7 Questions That Reveal If Your Company Is Ready for AI

Ask a leadership team about AI and the first question is almost always "which AI should we buy?" It is the wrong first question.

The right one costs nothing and decides everything: which customer problem are we trying to solve — and are we ready to solve it?

The cost of skipping it is now measured, not argued: S&P Global found that 42% of companies abandoned most of their AI initiatives in 2025 — up from 17% a year earlier — with the average organization scrapping 46% of its proofs of concept before they ever reached production. Those were funded projects with executive sponsors. They died anyway, and mostly for reasons that were visible before a single model was chosen.

McKinsey's latest global State of AI survey frames the same gap from the other side: 88% of organizations now use AI in at least one function, yet nearly two-thirds have not begun scaling it, and just 39% can point to any enterprise-level profit impact.

Adoption is everywhere; readiness is not.

After two decades running customer experience and service transformations inside global technology companies, I have learned that readiness is not a feeling — it is checkable.

So here is the diagnostic I use: seven questions, each with the evidence to demand, the pass signal, and what to do if you fail. Run it with your leadership team in one honest meeting. No slides. Artifacts only.

A word on format, because format decides honesty. Book ninety minutes. Assign each question to the executive who would own it in production — not to the AI sponsor, who has every incentive to see green. The owner brings the artifact, the room challenges it and everyone scores pass or fail out loud. No partial credit; partial credit is how companies talk themselves into the 42%. If a question triggers a long debate, that is not a delay — that is the diagnostic working.

FAQ: AI Readiness and the 70-20-10 Budget Rule

Editor's note: These questions address the readiness diagnostic, budget allocation and adoption statistics referenced throughout this piece.

1. Do We Know Which Customer Problem We're Actually Solving?

Why it decides everything: strategy is the widest readiness gap there is. Cisco's AI Readiness Index, built from more than 8,000 senior leaders across 30 markets, finds that 99% of the fully-ready "Pacesetter" organizations have a well-defined AI strategy — against 58% of everyone else.

• Evidence to demand: a one-page problem statement naming the customer pain, the metric it moves and today's baseline number.
• Pass signal: anyone in the room can say what changes for the customer, by how much, by when.
• If you fail: stop the vendor conversations and run a proper gap discovery first — I published my framework for deciding when AI is necessary and when it isn't, because some gaps are process or design problems wearing an AI costume.

Why Do Most AI Projects Fail Before Model Selection?

Cisco's AI Readiness Index shows only 58% of non-Pacesetter organizations have a defined AI strategy, versus 99% of fully-ready companies — meaning strategy, not technology, is the widest readiness gap.

2. Who Owns the Outcome — and Does the Budget Match It?

Why it decides everything: BCG's research puts the ratio bluntly: successful AI programs devote roughly 10% of resources to algorithms, 20% to technology and data, and 70% to people and processes — and many organizations get this exactly backwards.

• Evidence to demand: one named executive accountable for the customer outcome (not the deployment), and the budget split across those three buckets.
• Pass signal: a single owner who keeps the accountability after go-live, and a spending plan where people and process work is the majority line, not the leftover.
• If you fail: rebalance before you spend another euro/dollar/Yen — an inverted 70-20-10 budget is the most reliable failure predictor I know.
  

Related Article: Your All-Green Customer Experience AI Dashboard Should Worry You

What Is the 70-20-10 Rule for AI Budgets?

BCG's research shows successful AI programs allocate roughly 10% to algorithms, 20% to technology and data and 70% to people and process — a ratio most organizations invert.

3. Is Our Customer Data Actually Ready Today?

This is the question most companies fail, and the numbers say so: Cisco's index finds only 19% of organizations have fully centralized data infrastructure, and S&P Global's abandonment study names data privacy and security among the top obstacles that kill projects.

• Evidence to demand: a data map for this specific use case — where each required element lives, how fresh it is, who governs access, and whether customer consent covers this use.
• Pass signal: your team can trace every data element the AI needs to a governed, current source without the word "probably." A useful stress test: ask what the AI would tell your most important customer about their own account, today. If the room winces, you have your answer.
• If you fail: resist the enterprise-wide data program that takes three years. Fix the data for one use case, prove value, then widen — readiness is built in slices, not oceans. And treat consent as part of the data, not as legal paperwork: an AI that uses customer information the customer never agreed to is not a data problem waiting to happen; it is a trust problem that already exists.

How Many Organizations Have Centralized AI-Ready Data?

Only 19% of organizations have fully centralized data infrastructure, per Cisco's index, and S&P Global names data privacy and security among the top reasons AI projects get abandoned. 

4. Will Our Processes Survive Being Automated?

Why it decides everything: automation is loyal to the process, not to the customer. Wire AI into a workflow that already produces friction, and it will produce that friction at scale — faster, tirelessly and with perfect confidence. McKinsey's survey shows the winners understand this: among the companies capturing the most value from AI, fully redesigning workflows is one of the practices most strongly linked to profit impact and company and customer outcome. That is exactly what we did in Samsung, instead of the other mistaken way around.

• Evidence to demand: the target process walked end to end by its owner, with the exception cases counted — the credits, the escalations, the "it depends" moments.
• Pass signal: the process owner can explain what happens in the 20% of cases that break the happy path, and the design covers them.
• If you fail: redesign the process first. It is cheaper to fix a workflow on a whiteboard than inside a deployed system your customers are already meeting.

Why Do Redesigned Workflows Outperform Automated Broken Ones?

McKinsey's survey found that fully redesigning workflows — not just automating existing ones — is among the practices most strongly linked to AI profit impact.

5. Are Our People Skilled Enough to Use This Safely?

Why it decides everything: only 31% of organizations rate their talent at a high state of AI readiness, per the same Cisco research. Tools without skills produce theater; skills without sanctioned tools produce shadow usage your security team will discover later.

• Evidence to demand: a skills baseline for every role the AI touches, and proof that an approved, secure tool exists for daily practice.
• Pass signal: each affected role has a training path with time allocated, and nobody needs a personal account on an unapproved app to do their job.
• If you fail: pause the deployment, not the learning. Fund role-specific skill paths with protected practice time on approved tools, and track fluency like any other operational KPI — because untrained users don't reject AI loudly; they route around it quietly, and your dashboards never find out.

What Share of Organizations Rate Their AI Talent as Ready?

Only 31% of organizations rate their talent at a high state of AI readiness, per Cisco, creating a gap between deployed tools and the skills needed to use them safely.

6. Can We Govern This Once It Faces Customers?

Why it decides everything: an ungoverned system in front of customers is a brand incident choosing its date. Governance is not the brake on customer experience; it is the license to move fast without gambling trust. So also design of the experience will be crucially important to make it right.

• Evidence to demand: a live inventory of every AI deployment, a named person who can stop any of them within the hour, and a tested escalation path to a human for the moments that matter.
• Pass signal: the room can answer "who sees everything, who stops anything, and where does a stuck customer land?" without checking. The strongest teams rehearse it: they run the shutdown drill before launch, the way factories run fire drills or get a POC case by case — because discovering your kill-switch doesn't work is only acceptable when nothing is burning. 😊
• If you fail: no customer-facing launch. Internal use cases first, until the control layer exists and you still can call the firefighters.

What Does a Tested AI Kill-Switch Require?

A governed AI deployment needs a live inventory of every system in use, a named person who can halt any of them within the hour, and a tested escalation path to a human.

7. Do We Know What Winning Looks Like?

Why it decides everything: Cisco finds its Pacesetters are four times more likely to move pilots into production, with over 90% reporting gains in profitability, productivity, and innovation against roughly 60% of everyone else. The difference is not luck; it is targets defined before deployment. And when targets are concrete, results read like this: Bell, Canada's leading telecommunications provider, reported a 25% improvement in customer response time and 90% positive customer feedback on AI accuracy after deploying AI agents built on ServiceNow's platform — two numbers a board can verify, not a vibe.

• Evidence to demand: written success metrics with baselines, review dates, and one named owner who reports the numbers after go-live — the same person who presented the problem statement in question one.
• Pass signal: every target in the plan is something a board could check from data — not adoption counts, not internal enthusiasm, but movement in the customer metrics this whole exercise started with.
• If you fail: do not fund the project. A deployment without a definition of winning will be declared a failure by whoever measures it first.

How Much More Likely Are AI Pacesetters to Scale Successfully?

Cisco's Pacesetters are four times more likely to move pilots into production, with more than 90% reporting gains in profitability, productivity and innovation.

Scoring Your Organization's AI Readiness: Seven Diagnostic Checkpoints

The following table highlights the most important lessons, actions and strategic considerations emerging from this seven-question AI readiness diagnostic.

Key Area	What Happened	Why It Matters	Recommended Action
Problem Definition	99% of AI-ready organizations have a defined strategy vs. 58% of others	Strategy is the widest readiness gap in the market	Write a one-page problem statement naming the customer pain and baseline metric
Budget Allocation	Successful programs spend 70% on people/process, not technology	Inverted budgets are the most reliable failure predictor	Rebalance spend before funding vendor contracts
Data Governance	Only 19% of organizations have centralized data infrastructure	Ungoverned or non-consented data kills projects and trust	Map data sources for one use case before scaling enterprise-wide
Process Design	Automation amplifies existing workflow friction	Redesigned workflows show the strongest link to profit impact	Walk the process end-to-end and design for exception cases first
Workforce Skills	Only 31% of organizations rate talent as highly AI-ready	Untrained users route around tools rather than rejecting them loudly	Fund role-specific training with protected practice time on approved tools
Governance & Control	No tested kill-switch or escalation path in most deployments	An ungoverned customer-facing system risks a brand incident	Build a live inventory, a named stop-authority and a human escalation path before launch
Success Metrics	Pacesetters are 4x more likely to scale pilots to production	Undefined success gets declared a failure by whoever measures first	Set written, board-verifiable metrics with baselines before deployment

What Does Your Readiness Score Actually Mean?

Count your passes. Five or more out of seven: you are readier than most of the market — start, start small, and start now. Three or four: your first AI project is the preparation itself, and that is not a defeat; it is the work. Two or fewer: be grateful you ran this meeting before the vendors did.

One last number for perspective: Cisco has now run its readiness index for three consecutive years, and the share of fully-ready organizations has stayed stuck around 13% the entire time. Three years of historic investment and near-universal urgency — and the needle has not moved. That tells you something most vendors won't: readiness cannot be purchased, which is exactly why it compounds for the few who build it. The 13% are not lucky and they are not richer; they simply did the seven answers first, and now every new model release widens their lead instead of their risk.

So my honest advice is simple: stop asking which AI. Start asking which problem — and then run these seven questions, this month, with the artifacts on the table. The companies winning with AI in customer experience are not the ones with the best models. They are the ones that could answer seven questions before they bought one.

Learn how you can join our contributor community.