Only 8% of businesses are getting ROI from AI, here's what they know
TL;DR
KPMG surveyed 2,100 executives globally and found 95% of organisations now have a formal AI strategy, yet only 8% report any measurable return on investment. The other 92% are, by the numbers, setting their money on fire. The gap is not a technology problem; it is a strategy problem. The 8% who win are automating the administrative drag around their core work; the 92% who lose are trying to automate the core work itself.
Why are 95% of AI strategies failing to produce ROI?
The data is brutal. According to CIO.com, 95% of AI pilots fail to generate any measurable P&L impact. Deloitte's State of AI in the Enterprise report found that while 54% of organisations expect to move 40% or more of their AI experiments into production within the next three to six months, only 25% have actually managed to do so.
Thousands of businesses are buying expensive licences, running exciting pilot programmes, and then hitting a brick wall when they try to scale the technology across their operations.
This is not a capability problem. The models work. The problem is that most organisations are deploying AI against the wrong problems, and have no meaningful framework for measuring whether it is working.
What does the 8% do differently?
The answer is almost insultingly simple. The 8% who are succeeding are not trying to automate the thing that makes them money. They are automating the administrative drag that sits around that thing.
The 92% who are failing are typically trying to automate their core product or service, the work that requires human judgment, deep expertise, or regulatory accountability. The 8% are doing the exact opposite: targeting high-volume, repetitive, low-judgment workflows where time disappears every single day. They automate the friction, not the function.
The four companies that prove the point
These are not hypothetical case studies. These are production-grade deployments with documented results:
- Quilter (UK wealth manager): Microsoft 365 Copilot is projected to save its highest-cost staff more than 13,000 hours per month on post-call admin, notes, CRM updates, follow-up emails. They are not using AI to give financial advice. They are using it to eliminate the friction around financial advice.
- Flynn Group: Used Workday Paradox to automate 90% of the administrative hiring process. Result: 900,000 recruiting hours saved annually and a 21% reduction in time-to-hire. The AI schedules, screens, and coordinates. Humans still interview and decide.
- Salesforce: Their own sellers have saved over 50,000 hours through automated call summaries and conversation insights. The AI logs 440,000 sales activities monthly without human intervention. It handles the paperwork; humans close the deals.
- ServiceNow: 89% of customer self-service requests were supported by AI in 2025, saving employees more than 2.3 million hours. That is not a pilot programme. That is a fully scaled, production-grade deployment delivering measurable, repeatable value every single day.
The common thread across every one of these success stories is identical: they automated the boring stuff. The parts nobody wanted to do, that consumed enormous amounts of time, and that required no strategic human judgment.
Globe Telecom in the Philippines is another data point, 80% Gemini adoption across their workforce, with employees saving three to four hours per week through AI automation and custom chatbots. The gains are not coming from replacing core telecommunications work. They are coming from eliminating the administrative overhead that was drowning their teams.
See where AI fits in your business. Free.
A 45-minute audit. We map the highest-value automations and what they're worth in time and money. No pitch, no pressure.
The SME advantage nobody is talking about
Large enterprises move slowly. They have legacy systems, twenty-year-old databases, and six-month procurement processes standing between identifying a problem and deploying a fix. SMBs have none of that.
The CIO.com report highlighted that the AI adoption gap between large enterprises and SMBs is narrowing rapidly, precisely because SMBs have fewer legacy systems and shorter decision paths. If a small business owner sees their sales team wasting ten hours a week on CRM updates, they can deploy an AI tool to fix that specific problem on Monday morning.
That structural advantage is real, but it only holds if the right governance is in place. Without governance, speed becomes a liability. Every deployment becomes a risk, and progress eventually grinds to a halt.
The governance paradox: Australia leads on policy, lags on results
KPMG identifies Australia as a global leader in responsible AI governance, and a laggard in productivity gains. Australian businesses are very good at writing policies and setting up ethical frameworks. They are struggling to convert those frameworks into actual business value. Only 25% of Australian organisations are successfully turning AI experiments into production.
Singapore is taking a different approach. The Infocomm Media Development Authority (IMDA) has proposed the world's first international standard for testing generative AI systems, focusing heavily on benchmarking and red-teaming methodologies, building trust in the systems themselves, rather than writing policies about how humans should use them.
As Grant Thornton points out, organisations with strong AI controls actually move faster, not slower. When your team knows exactly what data they are allowed to use, what tools are approved, and how to verify output, they can deploy with confidence. Governance done right is not a handbrake, it is an accelerator.
The rework trap: how AI wastes the time it promises to save
Workday research found that nearly 40% of AI time savings are lost because employees have to fix low-quality AI output.
If you use AI to draft a complex, nuanced report and your employee spends three hours rewriting it because the tone is wrong and the facts are hallucinated, you have not saved any time. You have moved the effort from drafting to editing.
The rework trap is most common when organisations over-automate, when they deploy AI against tasks that require nuanced judgment, variable context, or specialised expertise. The fix is straightforward: pull back, and focus AI only on tasks where it consistently produces reliable, high-quality output that does not require constant human correction.
The performance gap is accelerating, not narrowing
PwC's 2026 AI Performance Study found that the top 20% of AI-adopting companies are capturing 74% of all AI-driven economic returns. These leaders are:
- 2.6 times more likely to have reinvented their business model around AI
- Delivering 7.2 times higher returns than the average adopter
The gap is not narrowing. It is accelerating. Every month spent running aimless pilots is a month your competitors spend compounding their lead.
The tool itself is almost irrelevant. A $20-per-month AI subscription deployed against the right workflow will outperform a $500,000 enterprise AI platform deployed against the wrong one, every single time.
Whether you use ChatGPT, Copilot, Gemini, or Claude, the technology is broadly comparable for most business use cases. The difference between success and failure is not the model you choose. It is the workflow you target, the data you feed it, the governance you wrap around it, and the way you measure the outcome.
What to do this week
Stop buying broad AI licences and hoping something sticks. Do this instead:
- Map your time sinks. Identify the three to five administrative workflows where your team loses the most hours each week, post-call notes, CRM updates, scheduling, document drafting, data entry. High volume and high repetition are the signal.
- Measure cycle time, not prompts. Do not measure how many prompts your team writes or how many hours a vendor claims you will save. Measure whether the workflow is actually moving faster. Did time-to-hire fall? Has post-call admin shrunk? Are your sales reps spending more time talking to clients?
- Watch for the rework trap. If your team is spending more time fixing AI output than they used to spend doing the work manually, you have over-automated. Pull back and focus on tasks where the AI produces reliable output that does not require constant correction.
- Treat governance as an accelerator. Define what data is allowed, which tools are approved, and how to verify output. Teams with clear guardrails deploy faster and with more confidence than those without.
- Start surgical, then scale. Pick one workflow. Prove the return. Then move to the next. The 8% did not get there by scattering AI across every department at once. They got there by being precise.
Where to from here
Book a free 60-minute AI audit, we'll explore exactly what workflows are worth augmenting with AI.
Live with passion & AI,
Brett
Want this installed in your business?
Bespoke AI implementation across your operations: strategy, build, rollout, and ongoing drift maintenance.
Frequently asked questions
What percentage of businesses are actually getting ROI from AI?
+
According to KPMG's survey of 2,100 executives globally, 95% of organisations have a formal AI strategy, but only 8% report any measurable return on investment. The other 92% are failing to convert their AI spending into business value.
Why do most AI pilots fail to reach production?
+
Deloitte's State of AI in the Enterprise report found that while 54% of organisations expect to move 40% or more of their AI experiments into production within three to six months, only 25% have actually managed to do so. The gap comes down to poor workflow targeting, weak governance, and the absence of outcome measurement.
What is the AI rework trap?
+
Workday research found that nearly 40% of AI time savings are lost because employees have to fix low-quality AI output. If your team spends more time correcting AI-generated work than they used to spend doing it manually, you have over-automated and shifted effort from creation to editing.
How are the top 8% of companies using AI to save time?
+
They are automating administrative drag, not core work. Examples include Quilter saving 13,000 hours per month on post-call admin, Flynn Group saving 900,000 recruiting hours annually through automated hiring coordination, and ServiceNow having 89% of customer self-service requests handled by AI in 2025, saving over 2.3 million employee hours.
Why is Australia lagging on AI productivity despite leading on governance?
+
KPMG identifies Australia as a global leader in responsible AI governance but a laggard in productivity gains. Australian businesses are effective at writing policies and ethical frameworks, but only 25% are successfully turning AI experiments into production, suggesting governance is being treated as a compliance exercise rather than a business enabler.
How much more are the top AI adopters earning compared to the average?
+
PwC's 2026 AI Performance Study found the top 20% of AI-adopting companies capture 74% of all AI-driven economic returns and are delivering 7.2 times higher returns than the average adopter. They are also 2.6 times more likely to have reinvented their business model around AI.
Does it matter which AI tool I use, ChatGPT, Copilot, Gemini, or Claude?
+
Not significantly. For most business use cases the technology is broadly comparable across platforms. The difference between success and failure is the workflow you target, the data you feed the system, the governance you wrap around it, and how you measure outcomes, not the model you choose.
Brett is a four-time founder (Darra Tyres, Gladfish, EzyTrac, Anaboo) and the operator behind AIOS, Anaboo's AI Operating System. He writes from inside the build, installing AI in his own businesses first and reporting back what actually moves the numbers. Based between Singapore, the UK and Australia.
