Japan Post Insurance Rolls Out AI Role-Play to Sharpen Sales Skills

How do you spread the tacit know-how of your best salespeople across an entire team? A concrete answer to this old-yet-new problem has emerged using generative AI. In May 2025, Japan Post Insurance (Kampo) rolled out AI role-play across all branches of its corporate sales division. The AI plays the customer, so staff can run realistic sales practice on their own, anytime.

A frontline idea that scaled company-wide

What stands out is that the service grew from a young sales employee idea and spread quickly across the corporate sales divisions of all branches. It started not as a top-down DX initiative but from a real frontline pain point. Traditional role-play required booking a senior or manager as the counterpart, which sharply limited practice during busy periods. With an AI partner, staff can repeat the drill as often as they like — on the move or in spare moments. The biggest value is that the sheer volume of practice goes up.

In sports and language learning alike, improvement rests on repetition. Sales is no different: an environment that deliberately increases the number of reps is what shapes how fast junior staff ramp up. AI freed those reps from the constraints of cost and time.

A design of quantitative AI scoring plus qualitative human feedback

The cleverness of the system is its two-layer evaluation. The AI scores delivery and proposal content quantitatively, making weaknesses visible in numbers; then a training mentor adds qualitative feedback. The AI provides an objective yardstick, the human provides advice attuned to context and nuance — a clear division of labor.

Here is one correct shape for using generative AI: do not hand everything to AI, but lift the baseline with AI and finish with people. Especially for less experienced staff, seeing problems in numbers has outsized impact.

Generative AI use beyond sales

Kampo's generative-AI adoption goes beyond training. In its contact center, post-call processing time was reportedly cut from over five minutes to 90 seconds; even a few minutes saved per call, multiplied across volume, becomes an enormous reduction in workload. For fiscal 2026 the company plans an AI that reads documents in claims assessment, expected to halve the related clerical work. Automating routine tasks for labor saving, and upgrading talent development for a capability lift — using generative AI clearly for these two distinct purposes is what characterizes the effort.

Adapting it to your own organization

If uneven sales skill is your problem, turning role-play into AI drills is the easiest place to start. Freed from a partner's availability, staff can raise their practice volume, and know-how that once lived in a few individuals gets levelled into an organizational standard through repetition. The crucial design point, though, is not to hand everything to AI: let it score delivery and proposal content quantitatively, but keep people anchoring the qualitative mastery of the final form — that line is what protects quality.

In the back office, the standard move is to automate the time-eating routine first, such as post-call processing and document checks, where the savings show up clearly in hours and earn internal buy-in. This is a large-enterprise case, yet AI role-play and automated post-call processing are equally workable for small firms and individual teams. The point is not "replacing people" but "increasing their practice volume and judgment." Treat AI as an amplifier, and that is where the biggest results come.

Key takeaways

Kampo's case shows the value of treating generative AI as an amplifier of people rather than a replacement for them. AI role-play removes the constraint of needing a human counterpart, raising the sheer volume of practice, while quantitative scoring turns delivery and proposal weaknesses into numbers. The qualitative, context-aware coaching that finishes the job stays with people — and that division of labor is what protects quality.

The order of application is just as clear. In development, repetition levels individual know-how into an organizational standard. In the back office, start with the time-eating routine — post-call processing (from over five minutes to 90 seconds) and the document-reading claims AI planned for fiscal 2026 (expected to halve related clerical work) — where savings show up plainly in hours. This is a large-enterprise example, yet every piece of it is reproducible for a small firm or a single team. The question to ask before investing is simply: which task's practice volume and judgment can AI multiply first?