Support automation reports die in finance review when they lead with a vendor dashboard export. A percentage without denominator, a savings line without baseline, a satisfaction number from a different quarter — each erodes trust faster than a flat ROI would have. CFOs are not sceptical of automation. They are sceptical of stories that treat correlation as proof.
The report they accept is boring on purpose. One page, four sections, numbers you can rederive, and explicit gaps where you will not pretend certainty. It is the monthly version of the discipline in resolution vs deflection and satisfaction must travel.
Send the same shape every month, even when the story is flat. Finance learns to read it; you avoid redefining metrics when the news is bad.
One-page layout
- Volume vs baseline — half page max.
- Quality sample — table with definitions in footnote.
- Backlog — three bullets with owners.
- Money — only lines you will defend in questions.
Section one: volume in context
Show ticket volume vs the pre-launch baseline — same channels, same categorisation rules. Note known confounders: campaign, outage, pricing change. Show widget conversations started. If tickets and widget usage move in opposite directions, say so before finance asks.
- Tickets this month vs baseline month (not vs launch week).
- Widget sessions and unique visitors if you have them.
- Escalation count and median turns to human.
- One sentence on seasonality you already expect.
Section two: quality evidence, not vibes
Finance understands sampling if you show your work. Twenty conversations stratified across top themes. Count correct, wrong, honest refusal, bad handoff. Express as ranges if the sample is small — honesty beats false precision.
Use the same sample IDs month to month when you can: trend beats snapshot for quality, and finance learns what “wrong on covered content” means in plain language.
| Audit bucket | Count in sample | Action if rising |
|---|---|---|
| Correct with verifiable citation | — | Maintain |
| Wrong on covered content | — | Retrieval or chunking fix |
| Invented on gap content | — | Grounding regression — urgent |
| Honest refusal | — | Content backlog |
| Handoff friction | — | Routing fix |
Section three: backlog and next month
List top three refusal themes or content gaps with owner and ship date. This section proves automation is operated, not installed. Include one escalation quality fix if measuring escalation quality surfaced a pattern.
Executives tolerate flat ROI longer when the operating loop is visible. “We will write the returns exception page by the 15th” is a plan. “Model v3 next quarter” is not.
Section four: money, carefully
Only include dollar estimates when you can show the arithmetic on the same page. Common honest approaches:
- Cost of tool + implementation amortised — factual, not controversial.
- Hours saved on sampled escalations that clearly ended in the bot correctly — conservative multiplier, show sample.
- Avoid translating full ticket drop into dollars unless attrition checks from falling tickets passed.
“A CFO accepts ‘we saved twelve agent hours on shipping questions we audited’ faster than ‘we deflected 40%.’”
What to leave off the page
- Vendor answer rate without your audit.
- Containment alone as success.
- ROI payback months derived from launch-week spikes.
- Customer quotes without linking to measurable themes.
Send the same report every month so finance learns the shape. When a number moves, they will ask better questions because the definitions stayed put. That rhythm is how automation earns budget — not one heroic quarter that cannot repeat.
Attach one sampled wrong answer (redacted) when quality slips — not to dramatise, to anchor the quality section in something concrete. Abstract percentages feel negotiable; a misquoted refund window is not.
When automation is flat month-over-month, say so. Credibility compounds when bad months look as honest as good ones.
