M
M
e
e
n
n
u
u
M
M
e
e
n
n
u
u

November 1, 2026

November 1, 2026

49% of Leaders Scaled Back AI Agents Because Costs Beat Value

KPMG's Q2 2026 survey found 49% of leaders scaled back, narrowed, delayed or paused AI agents as costs outran value. Here's how a small business keeps AI spend visible and profitable.

KPMG's Q2 2026 survey found 49% of leaders scaled back, narrowed, delayed or paused AI agents as costs outran value. Here's how a small business keeps AI spend visible and profitable.

Nearly half of the senior leaders in KPMG's latest survey cut back AI agent rollouts because costs began to outweigh value. Only about a third could see their AI costs clearly. Here's how to make sure your AI spend pays for itself, with a one-page cost model you can fill in today.

The AI Bill Nobody Can See

Here's a number that should change how you buy AI. In KPMG's Global AI Pulse for Q2 2026, which surveyed 2,145 senior leaders across 20 countries, 49% said they had scaled back, narrowed, delayed or paused AI agent deployments because costs started to outweigh the value. Only 7% reported established return on investment. Only 35% said their AI operating costs were fully visible and actively monitored.

Those are enterprise numbers, with enterprise budgets. But the lesson travels down to a plumber, a dental practice or a law firm with ten staff. The businesses getting a return are not the ones spending the most. They're the ones who can see what they're spending and what it returns.

KPMG's own data makes the point. Organisations with full cost visibility reported established ROI at roughly 15%. Those without it reported about 3%. Treat the exact figures with the usual caution, because survey methodology varies. The direction is hard to argue with: what you can't see, you can't make pay.

This post is for service business owners who've bought, or are about to buy, AI tools. We'll cover where AI costs hide, how to build a one-page cost model, and which numbers to check every month.

Where AI Costs Actually Hide

When you sign up for a tool, the headline price looks small. The real cost is spread across five places.

1. Usage-Based Charges

Many AI tools charge per minute, per message, per lead or per "credit." A voice receptionist that costs £30 a month at low volume can cost much more when your phone rings 40 times a day. Always ask what happens at three times your current volume.

2. Tool Sprawl

One tool for calls, one for chat, one for email, one for reviews, one for content. Each has a subscription, and none of them talk to each other. You're paying five times and still copy-pasting between them.

3. Setup and Integration Time

Your time has a price. If you spend 15 hours getting a tool working, and your time is worth £50 an hour, that's £750 before it does anything.

4. Error Clean-Up

In the PYMNTS Intelligence SMB report published in October 2026, which surveyed 533 US small businesses, about 22% of AI users said AI errors had created extra work, and 28% said their AI spending had exceeded expectations. Cheap AI that needs constant correction isn't cheap.

5. The Cost of Doing Nothing With It

The most common hidden cost is a tool you pay for and don't use. It never made it into a workflow. Nobody owns it. It sits in your bank statement every month.

The One-Page AI Cost Model

You don't need a finance team. You need one table per AI system.

Line

What to record

Example (illustrative)

Monthly subscription

Fixed fee

£300

Variable charges

Minutes, messages, leads at current volume

£80

Setup cost, spread over 12 months

One-off cost ÷ 12

£40

Your time

Hours per month × hourly value

3 hrs × £50 = £150

Total monthly cost

Sum of the above

£570

Jobs it influenced

Booked jobs it touched

9

Average job value

Your number

£1,200

Share you'd have won anyway

Be honest

50%

Attributable revenue

9 × £1,200 × 50%

£5,400

Return

Revenue ÷ cost

9.5×

The numbers above are examples, not benchmarks. The structure is the point. If your return is below 3×, ask why before you add anything else. If you can't fill in the table, you've found the problem.

Five Numbers to Check Every Month

Put these on one page and review them in ten minutes.

  1. Total AI spend. All tools, all fees, one number.

  2. Spend per booked job. Total spend divided by jobs the system produced.

  3. Answer rate or response time. Is it doing its one job?

  4. Escalations and errors. How many times did a human have to step in, and why?

  5. Revenue attributed. What did it actually produce?

KPMG found that the most common cost controls among large firms were cost review in approval processes (54%), cost dashboards (53%) and usage budgets (40%). You can do the small-business version of all three with a spreadsheet and a calendar reminder.

Why Owner Accountability Changes the Outcome

One more finding from the same KPMG survey is worth taking seriously. Where the CEO was accountable for AI outcomes, established ROI was reported at around 14%, compared with about 4% elsewhere.

In a small business, you are the CEO. If AI is "someone's side project," it will drift. If you look at the numbers monthly and decide what to keep, expand or cut, it behaves like any other investment. We come back to this in another post this week, because ownership is the cheapest way to improve almost every AI result.

Four Ways to Cut AI Cost Without Cutting Results

Consolidate

Replace three disconnected tools with one connected system. We run call answering, follow-up, booking and reviews through GoHighLevel as the control layer, so there's one place to look and fewer duplicate fees.

Narrow the Job

A system with one clear job uses fewer resources and is easier to measure. "Answer inbound calls and book qualified appointments" is cheaper to run and easier to defend than "handle customer communication."

Cap Variable Spend

Set a usage ceiling and an alert. If a tool charges per minute, know what happens when you hit your limit.

Shorten the Contract

For a first deployment, avoid long lock-in. A 30-day cancellation keeps everyone honest. It's what we offer, because we'd rather earn the next month than trap you in it.

How We Handle This at My Revue

We'll be direct about our model, because it relates to the topic. Our systems have a monthly price, and on the performance model there's a 15% fee per qualified lead. That structure ties part of our income to outcomes you can see. We deliver the first build in 14 days, then spend the first month listening to calls and tuning.

We still tell some businesses not to buy. If you're getting only a handful of enquiries a week, the sums often don't work yet. A booking link and a disciplined follow-up habit may be the better spend. Honest answers like that are cheaper for you than a tool you cancel in month three.

What to Do This Week

  1. List every AI tool you pay for, with monthly cost.

  2. Next to each, write its one job and the number that proves it's working.

  3. Cancel or pause anything with no number.

  4. Build the one-page cost model for your best tool.

  5. Put a monthly 10-minute review in your calendar.

That's the whole programme. It will take less than two hours and will probably save you more than it costs.

Frequently Asked Questions

Are AI tools actually worth it for a small business?

Often, yes, but not automatically. In the PYMNTS survey, 44% of AI-using small businesses described the financial impact as mostly or completely positive, 27% saw no impact yet, and about 7.5% saw a mostly negative one. Value depends on picking a measurable job and tracking it.

What did KPMG's survey actually find?

In its Q2 2026 Global AI Pulse, 49% of surveyed leaders had scaled back, narrowed, delayed or paused AI agent deployments because costs began to outweigh value, and 7% reported established ROI. Only 35% said AI operating costs were fully visible and monitored.

How do I know if my AI tool is paying for itself?

Compare its total monthly cost, including your time, with the revenue it can credibly claim, adjusted for jobs you'd have won anyway. If the ratio is under about 3×, investigate.

Should I choose usage-based or flat pricing?

Flat pricing is easier to budget. Usage-based can be cheaper at low volume and much more expensive at high volume. Ask for a worked example at three times your current volume.

What's the fastest way to reduce AI spend?

Cancel tools with no measurable job, and consolidate overlapping ones into a connected system.

Want Us to Run the Numbers With You?

Book a free AI cost and leak audit. We'll look at what you're paying for, what it should be producing, and where the biggest gap is. If a cheaper option works better for you, we'll say so. No sales pitch. No obligation.

[Book a free AI cost audit]

Nearly half of the senior leaders in KPMG's latest survey cut back AI agent rollouts because costs began to outweigh value. Only about a third could see their AI costs clearly. Here's how to make sure your AI spend pays for itself, with a one-page cost model you can fill in today.

The AI Bill Nobody Can See

Here's a number that should change how you buy AI. In KPMG's Global AI Pulse for Q2 2026, which surveyed 2,145 senior leaders across 20 countries, 49% said they had scaled back, narrowed, delayed or paused AI agent deployments because costs started to outweigh the value. Only 7% reported established return on investment. Only 35% said their AI operating costs were fully visible and actively monitored.

Those are enterprise numbers, with enterprise budgets. But the lesson travels down to a plumber, a dental practice or a law firm with ten staff. The businesses getting a return are not the ones spending the most. They're the ones who can see what they're spending and what it returns.

KPMG's own data makes the point. Organisations with full cost visibility reported established ROI at roughly 15%. Those without it reported about 3%. Treat the exact figures with the usual caution, because survey methodology varies. The direction is hard to argue with: what you can't see, you can't make pay.

This post is for service business owners who've bought, or are about to buy, AI tools. We'll cover where AI costs hide, how to build a one-page cost model, and which numbers to check every month.

Where AI Costs Actually Hide

When you sign up for a tool, the headline price looks small. The real cost is spread across five places.

1. Usage-Based Charges

Many AI tools charge per minute, per message, per lead or per "credit." A voice receptionist that costs £30 a month at low volume can cost much more when your phone rings 40 times a day. Always ask what happens at three times your current volume.

2. Tool Sprawl

One tool for calls, one for chat, one for email, one for reviews, one for content. Each has a subscription, and none of them talk to each other. You're paying five times and still copy-pasting between them.

3. Setup and Integration Time

Your time has a price. If you spend 15 hours getting a tool working, and your time is worth £50 an hour, that's £750 before it does anything.

4. Error Clean-Up

In the PYMNTS Intelligence SMB report published in October 2026, which surveyed 533 US small businesses, about 22% of AI users said AI errors had created extra work, and 28% said their AI spending had exceeded expectations. Cheap AI that needs constant correction isn't cheap.

5. The Cost of Doing Nothing With It

The most common hidden cost is a tool you pay for and don't use. It never made it into a workflow. Nobody owns it. It sits in your bank statement every month.

The One-Page AI Cost Model

You don't need a finance team. You need one table per AI system.

Line

What to record

Example (illustrative)

Monthly subscription

Fixed fee

£300

Variable charges

Minutes, messages, leads at current volume

£80

Setup cost, spread over 12 months

One-off cost ÷ 12

£40

Your time

Hours per month × hourly value

3 hrs × £50 = £150

Total monthly cost

Sum of the above

£570

Jobs it influenced

Booked jobs it touched

9

Average job value

Your number

£1,200

Share you'd have won anyway

Be honest

50%

Attributable revenue

9 × £1,200 × 50%

£5,400

Return

Revenue ÷ cost

9.5×

The numbers above are examples, not benchmarks. The structure is the point. If your return is below 3×, ask why before you add anything else. If you can't fill in the table, you've found the problem.

Five Numbers to Check Every Month

Put these on one page and review them in ten minutes.

  1. Total AI spend. All tools, all fees, one number.

  2. Spend per booked job. Total spend divided by jobs the system produced.

  3. Answer rate or response time. Is it doing its one job?

  4. Escalations and errors. How many times did a human have to step in, and why?

  5. Revenue attributed. What did it actually produce?

KPMG found that the most common cost controls among large firms were cost review in approval processes (54%), cost dashboards (53%) and usage budgets (40%). You can do the small-business version of all three with a spreadsheet and a calendar reminder.

Why Owner Accountability Changes the Outcome

One more finding from the same KPMG survey is worth taking seriously. Where the CEO was accountable for AI outcomes, established ROI was reported at around 14%, compared with about 4% elsewhere.

In a small business, you are the CEO. If AI is "someone's side project," it will drift. If you look at the numbers monthly and decide what to keep, expand or cut, it behaves like any other investment. We come back to this in another post this week, because ownership is the cheapest way to improve almost every AI result.

Four Ways to Cut AI Cost Without Cutting Results

Consolidate

Replace three disconnected tools with one connected system. We run call answering, follow-up, booking and reviews through GoHighLevel as the control layer, so there's one place to look and fewer duplicate fees.

Narrow the Job

A system with one clear job uses fewer resources and is easier to measure. "Answer inbound calls and book qualified appointments" is cheaper to run and easier to defend than "handle customer communication."

Cap Variable Spend

Set a usage ceiling and an alert. If a tool charges per minute, know what happens when you hit your limit.

Shorten the Contract

For a first deployment, avoid long lock-in. A 30-day cancellation keeps everyone honest. It's what we offer, because we'd rather earn the next month than trap you in it.

How We Handle This at My Revue

We'll be direct about our model, because it relates to the topic. Our systems have a monthly price, and on the performance model there's a 15% fee per qualified lead. That structure ties part of our income to outcomes you can see. We deliver the first build in 14 days, then spend the first month listening to calls and tuning.

We still tell some businesses not to buy. If you're getting only a handful of enquiries a week, the sums often don't work yet. A booking link and a disciplined follow-up habit may be the better spend. Honest answers like that are cheaper for you than a tool you cancel in month three.

What to Do This Week

  1. List every AI tool you pay for, with monthly cost.

  2. Next to each, write its one job and the number that proves it's working.

  3. Cancel or pause anything with no number.

  4. Build the one-page cost model for your best tool.

  5. Put a monthly 10-minute review in your calendar.

That's the whole programme. It will take less than two hours and will probably save you more than it costs.

Frequently Asked Questions

Are AI tools actually worth it for a small business?

Often, yes, but not automatically. In the PYMNTS survey, 44% of AI-using small businesses described the financial impact as mostly or completely positive, 27% saw no impact yet, and about 7.5% saw a mostly negative one. Value depends on picking a measurable job and tracking it.

What did KPMG's survey actually find?

In its Q2 2026 Global AI Pulse, 49% of surveyed leaders had scaled back, narrowed, delayed or paused AI agent deployments because costs began to outweigh value, and 7% reported established ROI. Only 35% said AI operating costs were fully visible and monitored.

How do I know if my AI tool is paying for itself?

Compare its total monthly cost, including your time, with the revenue it can credibly claim, adjusted for jobs you'd have won anyway. If the ratio is under about 3×, investigate.

Should I choose usage-based or flat pricing?

Flat pricing is easier to budget. Usage-based can be cheaper at low volume and much more expensive at high volume. Ask for a worked example at three times your current volume.

What's the fastest way to reduce AI spend?

Cancel tools with no measurable job, and consolidate overlapping ones into a connected system.

Want Us to Run the Numbers With You?

Book a free AI cost and leak audit. We'll look at what you're paying for, what it should be producing, and where the biggest gap is. If a cheaper option works better for you, we'll say so. No sales pitch. No obligation.

[Book a free AI cost audit]

YOUR FIRST STEP

Book A
30-Minute Call.

We review your current funnel, CRM, and sales process — identify exactly where leads are leaking — and give you a specific recommendation for what to fix first.

YOUR FIRST STEP

Book A
30-Minute Call.

We review your current funnel, CRM, and sales process — identify exactly where leads are leaking — and give you a specific recommendation for what to fix first.

YOUR FIRST STEP

Book A
30-Minute Call.

We review your current funnel, CRM, and sales process — identify exactly where leads are leaking — and give you a specific recommendation for what to fix first.

Ready to start?

Get in touch

Whether you have questions or just want to explore options, we’re here.

By submitting, you agree to our Terms and Privacy Policy.

We are Based in London

Soft abstract gradient with white light transitioning into purple, blue, and orange hues

Ready to start?

Get in touch

Whether you have questions or just want to explore options, we’re here.

By submitting, you agree to our Terms and Privacy Policy.

We are Based in London

Soft abstract gradient with white light transitioning into purple, blue, and orange hues

Ready to start?

Get in touch

Whether you have questions or just want to explore options, we’re here.

By submitting, you agree to our Terms and Privacy Policy.

We are Based in London

Soft abstract gradient with white light transitioning into purple, blue, and orange hues