The short version is that reporting AI customer service to executives takes four numbers and one judgment, delivered on one slide, repeated every month. Resolution rate (not deflection), CSAT parity between AI-handled and human-handled tickets, cost per resolved contact against the pre-AI baseline, and audit readiness against the governance checklist. The judgment is what you would change next quarter and why. A board does not want a dashboard. It wants to know whether the programme is working, cheaper, safe, and liked.
91% of customer service and support leaders feel pressure from executives to adopt AI in 2026, per Gartner's February 2026 release, and the pressure is as much to report well as it is to deploy. The common failure is a thirty-metric dashboard the executive team cannot act on. The alternative below is reporting AI customer service to executives in a format they will actually read.
Why deflection does not belong on the slide
Deflection is the number vendors like most and executives trust least once they understand it. A high deflection rate just means fewer tickets reached a human; it does not mean the customer got an answer. One analysis puts it bluntly: a chatbot with a 90% deflection rate can have a 40% resolution rate. The gap is real. If the board sees "92% deflection" and the CFO later discovers the resolution rate is less than half of that, the slide loses credibility and the programme with it. Our deflection rate note walks through the mechanics.
Report resolution. Define it the same way every month: a resolved ticket is one the customer did not reopen within seven days and did not score below 3 on CSAT. That definition is strict enough to survive scrutiny and simple enough to compute from the inbox.
The four numbers for the one-page slide
Keep the whole report to four numbers plus a trend arrow, with the governance line as the fifth element and judgment as prose.
1. Resolution rate on AI-handled tickets
The percentage of AI-replied tickets that were not reopened in seven days and did not score below 3 CSAT. Report the current month, the trailing three-month average, and the baseline. The useful question for the board is whether it is going up, not whether it is high. Top-decile AI deployments reach 80% or higher autonomous resolution, which gives one end of the range; most teams start much lower.
2. CSAT parity: AI versus human
The CSAT score on tickets AI resolved end-to-end, next to the CSAT score on tickets humans resolved end-to-end. Parity is the signal; convergence over time is the trend. Divergence of more than five points is the thing a board should ask about, because it says the AI is handling tickets it should not be, or it is handling them in a way the customer disliked. TechTarget cites 75% of leaders using CSAT as the primary AI-vs-human comparison, and the reason is that CSAT catches tone and resolution together.
3. Cost per resolved contact
Fully loaded support cost (salaries, tooling, overhead) divided by resolved contacts. Report AI-resolved and human-resolved separately, then the blended number. The point is to show the trend against the pre-AI baseline, not to compete with vendor marketing numbers. One industry benchmark cites AI-resolved inquiries at $0.62 against $7.40 for human agents, but your own baseline is the only one that matters. Our cost per ticket and measuring AI support ROI notes have the formulas.
4. Audit readiness
A single pass/fail against your governance checklist (data retention, DPA coverage, logging, human-in-the-loop for sensitive actions, DPIA current). Grant Thornton's April 2026 CFO survey found that 78% of executives could not pass an independent AI governance audit. The board number is binary this month: yes or no. If no, the one-sentence judgment names what to fix. Our EU AI Act August 2026 note covers the SMB framing.
The one-page template
| Row | What it shows | Trend | Target | | --- | --- | --- | --- | | Resolution rate (AI) | % tickets resolved end-to-end, no reopen, CSAT >= 3 | Last 3 months | Up and to the right | | CSAT parity | AI CSAT minus human CSAT, in points | Last 3 months | Within 5 points | | Cost per resolved contact | Blended, with AI and human sub-rows | Last 3 months vs pre-AI baseline | Below baseline | | Audit readiness | Pass/fail against the governance checklist | This month | Pass | | Judgment | Two to three sentences on what we are changing | This month | Specific |
The judgment line is what distinguishes a report from a dashboard. "Resolution is up two points because we expanded the AI's scope to shipping questions last month. CSAT parity narrowed to three points. We missed audit readiness on retention logs; the fix is scoped and in next week's release."
What stays off the slide
Three categories of metrics tempt their way onto executive reports and should not land there.
Volume numbers alone. "Replies this month" and "tickets handled" are operational, not strategic. The resolution rate contains them.
Model and vendor names. The board does not need the stack. Executives benefit from knowing the programme works, not what it is built from. Fin's 2026 report notes that as AI proves its value, focus shifts from cost savings to strategic impact; keep the slide at that level.
Vanity benchmarks. Industry average X, top-decile Y: these calibrate the ambition, not the report. If you do include a benchmark, include it once, with the source URL, and never as a bar chart next to your own number.
The monthly cadence that actually ships
Three roles, three windows, forty-five minutes total.
- The ops lead pulls the four numbers from the inbox and the finance ledger. Twenty minutes.
- The programme owner writes the judgment, two to three sentences, naming what will change next month. Fifteen minutes.
- The sponsor (COO or CS head) reviews, signs off, and sends it. Ten minutes.
A one-page report that lands reliably every month is worth more than a quarterly deep-dive that slips once a year. The compounding effect across quarters is where executive trust in the programme is built.
Mature versus immature AI programmes
Fin's report finds that 87% of mature AI teams see improved metrics, versus 62% overall. The usual difference between "mature" and "the rest" is not model quality or budget. It is this report, run monthly, with the judgment line taken seriously. The report forces three habits: a single resolution definition everyone agrees on, a parity view against humans the team cannot hide behind, and a public audit-readiness number that keeps governance in the operating cadence.
Gartner also finds that 87% of customers say a route to a human is essential when companies use generative AI; the parity view is where that trust is kept honest. If AI CSAT dips below human CSAT by more than five points, the handoff rules need widening, not the AI scope.
How Keloa approaches board-ready reporting
Keloa's unified inbox rolls the four numbers up natively (resolution, CSAT per channel, cost per contact against the baseline, audit log status), and the monthly snapshot is one export rather than an SQL afternoon. Per-reply pricing makes the cost line directly comparable across months without a model-change asterisk. The governance register that feeds the audit-readiness pass/fail is a single page in the admin; see our data residency note for the EU framing. Report what matters, repeat it monthly, be ready for the audit on the first ask.
Frequently asked questions
Can we start with just two of the four numbers? Yes, resolution and audit readiness are the two to start with. CSAT parity needs a few months of volume to be meaningful, and the cost line needs the pre-AI baseline captured first. The two-number report is still a useful executive brief.
How do we define the "pre-AI baseline" for cost per resolved contact? Take the three months before AI handled any tickets. Fully loaded support cost (salary, tooling, overhead) divided by resolved tickets, with the same resolution definition. If the data is not clean, say so explicitly; a transparent baseline survives scrutiny better than a precise fabricated one.
Should resolution include tickets handed off to a human? Only if the human closed them without the customer reopening within seven days. A handoff is not a resolution for the AI; it is a resolution for the system. Report both: AI-only resolution and end-to-end resolution. The gap between them is a routing metric.
What governance checklist do we score against? The one your DPO signed off. It typically covers data retention, DPA coverage, logging, human-in-the-loop for sensitive actions, and a current DPIA. If you do not have one yet, the first "fail" is the right number to report; the following month is the fix.
How often should executives see this? Monthly for the operating review, quarterly as a slide in the business review. Weekly is operations and belongs in the team channel. Yearly is governance and belongs in the audit file. One monthly page, repeated, is the right cadence.
What if the numbers get worse one month? Report them anyway, with the judgment line explaining why. A reporting cadence that only ships good months is not a reporting cadence; it is marketing. The board rewards an honest down-month more than a quiet one.
Want a board-ready one-pager pulled straight from your inbox? Book a demo and we will configure the four numbers against your baseline.
Sources
- Gartner: 91% of customer service leaders under pressure to implement AI in 2026
- Gartner: customers expect a route to a human when companies use generative AI
- TechTarget: 7 customer experience trends for 2026
- Maven AGI: State of AI in CX 2026
- Fin: 2026 Customer Service Transformation Report
- Grant Thornton: 2026 responsible AI governance survey
- Lorikeet CX: resolve, do not deflect
- Fini: measuring AI customer support ROI