ANALYSIS
McKinsey and PwC don't often agree on anything substantive. When two firms that size run separate studies and land within shouting distance of each other on the same finding, that's a data point worth taking seriously.
Both found the same thing independently: somewhere between 6 and 20 percent of organizations are capturing essentially all of the economic value being generated by the current wave of AI adoption. Not most of it. Not a disproportionate share. Essentially all of it.
Sit with that number.
This is not a situation where early movers got a head start and the rest are catching up. This is concentration. The kind that compounds. The kind where the gap doesn't close with time. It widens with it.
If you run a business, manage a campaign, or lead anything that competes for resources, attention, or outcomes, this is the number that matters. Not the hype. Not the backlash. The distribution.
Six to twenty percent. Everyone else is somewhere between "running pilots that won't compound" and "waiting for a better moment."
There is no better moment coming.
The Adoption Clock
We have a decent historical record on how transformative technologies get adopted. The internet went from novelty to business necessity over roughly 15 to 20 years, counting from commercial availability in the early 1990s. Mobile took a similar arc -- 12 to 15 years from the first smartphones to the point where not having a mobile strategy was genuinely indefensible.
Both followed an S-curve: slow early adoption, fast mid-curve penetration, then saturation. Companies that recognized the inflection point and moved during the fast phase captured most of the structural advantage. Companies that waited for stability found themselves playing catch-up against competitors who had already optimized their operations, built institutional knowledge, and accumulated the data flywheel effects that make the technology more powerful the longer you run it.
AI is moving through that same S-curve 3 to 4 times faster. That's an analytical inference, not a cited forecast -- it's labeled as such because the honest version of this argument doesn't require citing someone else's prediction to make its case. The inputs are observable: training costs have dropped faster than Moore's Law predicted for silicon. Deployment friction collapsed. Every major software category now ships AI-native features by default. The barrier to access effectively hit zero for most business use cases sometime in 2024 and 2025.
The historical 5 to 7 year decision window compresses, by that logic, to somewhere around 18 to 36 months. Some of that window has already closed.
The companies in the 6 to 20 percent aren't smarter than everyone else. They moved earlier and built faster. The compounding effects are now doing the work.
What Forrester Actually Found
There's a data point circulating right now that's getting cited as validation for the wait-and-see position. Forrester found 25 percent of enterprises deferring or reducing AI spending. Decision-makers are reading that as: "Even the big players are pulling back. There's time."
There isn't.
What Forrester actually found was enterprises pulling back on speculative pilots -- AI initiatives that were exploration theater rather than production deployments. That's a healthy signal. It means organizations are getting disciplined about where they invest. It does not mean AI adoption is stalling at the organizations that matter.
The organizations in the 6 to 20 percent are not running pilots. They are in production. Their AI implementations are generating real outputs -- in some cases, they are the outputs. They're not running experiments to "understand the technology." They're using it to undercut competitors' cost structures, accelerate decision cycles, and widen the capability gap.
Deferring a speculative pilot and pulling back from production deployment are not the same thing. Treating them as equivalent is how you use Forrester's data to justify a position it doesn't actually support.
Leaders are widening their gap every quarter. The 25 percent deferral finding doesn't change that. If anything, it concentrates the advantage further: the organizations that stayed in production while others trimmed pilots are now farther ahead than they were six months ago.
The Political Test Case
There is no industry where the AI disruption story is moving faster, or where the stakes of getting it wrong are more concrete, than political campaigns. The data is stark.
A 2025 study published in Nature examined the effect of AI-driven voter persuasion at scale. The finding: AI chatbot conversations shifted voter position by 3.9 percentage points in roughly six minutes of interaction. For context, that's approximately four times the effect size of traditional political advertising. Four times. In six minutes.
At-scale voter persuasion targeting now costs under
The American Association of Political Consultants found 83 percent of professional political consultants now use AI tools weekly. That's not AI-curious. That's operational integration. Down-ballot AI adoption is projected to hit 65 percent in 2026 election cycles.
For campaign operatives, these numbers describe a profession being restructured around you in real time. The question isn't whether AI matters in campaigns. It clearly does. The question is whether the organizations and candidates using it most effectively will have a decisive structural advantage over those who aren't.
If the Nature finding holds at scale, the answer is yes. By a wide margin.
The Disintermediation Vector
Here's the piece that goes underreported in every AI-and-politics conversation: what AI does to the consulting model itself.
Traditional campaign consulting was built on information asymmetry. The firm knew things the candidate didn't. Voter modeling, targeting, messaging optimization, media buy efficiency -- these were expert functions that required specialized firms because the candidate couldn't access or process the underlying data. The consulting model is a direct product of that asymmetry.
AI collapses information asymmetry. Not immediately, not universally, but directionally and fast. The candidate who can operate a modern political AI stack directly -- pulling targeting models, running persuasion analysis, iterating messaging -- no longer needs intermediaries for those functions at the same cost basis. The limiting factor right now isn't the technology. It's candidate confidence to operate the tools.
That confidence is growing. Fast.
For political consulting firms, the strategic question is whether your value proposition is still the information asymmetry or something else. Firms that have answered "something else" and are building around it will survive the transition. Firms that haven't answered the question yet are watching clients figure it out on their own.
The same dynamic plays out in every professional service industry where the billable function is analysis and interpretation of complex data. Legal. Finance. Healthcare strategy. Marketing. The firms that can clearly articulate what value they create beyond data access and interpretation will be fine. The ones that can't are in the window the McKinsey and PwC data describes -- watching concentration happen to their industry in real time.
Why Concentration Happens Instead of Diffusion
The McKinsey and PwC findings don't mean the bottom 80 percent get nothing from AI. They mean the value distribution is highly skewed, and the structural reasons for that are worth understanding.
Data flywheel effects favor incumbents. The longer you run an AI system on real operational data, the more precise it gets. An organization that started 18 months ago has 18 months of feedback loops, model refinement, and institutional knowledge built around AI-augmented workflows. A new entrant starting today starts from zero -- even with access to the same base models.
Workflow integration is harder to copy than capability access. The gap between "we have a subscription to an AI tool" and "AI is integrated into every decision and workflow" is enormous. The second state requires months of process redesign, institutional learning, and organizational change. You can't shortcut it by buying a better tool later.
The talent question resolves in favor of early movers. People who are good at working with AI systems want to work somewhere those skills matter. Organizations that built serious AI practices first attracted the capable people. Organizations building serious AI practices now are hiring from a smaller pool at higher cost.
None of this means late adoption is worthless. It means the ceiling for late adopters is lower and the path to it is longer than it was 18 months ago. That gap widens every quarter.
The Decision Window
If you accept the adoption trajectory argument -- 3 to 4 times faster than internet or mobile, an 18 to 36-month effective window, much of it already elapsed -- then the decision point is not in the future. It's now.
Stop treating AI as a tool category and start treating it as operational infrastructure. The organizations in the 6 to 20 percent are not asking "which AI tools should we buy." They're asking "which workflows does AI make us categorically better at, and how do we rebuild those workflows around that capability." Different question. Different budget. Different outcome.
Build in production, not in pilots. Pilots are education. Production is compounding. The Forrester finding about 25 percent deferral is a signal that speculative pilots are getting cut. Cut yours too. Redirect that investment into production deployments where you can measure real output improvement, real cost reduction, real speed gains.
Prioritize where competitive leverage is highest. In political campaigns, the Nature study data says that's targeting and persuasion, where the effect sizes are large relative to traditional methods. In professional services, it's the functions where speed and volume of iteration matter. In operational businesses, it's forecasting and process optimization where data already exists. Don't attempt everything. Find the two or three places where AI changes your output ceiling and move there first.
Accept that you're not waiting for the technology to mature. The technology is mature enough. The organizations in the 6 to 20 percent aren't running future-ready experimental systems. They're running current tools aggressively. What's unsettled is not the technology. It's your organization's willingness to operate with it at full commitment.
What Waiting Costs
The case for waiting has one argument: the technology is still changing fast, so wait for stability before committing serious resources.
It's not a terrible argument in isolation. It fails in context.
The organizations winning the AI transition did not wait for stability. They built during instability, acquired institutional knowledge, and are now positioned to absorb the next wave of capability improvements from a stronger operational base than you'll have if you start at the next wave.
Every quarter of productive AI deployment your competitor has that you don't is a quarter of compounding advantage. Not linear. Compounding.
In the political context, this is unambiguous. The consulting firms and campaigns running AI-augmented operations right now are developing capabilities that will be decisive in 2026 and 2028 cycle contests. Campaigns that adopt late will hire staff to run tools their opponents have already optimized, against opponents who've built 12 to 18 months of institutional knowledge around those tools. The 3.9-point persuasion advantage from the Nature study doesn't shrink because you're a slow adopter. It runs against you.
The gap between the 6 to 20 percent and everyone else is not primarily a technology gap. It's an execution timing gap. Execution timing gaps don't close when the technology improves. They close only when one competitor adapts fast enough and another doesn't.
The Conservative Case for Moving Now
The conservative instinct is skepticism toward hype. That instinct is healthy. The AI hype cycle has produced a lot of noise, burned budget, and some genuinely embarrassing corporate failures built on tools deployed before the organization understood what they were buying.
But skepticism toward hype is not the same as caution toward structural change. Free markets reward participants who correctly identify structural shifts early and allocate resources accordingly. The organizations in the 6 to 20 percent are not technology enthusiasts running on optimism. They're capital allocators who read the adoption curve correctly and made resource decisions based on it.
The conservative case for moving now is not "AI is amazing and will change everything." It's simpler: the data on competitive concentration is clear, the adoption window is compressing, and waiting compounds the cost of entry. That's not hype. That's reading the field.
Prudent resource allocation is not the same as strategic delay. Organizations that wait for certainty in a fast-moving competitive landscape are not being careful. They're making a resource allocation decision by default. And that default is increasingly expensive as the gap widens.
The 6 to 20 percent didn't get there by reading whitepapers. They got there by building.
The window is open. Not wide. Use it. Or don't. But don't complain when firms like Dark Horse Political take your clients to the woodshed.
Christopher Gergen is the founder of Dark Horse Political and the author of WIN., a campaign operations handbook used by Republican campaigns nationwide. DHP builds AI-driven political operations through the NOX platform.