AI Adoption Proposal Deck

About this template

Sixteen slides that ask for a phased approval

The deck for proposing AI tools inside a company that has not adopted them yet. It is deliberately not a technology pitch: five of the sixteen slides are about cost, risk, controls and rules, and only one is about what the tools do. The argument it makes is that you do not open access to everyone at once, and it asks for a pilot with a measurement plan attached rather than a company-wide license. The closing line, “A person makes the final check”, is the commitment the whole deck is built to make credible.

Eight agenda lines, five body pairs

Slide 2 runs where to use it first, cost and payback, phased rollout, risks and controls, operating rules, opening it all at once vs opening it in phases, key metrics, trend analysis. The first five carry a divider and a body slide from slide 3 to slide 12, line 06 is the comparison on slide 13, and the cards and bars are slides 14 and 15. If you have to cut, merge operating rules into risks and controls – never drop both, because the two of them are what an approver is buying.

The five body slides

1. Where to use it first: “3 chosen from 18 candidate tasks”, “Ranked by repetition, low cost of a wrong answer”, “Support drafts – document summaries – code review”. The ranking rule matters more than the three tasks; state it and the selection stops looking arbitrary.
2. Cost and payback: “Monthly fees estimated at KRW 7.2m”, “1,100 hours a month handed back”, “Payback lands 7 months after rollout”. Say what hourly rate turns hours into money, or the payback month is unverifiable.
3. Phased rollout: an 8-week pilot with 2 teams, then departments with training, then company-wide on a full contract.
4. Risks and controls: “Training on our own data switched off”, “A person signs off on generated output”, “No automatic sending to customers”. Written as settings and rules, not as intentions.
5. Operating rules: a published list of what must never be entered, usage and cost tallied by department, effects and rules revisited each quarter.

The metric cards and the trend bars

Slide 14 holds “Tasks in the first wave” 3, “Monthly cost (KRW 10k)” 720, “Hours saved per month” 1,100 and “Payback (months)” 7. The cost card is denominated in units of ten thousand won, so 720 is the KRW 7.2m quoted on slide 6 – change one and the other has to move. Slide 15 rises from “Pilot – 2 teams” 24 through “Dept. rollout” 86 and “Company-wide” 148 to “6 months in” 192, which reads as hours saved per month by stage rather than a calendar series. If you relabel it, the last bar should still be the 1,100 on the card or the two slides disagree. Clear the sample-values footnote and the “Unit: index” label before this goes to an approver.

The comparison slide

Slide 13 sets “Open to everyone at once” against “Phased rollout”. “No way to tell who uses it for what” faces “Effect measured in an 8-week, 2-team pilot”, “Unreviewed answers reach customers” faces “Nothing goes out without a human check”, and “Costs climb without warning” faces “Usage and cost tallied per department”. Each left-hand line is a failure someone in the room has already read about, which is why this slide usually does the persuading.

What goes wrong

• Quoting a license price with no usage assumption behind it. Per-seat and per-use pricing behave differently the moment phase 2 doubles the users.
• Promising 1,100 hours without saying what happens to them. Hours handed back are only savings if the plan says what fills them.
• Listing controls with no owner. “Training on our own data switched off” is an administrator setting on a specific plan – name who verifies it and when.