Every major technology cycle follows the same script. And every single time, smart people convince themselves that this one is different.
I've been working in and around digital technology for over two decades now, and the pattern is so reliable it's almost boring. A breakthrough happens. Early adopters get genuinely impressive results. The conference circuit catches fire. Vendors start making promises that stretch well beyond what the technology can actually deliver. Investment floods in. Implementations disappoint. The backlash begins. And then - this is the bit nobody talks about at the conferences - a quieter period follows where the firms that did the unglamorous groundwork start pulling ahead of everyone else.
Cloud computing followed this arc. Social media for business followed it. Blockchain followed it so dramatically that the word itself became a punchline for about three years before settling into genuinely useful applications in supply chain and financial settlement. AI - specifically generative AI and large language models - is following it right now.
The question isn't whether the AI hype cycle will deflate. It already is. The question is what you're doing about it while it happens.
Gartner's Hype Cycle is one of those frameworks that's become so ubiquitous in management consulting that people sometimes forget how useful it actually is. Five phases: an innovation trigger, a peak of inflated expectations, a trough of disillusionment, a slope of enlightenment, and a plateau of productivity. It's not a prediction tool - the timing varies enormously - but the pattern itself has held for every significant technology shift of the last thirty years.
What makes it genuinely valuable isn't the shape of the curve. It's the insight that different applications of the same underlying technology can sit at different points on the curve simultaneously. And that's exactly what's happening with AI right now.
Let me be specific, because this is where most commentary gets lazy.
Autonomous AI agents - the idea that you can deploy an AI system to make complex business decisions without meaningful human oversight - passed peak hype somewhere around mid-2024 and are firmly in the trough of disillusionment. The gap between vendor demos and operational reality became impossible to ignore for anyone who'd actually tried to deploy them. AI-generated creative work is in a similar position: the initial "this changes everything" energy has collided with the reality of quality control, brand consistency, and the fact that clients can tell when something reads like it was written by a machine.
But here's what's happening at the same time. Specific, narrower AI applications are quietly entering productive use. Document processing and summarisation. Structured data analysis. AI-assisted writing for internal purposes. Proposal generation. Research synthesis. These aren't the applications that make headlines at Davos. They're the ones that save your people two hours on a Tuesday afternoon. And they're working.
If you're treating "AI" as a single phenomenon - either all hype or all substance - you're misreading the landscape badly.
Here's a stat that should give you pause: only 7% of CFOs report seeing high ROI from their AI investments, despite widespread adoption across enterprise (McKinsey's State of AI survey, 2024). And in financial services specifically, 65% of institutions face AI implementation delays averaging fourteen months. That's not a technology failure. That's the trough of disillusionment in numerical form.
That 7% figure doesn't mean AI doesn't work. It means most organisations invested in AI the way most organisations invest in any hyped technology: too broadly, too quickly, with unclear success criteria and governance bolted on after the fact. The 7% that are seeing returns tend to share a specific set of characteristics I'll come back to.
I was in a conversation last month with the managing partner of a 200-person consulting firm. Good firm, well-run, solid reputation. He told me they'd "paused" their AI programme because "the results haven't justified the investment." When I asked what they'd actually implemented, it turned out they'd bought an enterprise licence for an AI platform, run a two-day training session, and then waited. No specific use cases. No success criteria. No governance. No measurement. They'd essentially bought a gym membership, attended the induction, and then wondered why they weren't fitter six months later.
That's not an AI failure. That's an implementation failure. And the difference matters enormously for what you do next.
I'll admit I've been guilty of a version of this myself. Early on, we ran an internal AI pilot at Distinction that was far too loosely defined - the brief was something like "explore how AI can improve our content process." Predictably, it produced a lot of activity and not much you could point to. We got sharper after that. Specific problem, specific measure, specific owner. It sounds obvious in hindsight, but the excitement of the technology makes it easy to skip the boring setup work.
The AI applications that will still be delivering measurable results in 2028 share a few characteristics. First, they were deployed against a defined problem with clear success criteria. Not "let's use AI to improve efficiency" but "let's reduce proposal turnaround time from five days to two, and we'll track every proposal through the system for ninety days." That level of specificity sounds obvious, but it's genuinely rare. Most AI implementations I've seen start with the technology and work backwards to the problem, which is exactly the wrong way round.
Second, they have governance frameworks that were designed in from the start, not added when something went wrong. Someone decided upfront who reviews the AI's output, what happens when it gets something wrong, how data quality is maintained, and what the escalation path looks like. It's not glamorous work. It doesn't make for good LinkedIn posts. But it's the difference between an AI application that scales and one that gets quietly shelved after the pilot.
Third - and this is the one most firms underestimate - they've built genuine internal capability through structured learning-by-doing. Not a training day. Not a lunch-and-learn. Actual experience of working with AI tools on real tasks, building intuition about what works and what doesn't. We've seen this pattern repeatedly at Distinction: the firms that accumulate twelve to eighteen months of hands-on operational experience with even one or two specific AI applications develop a kind of institutional intelligence that can't be bought or shortcut.
Right, let me be blunt about this, because someone needs to be.
The strategy document that nobody implemented. I've lost count of how many firms commissioned an AI strategy in 2023 or 2024 - sometimes paying serious money for it - and then filed it. It sits in a shared drive somewhere, beautifully formatted, occasionally referenced in board papers, and completely disconnected from anything operational. In three years, that document will be a monument to good intentions and wasted budget.
The marketing claims that weren't supported by reality. If your website currently says "AI-powered insights" or "leveraging artificial intelligence to deliver superior outcomes" and the actual AI involvement is a ChatGPT subscription that two people use occasionally - that's going to catch up with you. Clients are getting smarter about this. So are regulators.
The vendor contracts signed at peak hype. Between roughly mid-2023 and late 2024, AI vendors had extraordinary leverage. Firms were signing multi-year enterprise agreements for capability that wasn't operationally ready, at price points that reflected the vendor's narrative rather than the buyer's reality. Some of those contracts will deliver. Many won't. And the ones that don't will be painful to unwind.
Governance added as an afterthought. This one's the most insidious. A firm deploys an AI application, it works reasonably well for six months, then someone notices that the training data includes client-confidential information that shouldn't have been ingested, and suddenly there's a scramble to build a governance framework that should have existed from day one. I've seen this happen twice in the last year - once at a professional services firm where the remediation involved a full data audit and a very uncomfortable conversation with their legal team, and once at a financial services client where the fallout was a three-month delay to their wider AI programme. Both times, the remediation cost more than the original implementation.
If any of these sound familiar - look, no judgment. The hype cycle is designed to produce exactly these outcomes. The question is what you do about it now.
The AI hype is too much right now. We're going to wait until things settle down before we invest properly.
I hear this constantly. And honestly, I get it. The noise-to-signal ratio on AI right now is genuinely terrible. Every conference, every vendor pitch, every LinkedIn feed is saturated with claims that range from mildly overstated to completely detached from reality. The instinct to wait for clarity is understandable.
But here's what the "wait and see" approach actually produces - and I've watched this play out with cloud computing, with social media, with CRM before that. The hype deflates, and when it does, it deflates in both directions. Yes, the overblown promises disappear. But so does the easily available insight. The accessible vendor pilots dry up. The internal energy and board-level tolerance for AI investment evaporates, because the topic has been generating noise without results for two years and people are tired of it.
By the time "things settle down," you've missed the window where investment was easiest to justify, expertise was most available, and the competitive landscape was most forgiving of early mistakes.
The managing partner who waits for perfect clarity will discover that clarity arrives alongside a much harder internal conversation. Try going to your board in 2027 and saying "I'd like to invest in AI capability" after two years of visible inaction. The response won't be enthusiasm. It'll be scepticism. We've been hearing about AI for years. Why now? What's different? Show me the evidence. And you won't have any, because you didn't build any.
Meanwhile, the firm down the road that started with a single, specific AI application eighteen months earlier - say, automated document summarisation for their research team - will have real data, real productivity gains, and real institutional knowledge. They won't be an "AI leader." They'll just be quietly better at deploying technology that works, and they'll have the receipts to prove it.
The trough of disillusionment is, counterintuitively, the best time to invest in a technology. Vendors are more realistic about what their products can do. Consultants have been humbled by implementations that didn't work. The talent market is less overheated. And your competitors are pulling back.
I keep using the phrase "counter-cyclical" because it captures something important. In financial markets, the best returns come from investing when everyone else is selling. The same logic applies to technology adoption. When the conference buzz dies down and the breathless AI articles slow to a trickle, the firms that are still investing - steadily, specifically, with clear governance and honest measurement - are building compound advantage.
What does that actually look like? It's not a moonshot. It's closer to this:
Pick one process where your team spends significant time on structured, repeatable work. Proposal generation. Research compilation. Document review. Client communication triage. Something specific and measurable. Deploy an AI application against that process with clear success criteria - "reduce turnaround time by X%" or "reduce error rate by Y%." Measure it honestly for ninety days. If it works, expand it. If it doesn't, learn from it and try the next one.
At the same time, build a lightweight governance framework. Who reviews the AI's output? What data can it access? What happens when it gets something wrong? How do you audit it? This doesn't need to be a 50-page policy document. It needs to be a set of clear, practical rules that everyone understands and follows.
And invest in internal capability - not a training day, but genuine, structured experience. Get your people using AI tools on real work. Let them develop intuition. Build the organisational muscle that allows you to evaluate new AI applications as they mature, rather than starting from scratch every time a vendor sends a compelling email.
The thing about hype cycles is that the positioning decisions that matter most aren't made at the peak. They're made during the descent. While everyone else is either doubling down on the hype or retreating from it entirely, the firms that get the fundamentals right - specific use cases, honest measurement, governance from day one, steady capability building - are laying foundations that compound.
This isn't about being an AI leader. I'm genuinely not suggesting you need to be on the bleeding edge of anything. It's about being a steady, deliberate adopter who measures honestly and builds on what works. Less exciting than a keynote about the AI revolution, but considerably more likely to produce results you can actually stand behind in a board meeting.
The hype cycle will do what hype cycles do. It'll deflate, then quietly mature, and eventually the technology will become so embedded in how work gets done that we'll stop calling it "AI" - the same way we stopped calling things "cloud-enabled" or "internet-powered."
The firms that will be left standing aren't the ones that made the most noise during the hype. They're the ones that used the noise as cover to do the real work.
If you want to benchmark where you actually stand before deciding where to start, our AI readiness checklist for mid-market service firms is a sensible first step.