9.3 Communicating to Non-Technical Audiences

What Task C-3 Asks For

The final task in the Content Outline is to "communicate project findings to non-technical audiences, including the implications on business outcomes or decisions." The rubric sharpens it further: the candidate must make an effective persuasive argument to a non-technical audience about how the model addresses the business question, including the rationale for the creation of the model.

The word persuasive is doing real work. A correct explanation that leaves the audience unable to decide anything has not met the criterion.

Start With Audience and Action

Knaflic's method begins before any chart is drawn, with two questions:

1. Who is the audience? An underwriting chief, a product manager, a regulator and a CFO want different things from the same model. The underwriter wants to know which risks to look at differently. The product manager wants to know the impact on competitiveness and mix. The CFO wants the loss-ratio effect. The regulator wants to know why the rating variables are reasonable.
2. What do you want them to do? Approve a filing, redirect underwriting attention, fund an implementation, change an appetite. If you cannot state the requested action in one sentence, the communication has no target.

From those two answers comes the single message — the one sentence you would keep if you could keep only one.

Lead With the Conclusion

Technical writing builds to a conclusion. Business communication states it first.

The methodology still belongs in the report — task C-2 requires it — but it comes after the finding for this audience. A non-technical reader who has to wade through the build before learning what happened will form their own conclusion before reaching yours.

The practical structure for the non-technical portion:

1. The business question. One sentence restating what was asked and why it mattered — this is the "rationale for the creation of the model" the rubric requires.
2. The finding. What the model says, in business terms.
3. The consequence. What it is worth, or what it costs to ignore.
4. The recommendation. The action being requested.
5. The caveat. The main limitation, stated plainly.

Translate Into the Audience's Units

This is the most reliable way to lose or keep a business reader.

Useful business units: loss ratio points, premium dollars, percentage of policies affected, retention percentage, number of risks referred. Not: coefficients, deviance, AIC, p-values, dispersion.

One Exhibit, One Message

A technical reader will study a dense exhibit. A business reader will look at it for a few seconds and take away one thing — so decide what that one thing is.

• Emphasise deliberately. One colour, one bold label, one annotated point. Everything else in grey.
• Annotate the chart itself. A short label on the decile where the gap is widest does more than a paragraph beneath it.
• Strip what does not serve the message. Gridlines, extra series, legends that could be replaced by direct labels.
• Title with the finding. "Current rates are 18% short in the highest-risk decile" rather than "Figure 3: Decile analysis."

This is the infovis end of the Gelman and Unwin spectrum, and it is the correct register here — provided the underlying statistical work was done with the other kind of graphic first.

Anticipate the Business Questions

A persuasive argument survives the first three questions the audience will ask:

• "How much is this worth?" Have the number, and the assumptions behind it.
• "Who does it hurt?" Know which segments see increases and roughly how many policies are affected.
• "Why should I believe it?" One sentence of validation evidence in plain terms: "this was measured on policies the model never saw during fitting."
• "What if it is wrong?" The main limitation and what would be monitored after implementation.

Confidence Without Overclaiming

Two opposite failures are equally damaging. Hedging everything — "the model may possibly suggest a potential tendency" — gives the audience nothing to act on. Overclaiming — presenting an estimate as a certainty, or an association as a cause — destroys credibility the first time reality diverges.

The workable register states the finding directly, attaches the uncertainty to it, and keeps the recommendation intact:

"The model indicates the urban young-driver segment is under-priced by roughly 18%, within a range of about 12% to 24%. Even at the low end the correction is worth implementing this year."

[!WARNING] Do not present a modelled relativity as a causal claim. "Young drivers in urban territories cause 67% more loss" overstates what a GLM shows. "Risks with these characteristics have historically generated about 67% more loss per car-year, holding other rating variables constant" is both accurate and still persuasive.