Stop with the AI hype already. Nail the Business Case; and the Headlines Take Care of Themselves.  

I’ll hold my hands up, I’m absolutely guilty of putting big numbers on slide one of a pitch. 

“40% drop in handling time.” 

“£2m saved.” 

“50% containment.” 

But when I do it, it’s only because I’ve crunched the data and backed it up with over 25 years of hard contact centre operational experience.  

In an early-stage engagement, sure, I’ll put a speculative, research-based outcome on the table depending on the industry or use case to show the “potential”. But the second we get into the weeds; you absolutely must get into the detail. 

Unfortunately, that’s where most of the industry falls apart. 

Ask the average vendor where their headline number came from, what the baseline was, what occupancy and shrinkage they factored in, how attrition was built into the data, or what it’ll actually cost to get there, and the room goes completely quiet. 

Writing a flashy headline is the easy bit. 

But if we actually want AI to graduate from shiny pilot projects into the engine room of how businesses run, we have to drop the hype and bring back basic commercial discipline.  

Here’s what that actually looks like. 

1. Stop obsessing over the token cost and look at the “Return on Intelligence” 

Let’s be real, when we talk about Agentic AI voice, obsessing over the vendor’s per-minute or per-token rate is missing the point completely. Our Chief AI Officer, Stuart Dorman, talks about this in a few recently published blog posts (Why You Should Aim for 10x Return on Intelligence / The Discipline Behind Achieving a ‘Return on Intelligence’) as “Return on Intelligence.” I see companies panicking because their AI consumption is rising, and their immediate instinct is to figure out how to cut the bill. 

But if spending £1 on AI generates £5 to £10 in value, reducing consumption is exactly the wrong move. You’re literally destroying value. 

Think about it. What does it really matter if the cost of a fully contained call is £0.85 or £1.00 when the human equivalent is costing you £7? 

Well, it kind of does and doesn’t matter actually. You can absolutely go cheap and find a vendor who will charge you £0.25 a call. But if that bot is so basic that it only manages to contain 5% of your calls, you’re only banking that £7 saving on 5% of your volume. 

On the flip side, if paying £0.85 per call pays for a secure, well-governed, and heavily optimised agent that successfully contains 50% of your volume, which one is actually better for the business? That higher cost per call is what pays for the data cleanup, the CRM plumbing, the red-teaming, and the prompt tuning. You need to buy based on the total return on intelligence you are consuming, not the cheapest entry ticket. Chasing the lowest cost per call usually just buys you a very cheap failure. 

2. Your Return On Investment is only as good as your baseline 

Claiming a “30% drop in handle time” is frankly a joke if you didn’t measure where you started. Before I’ll stand behind any projected benefit, I want to see a few things. 

First: a real baseline, not a gut feeling. Pull months of hard data and break it down by channel and contact reason. Averages are liars. If the AI handles all the simple queries, your human agents are left with the messy, complex stuff. Your average handle time might actually go up while your operation gets way more efficient. 

Second, agree on how you’ll prove it before you launch. Control groups aren’t for science nerds; they’re how you prove to you Chief Financial Officer that the tech actually worked. 

Finally, draw a straight line to the cash. Saving 45 seconds a call is completely useless unless it changes reality: fewer hires, less overtime, or shifting people to higher-value work. But as Stu rightly points out, there is a much bigger opportunity here than just doing today’s work more cheaply. For decades, we’ve rationed customer conversations because human intelligence is expensive. AI changes those economics entirely. 

What if AI allows you to understand the context and proactively have a genuinely useful conversation with a customer before they ever need to call you furious? You haven’t just made an existing interaction cheaper; you’ve created a highly valuable interaction that previously didn’t make economic sense to have. But again, if you don’t track the financial impact of preventing that churn or driving that new revenue, the ROI isn’t real. 

3. Agree on your kill switch before you start 

This is where the industry gets incredibly sloppy. Treating AI funding like a rocket launch: approve, build, fire, celebrate. Once it’s live, the programme just stumbles onward because stopping feels too much like failing. 

I’d rather treat it like a venture bet: staged funding with brutal, pre-agreed evidence thresholds. Double down when the pilot proves itself against the baseline at a cost that makes sense at scale. Pivot when the tech works but the economics don’t. Maybe it crushes one type of call but hallucinates on another; shrink the scope, fix it, reset the case. That’s a win. 

But most importantly, pull the plug when the benefits are a coin toss or the maths falls apart. If token consumption is growing and we can’t point to better outcomes, improved CX, or additional revenue, then we absolutely should challenge the spend. Killing a failing project early isn’t a failure, it’s just a sign of a mature operation. The trick is to write these rules down before the results come in. 

Why it actually matters right now 

We are way past the era where just whispering “AI” in a boardroom unlocks a budget. Boards and CFOs are asking sharper questions, as they should, and so your numbers will need to able to hold up twelve months later For those of us leading these transformations, our job isn’t to bypass the discipline. It’s to enforce it. Own the full cost. Measure the real baseline. Focus on the true return on intelligence, and set your kill switches up front. 

Nail the business case, and the headline takes care of itself.