transferred directly.
• Multiple entry points kept the room wide. Pitching sessions at different experience levels meant participation wasn't limited to engineers who were already building with AI.
• The challenge framing stayed open. Prescriptive challenge tracks produce projects that answer the organizer's question. An open frame produced projects that answer the participant's question, which is where the operational value was.
• Mentors were available during the build. The gap between "I learned this" and "I built this" is where most internal programs lose people.
• Hybrid delivery removed a filter. Distributed teams could join without travel approval determining who participated.

Running something similar

The pattern applies well beyond financial services. Any organization with AI tooling in place, employees who haven't adopted it, and operational problems nobody has had time to solve is looking at the same setup.

BeMyApp designs and runs these programs end to end and fully white-labeled, from enablement content, participant recruitment, platform, and mentoring to judging and post-program reporting, so the program runs under your brand and your team stays focused on the outcomes rather than the logistics.

If you're working out how to move your organization from AI licenses to AI adoption, let's talk about what that would look like.