AI Strategy for Business Leaders: 7 Questions

June 5, 2026

AI strategy for business leaders: the seven questions a leadership team must answer before approving AI spend, stacked as a board-briefing card

Last updated: June 4, 2026

If you are the CEO, COO, CFO, or board member with twenty minutes on the executive calendar and a six- or seven-figure AI ask in front of you, you have a different problem than the strategist who wrote the deck. The forty-two slides answer "why AI." Your signature releases the spend, so you have to answer something narrower and harder: why this workflow, this owner, this data path, this quarter, and how this one ends if it does not work. Approve on a deck that handwaves any of those, and you are funding the pilot that quietly dies in Q3 with no kill criteria written. In this guide, you'll get the seven questions to run in a thirty-minute meeting, what a good answer to each sounds like compared with a vague one, and the cost-and-timeline numbers a leadership team should expect a real proposal to land inside.

The IBM IBV 2025 CEO Study of 2,000 CEOs across 33 countries found 64 percent of CEOs admit they are investing in AI before they fully understand the value, and the top barrier they name is "lack of expertise." The seven questions below are designed to close that gap inside a single executive meeting. A free AI Assessment walks a leadership team through all seven and returns a written one-page brief before the SOW lands on a desk.

If you want the operator's practical guide, see practical AI strategy for business. This piece is the twenty-minute briefing for the executive whose signature releases the budget. The methodology that sits behind it lives in the pillar on enterprise AI strategy, and the longer narrative on corporate AI strategy covers moving beyond pilot purgatory.

Quick Answer
• What it is: Seven questions a leadership team must answer before approving AI spend: workflow, owner, data path, human-in-the-loop design, eighteen-month operations cost, kill criteria, success metric.
• How to use it: Run the questions in a single thirty-minute meeting. Approve only when all seven have specific answers.
• Why it matters: Vague answers correlate almost perfectly with pilots that never reach production.

Why does a leadership team need seven questions instead of a strategy deck?

AI strategy for business leaders comes down to seven questions because the deck is almost always written for the executive who already approved the spend. The deck answers "why AI." The leadership team has to answer "why this AI, this workflow, this owner, this budget, this quarter." Those are the questions a board or a CEO is on the hook for, and they are the questions a sales pitch is least equipped to surface honestly.

The pressure to move is real. The PwC AI Agent Survey of 300 senior US executives found 79 percent of US businesses already adopting AI agents and 88 percent raising AI budgets, with 66 percent of adopters reporting measurable productivity gains. The Stanford HAI 2025 AI Index reports 78 percent of organizations used AI in 2024, up from 55 percent the year before. The work is being commissioned. The harder question is whether the work being commissioned has answers to all seven.

The seven questions, and what a good answer sounds like

Picture the boardroom moment. The slide says "Approve $480,000 AI investment." You have twenty minutes before the next item. Run the seven questions. Listen for specificity. A good answer names a workflow, a person, a system, a number, and a date. A bad answer names a category, a team, a vendor, a range, and a quarter.

THE SEVEN QUESTIONS

What a leadership team must answer before approving AI spend

Listen for a workflow, a person, a system, a number, and a date.

QUESTION 01

Which workflow ships first?

A good answer names a single workflow with a measurable bottleneck and a clear before-and-after. Not "finance." Not "customer support." One workflow that one named person currently runs.

QUESTION 02

Who owns it?

A good answer is a job title and a calendar, not "the innovation team." The owner has a P&L stake in the workflow and authority to change the operating rhythm around it.

QUESTION 03

Where does the data live?

A good answer chooses between a private deployment where data never leaves the building and a hosted SaaS path, with the trade-offs explicit, before the contract is drafted.

QUESTION 04

What is the human-in-the-loop design?

A good answer names where a human approves the agent's action: every action, exceptions only, or a sampled audit. Said out loud, before go-live.

QUESTION 05

What is the eighteen-month operations cost?

A good answer is a run-rate, not a build budget. It covers monthly inference, monitoring, retraining cycles, and the internal time the SOW does not price.

QUESTION 06

What are the kill criteria?

A good answer is a specific outcome by a specific date that, if missed, ends the program. Not "we will reassess in Q4." A line in the sand, agreed before the spend.

QUESTION 07

What metric proves it works?

A good answer is a single measurable agreed at kickoff. Cycle time, error rate, cost per transaction, conversion. One number. Tracked weekly. Visible to the executive sponsor.

Approve spend only when all seven have answers. Otherwise you are funding pilot purgatory.

The reason this matters is the pilot-to-production gap. The Deloitte State of Generative AI Wave 4 study of 2,773 C-suite respondents found more than two-thirds expect 30 percent or fewer of their generative AI experiments to scale within three to six months. BCG's Where's the Value in AI? report from October 2024 found 74 percent of companies struggling to capture value from AI. Almost every one of those stalled pilots was approved on a deck that left at least three of the seven questions vague.

Get the seven answers before you approve the spend
The free 60-minute AI Assessment walks a leadership team through all seven questions in a single working session and returns a one-page summary with the answers in writing.
Book Your Free AI Assessment →

What do good and bad answers actually look like?

The questions are useless without an ear for the answers. Picture the budget-approval meeting. The same seven questions get asked. Two proposals come back. One reads as an operating plan. One reads as a sales pitch with a budget attached. Side by side, the difference is uncomfortable to miss.

GOOD ANSWER VS BAD ANSWER

What each of the seven sounds like in the room

A bad answer is not always wrong. A vague answer almost always is.

QUESTION 01 | WORKFLOW

Good answer: First-notice-of-loss triage, 1,200 documents a week, 14-minute average handle time today.

Bad answer: "We are starting with claims operations."

QUESTION 02 | OWNER

Good answer: Director of Claims Operations, 20 percent of calendar reserved through Q4, P&L stake in cycle time.

Bad answer: "The innovation team will own it."

QUESTION 03 | DATA PATH

Good answer: Private deployment inside our VPC, data never leaves the building, models hosted in our identity boundary.

Bad answer: "We will work with the vendor's enterprise tier."

QUESTION 04 | HITL

Good answer: Human approves every payment release; agent auto-acts on classification only; sampled audit on the rest weekly.

Bad answer: "The agent will be fully autonomous from day one."

QUESTION 05 | 18-MONTH COST

Good answer: Build $30K, monthly inference and monitoring $4K, 0.3 internal FTE at $48K loaded, 18-month all-in $162K.

Bad answer: "The build is in the budget."

QUESTION 06 | KILL CRITERIA

Good answer: If average handle time is not under 6 minutes by March 31, the program ends and the spend stops.

Bad answer: "We will reassess at the end of the fiscal year."

QUESTION 07 | METRIC

Good answer: Average handle time on first-notice-of-loss triage. Tracked weekly. Reported to the executive sponsor.

Bad answer: "We will measure productivity and efficiency."

Bad answers are not always wrong. Vague answers always are.

How does a board evaluate an AI strategy in twenty minutes?

The seven questions are the agenda. Five minutes on workflow and owner. Five on data path and human-in-the-loop. Five on the eighteen-month cost. Five on kill criteria and the success metric. If any of the seven cannot be answered concretely, the strategy is not ready and the spend should not be approved yet. The IBM IBV study cited above also found 54 percent of CEOs already hiring for AI roles that did not exist a year ago. That hiring pressure is exactly why Question 2, the owner question, is the one most often handwaved: nobody on staff today has run an AI program before, so the proposal hides the gap inside "the innovation team will own it."

What does a real answer to the cost question sound like?

The honest blunt truth: most AI proposals brought to a leadership team are scoped to win approval, not to ship. The seven questions exist to expose that gap in a single meeting. Arkeo's build experience says a scoped single-workflow agent runs roughly $15,000 to $40,000 to build with a 6 to 10 week timeline to production, or 8 to 12 weeks for a private or enterprise deployment where data never leaves the building. Off-the-shelf copilots come in at roughly $20 to $30 per user per month and go live in days. The first quick win typically lands in 30 to 90 days. Arkeo operates under an Assess, Deploy, Manage rhythm and runs its own operations on the same private agents it deploys for clients (we use what we sell), which is the only reason answers can be this specific rather than aspirational. The seven questions force a vendor or an internal team to be that specific before the budget gets approved, instead of after.

Walk the seven questions with your leadership team
The free 60-minute AI Assessment runs the seven questions live with your executives and returns a written one-page brief covering all seven before the next board meeting.
Book Your Free AI Assessment →

Frequently Asked Questions

What should a CEO ask before approving AI spend?

A CEO should ask seven questions before approving AI spend: which single workflow ships first, who owns it by name and job title, where the data lives (private or hosted), what the human-in-the-loop design is, what the eighteen-month operations run-rate is, what the kill criteria are with a specific date, and what single measurable proves the program works. Approval should only follow when all seven have specific answers. Vague answers correlate almost perfectly with pilots that do not reach production.

What is the difference between AI strategy for business leaders and AI strategy for business?

AI strategy for business leaders is the twenty-minute executive briefing version, structured as the seven questions the leadership team must answer before approving spend. AI strategy for a business is the operator-grade methodology that sits behind the spend, covering current-state diagnostics, workflow selection, sequencing, change management, and the eighteen-month operating plan. A leadership team uses the briefing version to decide; an operator uses the methodology to deliver.

How does a board evaluate an AI strategy?

A board evaluates an AI strategy by walking the seven questions in roughly twenty minutes and listening for specificity. A workflow named at the document level, not the function level. An owner named by job title with calendar reserved. A data path chosen between private and hosted with the trade-offs explicit. A human-in-the-loop design said out loud before go-live. An eighteen-month operations cost that includes monthly inference and internal time, not just the build budget. Kill criteria tied to a specific outcome and date. One measurable success metric agreed at kickoff. Any handwaved answer is the answer to disqualify the proposal on.

Who is accountable for AI strategy in a mid-market business?

In a mid-market business, accountability for AI strategy sits with the CEO or COO at the portfolio level and with a named line-of-business leader at the workflow level. The executive sponsor approves the spend and owns the eighteen-month outcome. The line-of-business leader (a Director of Claims Operations, a VP of Finance, a Chief of Staff) owns the specific workflow being automated, holds the P&L stake in the metric, and reserves calendar for the build and run phases. "The innovation team will own it" is a red flag, not an answer.

What red flags signal a bad AI proposal?

The cleanest red flags are vague answers to the seven questions. The workflow is named at the function level ("finance," "customer support") rather than the document or transaction level. The owner is a team rather than a person. The data path is left to the vendor's enterprise tier without a private-versus-hosted decision. The human-in-the-loop design is unspecified. The eighteen-month cost shows the build but hides monthly run-rate and internal time. The kill criteria are "we will reassess." The success metric is "productivity and efficiency." Any one of these is a yellow flag. Two or more is a proposal that should be sent back before the spend is approved.