The 10-20-70 Rule for AI Success

Todd Curzon

[calm] Welcome to the show. Daniel, I want to start with a ratio that I think explains a great deal of the confusion in boardrooms right now: 10, 20, 70. Ten percent of AI success is the model. Twenty percent is infrastructure. And the remaining SEVENTY percent is people and process. That is the uncomfortable math. Because if that ratio is even roughly right, then the executive team spending six months comparing vendors, debating models, and touring demos is often lavishing attention on the visible 30 while neglecting the harder, slower, and more decisive 70.

Daniel Carter

[questioning tone] Wait -- the 10, 20, 70 split is the part I want people to sit with. Ten on the model, twenty on infrastructure, seventy on people and process. So if a leadership team has twelve meetings on model selection and one vague town hall on how work will actually change... they've inverted the math.

Todd Curzon

Exactly. And I think the reason this happens is rather human. Tools feel concrete. You can point to them. You can price them. You can arrange them neatly on a slide with logos and capability maps and little green checkmarks. Organizational readiness is much less photogenic. It lives in awkward questions: Who owns the output? What data is trustworthy enough to use? Which decisions may be accelerated, and which still require human review? What happens when the tool is wrong -- and it will be wrong, at least sometimes.

Daniel Carter

[skeptical] I slightly disagree on the reason, though. I don't think it's just that tools are photogenic. I think tech leaders, and honestly plenty of business leaders too, get a little emotional relief from the stack. The stack gives them the feeling of progress. You sign a contract, spin up a pilot, post the announcement on LinkedIn... and everybody gets to feel modern without touching incentives, accountability, or management behavior.

Todd Curzon

[reflective] That's fair. You're right -- it's not merely visibility, it's emotional convenience. The tooling decision can masquerade as transformation. Whereas genuine readiness asks for a more humbling admission: perhaps the bottleneck is not that we lack a model, but that our operating habits are muddled. And leaders do not always enjoy discovering that the impediment sits much closer to home.

Daniel Carter

And for newly promoted VPs and Directors, that's the real test. Because they inherit this pressure from above -- "What's our AI strategy?" -- and they feel pressure from below -- "Are we replacing jobs? Are we changing workflows? Who decides?" If you're sitting in that seat, the wrong move is to answer a leadership problem with a shopping exercise.

Todd Curzon

Yes. AI transformation is mostly a leadership and operating-model challenge, not a model-selection challenge. Or, to put it a bit more bluntly, the model is just the beginning of the sentence. The organization has to finish it. If teams do not trust the outputs, if managers do not know when to override them, if workflows were never redesigned to incorporate them, then the most impressive tool in the world becomes a slightly expensive curiosity.

Daniel Carter

[matter-of-fact] And I think that's why leaders keep solving the wrong 30 percent. The wrong 30 percent is easier to buy, easier to demo, easier to explain upward. The right 70 percent is slower. It involves training plans, role clarity, governance, and people saying, "I don't actually know how this fits my week." That's messier. But that's where adoption lives. And adoption -- not novelty -- is what creates value.

Todd Curzon

[warmly] Beautifully put. Novelty attracts attention. Adoption creates return. So as we go through this, I think the invitation is simple: each time you feel yourself drawn toward the glamour of the stack, ask whether you are solving the visible 30 or the decisive 70.

Daniel Carter

[calm] The reason leaders over-index on vendors and platforms is pretty straightforward: those are the easiest things to evaluate in a meeting. You can compare Vendor A against Vendor B. You can watch the demo. You can score security, integration, cost, speed. That's a familiar executive muscle. It feels like procurement mixed with strategy. But the moment you leave that visible layer, the conversation gets murky fast. Is the underlying data reliable? Is the workflow stable enough to automate? Who has decision rights if the AI output conflicts with a manager's judgment? Those questions are harder, and they don't fit nicely into a demo.

Todd Curzon

[curious] The phrase "decision rights" there matters. Because if those rights are unclear, even a strong model can fail in a very ordinary way. Not dramatically -- just quietly. The tool produces something useful, nobody is certain who may act on it, and so the output sits there like a beautifully drafted memo that never leaves the drawer.

Daniel Carter

Exactly. Picture a sales organization using a very capable model to generate next-best-action recommendations for live deals. On paper, brilliant. But if the CRM data is patchy, the account executives don't trust the recommendations, and the regional managers haven't agreed on when reps should follow the AI versus when they should use judgment, the model doesn't drive behavior. It creates friction. Some reps ignore it. Some overuse it. Some work around it in private. So leadership says, "The model underperformed." No -- the system around the model underperformed.

Todd Curzon

[questioning tone] Let me try to explain that back. The failure wasn't in the recommendation engine itself. The failure was in the chain surrounding it: incomplete CRM inputs, low trust from the reps, and blurry authority between rep and manager. So the AI was inserted into ambiguity and then blamed for reflecting it.

Daniel Carter

Yes -- "inserted into ambiguity" is the phrase. I'm stealing that. [chuckles] Because that's what happens all the time. Leaders drop a tool into a workflow they haven't truly examined in years, and then they're surprised the mess gets faster rather than better.

Todd Curzon

I had a version of this earlier in my career, though not with AI. We introduced a new executive communications platform -- expensive, elegant, very polished -- and I remember thinking the rollout itself signaled progress. We had training decks, launch emails, dashboards... activity everywhere. But after a month, very little had changed. And the reason, embarrassingly, was simple: we had upgraded the channel without clarifying the behavior. Nobody knew what a good update looked like, what cadence mattered, or which decisions those messages were supposed to accelerate. We improved the instrument panel and forgot the vehicle.

Daniel Carter

[laughs softly] "Improved the instrument panel" -- that's memorable. Mine was a little more blunt. Years ago, I coached a leadership team that kept telling me, "We've got tremendous momentum on digital transformation." I finally asked them for one operating metric that had actually improved. Cycle time? Error rate? Revenue per rep? Escalations? Anything. And there was this... [pauses] silence. They had meetings about the technology every week, but the business itself wasn't moving. They mistook organized attention for progress.

Todd Curzon

Which is such an important distinction. Activity around technology is not the same thing as changed performance. And with AI, because the demos are so compelling, the confusion becomes even more seductive. The visible part of the iceberg is impressive. The submerged part -- data quality, workflow design, trust, governance, decision rights -- is what determines whether the ship actually clears it.

Daniel Carter

[skeptical] And this is where I push leaders a little. If your AI review is all about platform capability and none of it is about manager behavior, you've probably wandered into theater.

Todd Curzon

[reflective] So let us descend, for a moment, into the less glamorous and more consequential territory where the 70 percent lives. AI impact is usually driven by six rather unfashionable things: adoption, training, trust, role clarity, governance, and workflow redesign. None of those will dazzle an audience in a product demo. Yet every one of them determines whether the tool becomes part of the working day or remains a decorative experiment.

Daniel Carter

Adoption is the first hurdle. And I mean actual adoption, not log-in counts. A team can log in once because the VP told them to. That's compliance theater. Real adoption means the tool gets used when the pressure is on -- on the busy Tuesday, in the live deal, during month-end close, in the customer escalation. If people abandon it the moment the stakes go up, then it was never adopted.

Todd Curzon

[calm] That distinction between log-in counts and busy-Tuesday behavior is excellent. Training comes next, and here leaders often underinvest badly. They assume a short demo equals capability. It does not. People need to know what the tool is for, where it tends to be strong, where it is unreliable, and how its output should be checked. In other words, they require judgment training, not merely button training.

Daniel Carter

And trust is peculiar, because it cuts both ways. Too little trust and people ignore the system. Too much trust and they stop thinking. The sweet spot is calibrated trust: "I know when this helps me, I know when to verify, and I know when to override." That's a management issue as much as a technical one.

Todd Curzon

Then role clarity. Who owns the prompt design? Who validates the output? Who is accountable if the answer is wrong? Who decides whether a workflow should be redesigned around the tool? If those lines are blurred, teams become tentative. And tentative teams tend to work around the tool quietly rather than through it visibly.

Daniel Carter

[matter-of-fact] Which brings us to the most underrated group in this whole story: middle managers. Not the C-suite on stage, not the vendor in the demo -- the middle managers. They're the ones translating broad strategy into Tuesday morning behavior. They decide whether AI use is encouraged, tolerated, or subtly punished. They set the tone in one-on-ones. They decide what gets reviewed in staff meetings. They tell people, sometimes without saying it directly, what's safe.

Todd Curzon

Yes, and frontline teams are exquisitely sensitive to those signals. If leadership announces an AI strategy with tremendous fanfare but leaves incentives untouched, meeting cadence unchanged, and operating norms exactly as they were, employees infer the truth very quickly: this is optional, symbolic, or dangerous. One might say, with some sadness, that the organization has issued a press release to itself.

Daniel Carter

[chuckles] A press release to itself -- yes. And you see this trap all the time. The CEO says, "AI is now a top priority." But managers are still rewarded for output volume, not process improvement. Team meetings still review the same old metrics. Nobody has time blocked to test new workflows. There's no guardrail for risk, so people get cautious. Then six months later leadership says, "We're disappointed in adoption." Well... what exactly did the system reward?

Todd Curzon

[softly] This is why AI transformation is, at heart, managerial. It is about the redesign of work, the permission structure around experimentation, and the discipline to change routines. Strategy announced without norms revised is merely aspiration with a microphone.

Daniel Carter

And if you're a new VP or Director, that's actually good news. Because you may not control the whole enterprise stack, but you absolutely can influence incentives, meeting rhythms, role clarity, and local trust. That's where the 70 percent starts becoming real.

Daniel Carter

[warmly] So what do AI-ready leaders actually do? First, they treat their job as redesigning work, not just approving tools. That's the shift. They don't ask, "What platform should we buy?" and stop there. They ask, "Which decision, workflow, or bottleneck are we trying to improve, and what would better look like in measurable terms?" Cycle time, accuracy, conversion, response speed, error reduction -- something concrete. They measure outcomes, not novelty.

Todd Curzon

[calm] I think that is the essential move from curiosity to readiness. Curiosity says, "This technology is interesting." Readiness says, "This organization can absorb it." And those are not the same thing at all. Many firms are AI-curious in the way one might admire an exquisite mechanical watch in a shop window. Far fewer are AI-ready in the sense of being able to incorporate it into the practical rhythm of working life.

Daniel Carter

Let's make it concrete. Pick ONE high-value use case. Not twelve. One. Something painful enough that people care, frequent enough that learning compounds, and bounded enough that you can manage risk. Then assign a named owner. Not a committee, not "the innovation team," not a vague cross-functional working group. One owner.

Todd Curzon

[questioning tone] And when you say "named owner," you mean the person accountable for both adoption and outcome, yes? Not simply the technical setup.

Daniel Carter

Exactly. Accountable for the business result. Then set guardrails. Where may the tool be used? Where may it not? What requires human review? What data is in bounds? What is the escalation path if the output is wrong or risky? Guardrails reduce fear because they replace ambiguity with judgment.

Todd Curzon

And then, crucially, create a feedback loop. This is where leaders often stop too soon. They launch, they circulate a note of triumphant optimism, and then they disappear. Far better to establish a regular cadence -- perhaps weekly at first -- where teams report not just successes but friction. What failed? What was ignored? Where did trust break down? Which step in the workflow still makes the tool cumbersome? Momentum is created by structured follow-through, not by ceremonial launch.

Daniel Carter

[reflective] There's also a deeper question I think leaders should ask before any deployment: can the organization actually absorb this tool right now? Do managers have the bandwidth to coach new behavior? Is the process stable enough to redesign? Are the incentives aligned, even a little? Because if the answer is no, the honest move may be to build the muscle first rather than rush the implementation.

Todd Curzon

Which may sound cautious, but is in fact rather ambitious. It asks the leader to resist the seduction of appearing advanced in favor of becoming capable. And capability, unlike enthusiasm, has a way of compounding over time.

Daniel Carter

[skeptical] And just to keep us honest, this is where some leaders bristle. They say, "If we don't move fast, we'll fall behind." Fair concern. But speed without absorption isn't speed. It's skid. You burn trust, confuse teams, and teach the organization that AI equals noise.

Todd Curzon

[warmly] Beautifully said. Speed without absorption is skid. The AI-ready leader is not the one collecting the most demos, but the one building a system in which useful change can actually take root.

Todd Curzon

[reflective] The final point, and perhaps the most strategic, is that AI readiness should be understood as a leadership capability that compounds. It is not a one-time implementation project, neatly bracketed by kickoff and closeout. It is a muscle. The organization learns how to evaluate use cases, how to set guardrails, how to redesign workflows, how to train judgment, how to gather feedback, and how to scale what works. Once those muscles strengthen, each subsequent wave of technology becomes more manageable. Without them, each wave feels like a fresh emergency.

Daniel Carter

And that compounding matters in high-growth tech because the market will not wait for a perfectly timed retreat. Pressure arrives unevenly. A competitor cuts response time. A customer expects faster proposals. A board asks tougher questions. The leaders who do well aren't always the ones with the flashiest prototype. They're the ones whose teams can absorb change without drama.

Todd Curzon

Which is precisely why we've been speaking with leaders about the AI Ready executive cohort course. It's designed for executives who do not merely want to admire the landscape, but to navigate it. The emphasis is on operating model, decision framework, and change-management skill -- the practical disciplines required to make AI real inside an organization rather than merely present in a strategy deck.

Daniel Carter

[matter-of-fact] Yeah, and that's the tasteful but important distinction. This isn't "come learn a bunch of prompts and feel clever for an afternoon." It's about building executive readiness: how to pick the right use case, how to assign ownership, how to set governance, how to create adoption, how to manage the human side of rollout. In other words, how to handle the 70 percent people keep underestimating.

Todd Curzon

[calm] I think many leaders secretly know this. They sense, even if they have not yet articulated it, that the gap is not information but integration. They have seen the demos. They have read the memos. What they require now is a disciplined environment in which to convert scattered interest into operating competence.

Daniel Carter

And honestly, the phrase "operating competence" is the one I'd underline. Because markets reward execution, not fascination. The company that wins with AI is not necessarily the one with the most dazzling internal showcase. It's the one that can mobilize managers, clarify decisions, redesign workflows, and move people and process at speed.

Todd Curzon

[softly] Yes. The flashiest demo may command the room for a moment. But the quieter achievement -- an organization that can adapt, learn, and implement with discipline -- tends to command the future. And perhaps that is the lingering question for any leader listening: if the market forced real AI adoption in your business within the next ninety days, would your organization be impressed by the technology... or genuinely ready to work differently?

Daniel Carter

[short pause] That's the question. Thanks, Todd.

Todd Curzon

Thank you, Daniel.