From Informal Help to Structured Practice

New agents finished a month of corporate onboarding and a week of regional practice before taking customer calls. That preparation covered the organization’s own products well. It did not cover the partner-carrier side of the job — the quoting, binding, and documentation agents are expected to handle when a customer’s situation falls outside the primary product line.

Those transactions still came up. When they did, new agents stopped and asked an experienced colleague, often with the customer still on the line. The help was generous and it worked, but it interrupted two people’s work, slowed the transaction, and left the newer agent no better prepared for the next one. This was never a knowledge gap anyone had named. It was a workflow quietly dependent on informal help.

I was the instructional designer on this project and a frontline agent doing the same work — which shaped both what I could see and what I had to verify rather than assume.

I ran the analysis, surveyed four recently hired agents, and consulted regional leaders and subject-matter experts. From that I defined the performance sequence — what an agent actually does, in order, to complete one of these transactions — then designed the learning architecture and the assessments, developed the prototype and the point-of-need support, coordinated review, and revised the program through alpha and beta testing.

1. I kept a topic in scope that the learner survey said to cut.

Three of four surveyed agents said they already understood brokered-product qualifications, so the first draft skipped them. The regional sales leader disagreed. That leader reviews underwriting outcomes and sees what happens when a brokered policy is written that did not qualify: qualification errors can trigger underwriting corrections and avoidable rework. Self-reported confidence and observed errors pointed in opposite directions, so I followed the error evidence. Qualifications became a short pre-assessment and a review built around real cases where the rules were not followed. Self-reported confidence is not evidence of accurate performance, and this was the cheapest correction in the project.

2. I chose the job aid over memorization.

Agents work with multiple carrier systems, and each has its own eligibility rules. Those rules change, and some carriers come up only a few times a year. Asking new agents to memorize all of it was not realistic. The training teaches them to find the answer quickly instead: a redesigned screen-side job aid, plus a pointer to the company’s existing repository of underwriting updates. In beta testing, reviewers named the job aid as the resource they expected to use most after training.

3. Practice happens in the real systems, not on slides.

Agents already work with two monitors. The facilitator demonstrates on one screen, and learners repeat each step in the live carrier system on the other. The sequence moves from demonstration to guided practice to independent work, and it follows the order of a real sale: choose the carrier, quote, bind and take payment, then complete the signature documents. Agents practice in the same order they will perform the work.

The obvious alternative was a tour of partner products. I organized the program around the transaction instead, because agents were not stalling for lack of product facts — they were stalling at specific handoffs, and a catalogue would not have found those.

4. The facilitator asks “why,” not just “what.”

Assessment follows one customer through the complete workflow rather than testing isolated facts, so the learner has to demonstrate the whole sequence rather than recall its parts. Beta reviewers noticed that learners could pick the right carrier without being able to explain the choice, so I rewrote the facilitator guide’s prompts around the reasoning: “What information led you to that decision?” and “Why is that option right for this customer?” Carrier selection is a judgment call on the job, so the practice had to test the judgment at each decision point, not just whether the task was finished.

5. I designed for the newest learner, not the most experienced reviewer.

In alpha testing, the subject-matter expert and the regional sales leader rated usability 5 out of 5. The new agent rated it mostly 3s and asked for clearer cues about where to go next. Experienced users already knew the workflow. The new agent, the actual audience, did not. I added navigation cues, standardized the transitions, and wrote directions before each assessment.

6. I identified a barrier training couldn’t solve.

A prerequisite onboarding step is completed well before the point where agents actually need it, so some reach this training without working system access. More instruction can’t fix a sequencing problem in another process, so I documented it for stakeholders instead of adding course content.

The proposed experience combined virtual instruction, existing self-paced material, demonstration, realistic practice, assessment, and support an agent could reach at the moment of need.

I used generative AI as a design partner to explore options, draft content, and challenge my thinking. I did not treat its recommendations as final design decisions.

Used AI for: brainstorming scenarios and interactions, refining learning objectives, drafting assessment items and feedback, restructuring facilitator language, and generating alternative approaches during development.

Decisions that stayed mine: the scope of the training, the carrier-selection logic, the sequence of job tasks, the role of the job aid, the assessment strategy, and the final choice of media and interactions.

What I rejected or changed. An early AI-generated activity framed carrier selection as if each customer had one correct carrier. I rewrote the decision logic to match the actual work: establish eligibility first, evaluate product fit second, and consider price only after those requirements are met. The learner’s task became listening for the customer statement that rules out a branded product, investigating further, and identifying the best starting point for that customer — not hunting for one universal “right” carrier.

I also rejected a suggestion to open the first lesson with short HeyGen explainer videos. Rise interactions communicated that information more efficiently, so I saved video for the customer scenarios, where seeing and hearing the conversation added realism and instructional value.

Both corrections pointed to the same lesson: AI produced useful possibilities quickly, but each one still had to be checked against the job, the learners, subject-matter expert input, and the performance objectives.

I completed alpha and beta testing but did not implement the program across the organization. No performance or business results exist, and none are claimed here.

Alpha review. A subject-matter expert, the regional sales leader, and a new agent reviewed the course. They rated satisfaction, aesthetics, and perceived effectiveness 4.8–5.0 out of 5. These reviewers tested the prototype; they were not learners in a live rollout.

Beta review. Five reviewers represented instructional design, subject expertise, sales leadership, a new agent, and an experienced agent. They described the course as well organized and relevant to the job, and called for refinement rather than redesign.

Beta feedback produced concrete changes: more visual and interactive support, clearer on-screen text, and less unnecessary repetition.

A front-end survey of four new agents established the starting point: two felt “not at all” prepared to sell brokered products, none felt “very prepared,” all relied on an in-office colleague for help, and none had received training in the carrier quoting and binding systems.

Because the program has not been rolled out, I would evaluate its success at four levels:

The sample below is one piece of the larger design: the customer-scenario practice. It shows how real conversations, decision points, feedback, and follow-up questions were translated into interactive practice.