From Informal Help to Structured Practice

The gap onboarding left

New agents at this organization complete a month of corporate onboarding, then a week of regional practice in their own offices. Onboarding covers the products the organization underwrites itself — the systems, the quoting path, the documentation. It does not cover the partner carriers whose products those same agents are also expected to sell.

The consequence was not that agents failed. It was that they stopped. Faced with a customer who needed a partner product, a new agent had no reliable way to identify the right carrier, reach its system, or complete the paperwork — so the transaction paused while they found someone more experienced. That interruption cost two people at once, and it happened in front of the customer.

What the analysis found

I surveyed four newly hired agents with seven questions about what their training had and had not prepared them to do. The pattern was consistent. Most had not been introduced to the partner carriers at all. None had been trained on the systems used to quote and bind their products. When asked how prepared they felt immediately after training, not one chose the highest rating. Asked who they turned to for help, every respondent named the same source: another agent in the office.

The survey also told me what not to build. Agents reported being reasonably confident about which customers qualified for partner products, so I scoped eligibility down to a brief review rather than a module. That decision did not survive stakeholder review, and I think the case is more useful because of it.

Four goals, sequenced to the transaction

I wrote four goals and put them in the order the work actually happens: know the partner carriers and what they offer; navigate each carrier system to produce an accurate quote; bind the policy and take payment; then generate and send the signature documents that complete the sale.

The sequence is the design decision, not the list. An agent who can quote but cannot bind has an unfinished transaction. An agent who can bind but cannot produce documents has an unpaid one. Each goal is a prerequisite for the next, so the training moves from general orientation to specific procedure in the same order a customer conversation does.

Three stakeholders and one useful disagreement

Getting the reviewers into one room was not practical, so I interviewed them separately: a customer-support specialist who is regularly pulled off his own work to rescue partner-product transactions, a regional manager accountable for whether agents hit their targets, and a regional sales leader who would deliver the training.

The first two confirmed what the survey said. The third contradicted it. He argued that agents only believe they understand eligibility, and that he routinely sees policies submitted that should never have been written — a problem invisible to the agents themselves, because the consequences surface later and elsewhere.

My data said one thing and my subject-matter expert said another. I added the eligibility review back as a substantive component with a knowledge check, and placed it at the front of the day. Self-reported confidence is evidence of confidence; it is not evidence of accuracy, and the person who sees the downstream errors had information my four respondents could not have.

Designing for a distributed workforce

The agents are spread across offices; the trainer covers several of them. That single constraint set the delivery model. The program runs as one day of virtual instructor-led training over a conferencing platform, with agents at their own desks, on their own equipment, inside the systems they will use with real customers an hour later.

Around that spine I blended three other modes: a vendor's own self-paced module where one existed, live demonstration by the trainer, and hands-on practice in the partner-carrier systems on a second screen. Practice happens in the real environment rather than a simulation, using the fictitious customer records the organization already maintains for training — a legal requirement, since agents may not quote a real person without permission.

Learning and performance outcomes

Each goal became a learning outcome, and each learning outcome resolved into observable performance. The distinction matters: “understands the quoting process” cannot be assessed, while “produces an accurate quote within the target time, with no data-entry errors” can be watched and scored.

Assessment strategy

Formative assessment runs continuously through scenario discussion and guided click-through, so the trainer sees where an agent is losing the thread while there is still time to intervene. Summative assessment is a live role-play scored against a checklist, because the outcomes are procedural and a checklist records whether each step happened in order.

One decision I would defend anywhere: the same fictitious customer carries across all four assessments. An agent quotes that customer, binds that customer's policy, and generates that customer's documents. Assessing each goal against a fresh scenario would have tested four isolated skills; assessing them against one continuous customer tests the transaction, which is what the job actually is.

Performance support as a design decision

The job aid was in the design from analysis onward, not added at the end. Two findings put it there. Partner-carrier requirements change often enough that memory is the wrong storage medium. And the organization already had job aids — outdated, and never mentioned to new agents during their week of regional practice, so nobody used them.

So the aid is organized by carrier and covers eligibility, products, underwriting considerations, documentation, payment, and which carrier fits which customer. It is designed to sit at the edge of the agent's screen during a live call. Alongside it, the design routes agents to the organization's existing repository of procedural updates rather than duplicating that content, and asks supervisors to coach during the first live transactions. Training ends; the work does not, and the support has to outlast the day.

Decision logic

The sequence was the same one I use elsewhere: clarify what is actually being asked, listen to the operating reality, define what good performance looks like in observable terms, diagnose the barriers, equip proportionately, and stay through transfer. Its most useful moment here was diagnosis. The presenting problem was framed as a training gap, and training was part of the answer — but outdated, unmentioned job aids were a resource problem, and the absence of coaching during early transactions was a management one. A program that only taught would have addressed a third of what was wrong.

Alpha review, beta test, and revision

Two regional sales leaders and a subject-matter expert reviewed the completed design. They asked for no changes. I record that because a review returning nothing is easy to leave out of a portfolio, and leaving it out would misrepresent how the project actually went.

The beta test was more productive. Five reviewers worked through the built program, representing instructional design, regional sales leadership, subject-matter expertise, new-agent experience, and experienced-agent practice. They found the structure sound and the scenario-based practice realistic, and they identified the job aid as the piece most likely to survive into daily work.

Their recommendations were consistent enough to act on directly. The first module leaned too heavily on text, so it gained visuals and interaction. Several videos needed clearer on-screen text. The assessment repeated itself in places. Most usefully, a reviewer pointed out that the facilitator prompts asked agents for the right answer rather than the reasoning behind it — so I rewrote them to ask why a carrier was the right fit, which is the judgment the job actually requires. I prioritized the revisions by instructional impact against effort, and the high-impact items were completed first.

Lessons learned

Ask the people who see the consequences. My survey respondents could not report a problem whose effects reached them weeks later and in another department; the subject-matter expert could.

Sequence teaches. Ordering the goals to match the transaction did more for coherence than any single activity in the program.

Support is part of the design. A job aid added at the end is a document. One scoped during analysis is a decision about where knowledge should live.

What I would improve

The survey is the weakest evidence in this project and I knew it while running it. Four responses by convenience gave me direction, not measurement. Given the chance again I would widen the sample, separate self-reported confidence from demonstrated accuracy, and establish a baseline before designing anything — so that the evaluation, when it runs, has something to compare against.

Why this transfers beyond insurance

The shape of this problem is common. An organization trains thoroughly on its core product and leaves an adjacent responsibility to be picked up informally — from a colleague, from a document nobody mentioned, from trial and error in front of a customer. Informal transfer looks free because its cost is distributed: a few interrupted minutes here, a slower transaction there, an error that surfaces somewhere else entirely.

What transfers is the method rather than the content: establish the gap with evidence instead of assumption, check that evidence against people positioned to see what learners cannot, separate the knowledge problem from the resource and management problems, assess against observable performance, and put support where the work happens rather than where the training happened.