Calibration in writing is the same as calibration in detection: matching your stated confidence to your actual confidence. A well-calibrated memo lets the reader act on each claim at the appropriate level of risk; a poorly-calibrated one forces them to either trust everything or trust nothing.
Two patterns to avoid
- Flat overconfidence. Every claim stated as fact. "This will increase conversion by 12%." A reader who knows the data is thinner than that ends up discounting everything in the memo.
- Flat hedging. Every claim wrapped in the same "could potentially", "may possibly", "appears to indicate". A reader can no longer tell which claims you'd defend and which you wouldn't.
What calibrated memo language looks like
A calibrated memo signals different levels of confidence with different verbs. "The data shows" for things you can demonstrate. "We estimate" for modelled projections. "We think but can't yet prove" for the gut call you still want on the page. The reader can act on each at the right level.
Before
The new pricing will significantly improve revenue and may potentially increase customer satisfaction. This will fundamentally transform our market position and could possibly lead to increased market share.
After
Q3 data shows a 6% lift in revenue per account from the new pricing on the cohort that's seen it. We estimate this generalises to the full base, with the caveat that the cohort was self-selected. Satisfaction is harder to read — NPS hasn't moved, but unprompted feedback is positive. We'd hold off on claiming a market-share effect until at least Q1.