The work of becoming AI-ready is not glamorous—clean data, honest culture, enforceable governance, a transformed workforce. However, it is the only path that turns AI from a liability you finance into an advantage you compound.
By Evan J. Schwartz
As AI moves from experimentation to mission-critical infrastructure, waste and recycling organizations are encountering a new problem: AI technical debt. Without the right foundations—clean master data, automation-ready processes, safety controls, governance, and workforce enablement—AI systems degrade, mispredict, or fail to scale. This article outlines a practical roadmap for becoming an AI-ready waste organization, drawing on lessons from our AI Governance & Risk Framework (AIGRF), product-design principles, and global implementations. It is written as a playbook for executives who want strategic clarity rather than hype.
When the internet arrived in the business world in the late 1980s and mid-1990s, most companies threw it over their operations like a blanket and asked, “So what?” The file cabinets were still overflowing. The organization chart had not moved. Where was the value? We are living through that moment again—only this time the technology is artificial intelligence, and the stakes are far higher. You cannot drape AI over your business and expect the multiples of return it promises. Do it carelessly and you will not get nothing—you will get something worse than nothing. You will get AI technical debt: a sprawl of half-integrated pilots, ungoverned agents, leaking intellectual property, and degraded trust that becomes extraordinarily expensive to unwind.
The warning signs are already in the data. MIT’s Project NANDA studied 300 public AI deployments, surveyed 350 employees, and interviewed 150 leaders, and found that roughly 95 percent of enterprise generative-AI pilots were delivering no measurable return on the income statement (MIT Project NANDA, 2025, as reported by Mauran, 2025). The researchers were emphatic that the failure was rarely the model. It was the absence of the operating model around the model—governance, integration, metrics, and the human systems that make AI actually work inside a business.
That is the gap this article is written to close. For waste and recycling leaders, the path to an AI-ready organization runs through five foundations: 1) an honest reckoning with your culture, 2) the discipline to ask the right question, 3) a context system you can trust, 4) an enforceable governance plan, and 5) a workforce transformed from producers of output into producers of outcomes. Build those, and AI compounds. Skip them, and you are simply financing tomorrow’s debt.
Know Thyself: The Culture Challenge
Before you choose a single use case, you must answer a harder question: Who are you? The Romans had a phrase for it—nosce te ipsum—know thyself. An organization’s cultural risk posture dictates how it must sequence its AI journey, and a mismatch between strategy and culture is one of the leading causes of stalled programs. There is no universal playbook. There are, in my experience, four cultures, and each can win—but only by running its own game.
#1: Risk-Averse Cultures
These cultures begin with behavior and data governance, investing heavily in clean, dependable data sources before applying AI anywhere. These organizations take longer to get off the ground; in fact, given the chaotic and rapidly evolving nature of AI today, they often struggle to escape the planning phase at all. But once they land on a valid plan, they become juggernauts—plowing through obstacles with a force that is hard to stop and harder to pivot. Their success depends entirely on the quality of that upfront strategy.
#2: Pragmatic Cultures
This means investing in proofs-of-concept and minimum viable products to prove value before putting the business at risk. This is faster to production than the risk-averse path, but it demands greater upfront investment in single use cases that must each justify their own ROI. The approach is kind to investors—progress is visible early—and efficient in capital, but it grows methodically and misses the big picture. Pragmatists rarely architect the whole board in advance, so they accumulate overlapping POCs that do not quite stitch together. This is AI sprawl. Eventually they hit a pain threshold, consolidate, and re-architect—at which point their investment profile starts to look exactly like a risk-averse company’s, except they also pay for the rework and absorb the production interruptions that consolidation requires. The trade is fewer unknowns in exchange for less flexibility—a poor fit for a technology that evolves as fast as this one—but, like the risk-averse, once their juggernaut is built it steamrolls smaller initiatives with ease.
#3: Innovator Cultures
Innovator cultures take on more risk in exchange for agility. They can pivot and course-correct, and they do not build governance or data platforms that they are not yet sure they need; they build as the use case demands. Whether that agility reflects genuine confidence or merely well-funded experimentation is revealed only when the bet resolves. The downside is the cost of correction: when a new requirement appears, an innovator may have to go all the way back to the governance step and rebuild everything to support it. Whether the strategy wins depends heavily on whether the use case carries a valid, compelling ROI.
#4: Early Adopters
These groups resemble innovators, but refuse to pay the price of building something new. They would rather plug in something “good enough” that someone else built—not fully proven, but close enough to fit the vision and be made to work. They are forgiving of imperfect fit and willing to accept variance in their vision statement to avoid the cost of innovation. They are also the most prone to “moving the goalposts.”
I am not here to crown a winner. Each strategy can work. What matters is alignment: if you put an innovative leader in a risk-averse company and ask them to run an innovation playbook, it will fail—not because either is wrong, but because they are mismatched. This is consistent with what the broader research keeps finding: the right strategy is the one calibrated to your organization’s risk tolerance, decision-making speed, and change capacity, and choosing it is the first strategic act of AI leadership.
Do Not “Take the Hill”
The second cultural trap is the opposite of paralysis. The pressure to adopt AI is overwhelming, and many leaders respond by ordering everyone to take the hill—handing every department a budget and telling them to be creative and “see what AI can do.” This almost always ends in AI sprawl that is unmanageable, whose cost in risk and lackluster performance dwarfs any gains. Worse, a run-and-fall approach kills morale and poisons the well for future efforts to do it right.
Doing AI wrong is far worse than doing it slowly. MIT’s findings reinforce the point: tools bought and integrated into real workflows succeeded far more often than scattered internal experiments, which succeeded at roughly a third of the rate (MIT Project NANDA, 2025, as reported by Mauran, 2025).
Stop Asking “What Can AI Do?”
So, what is the right process? Set culture aside for a moment and look at the generic workflow you will later tune to fit it. First and most important: stop asking “What can AI do?” It is the wrong question. The right question is “What should AI do?” That single word—should—invites a deeper conversation about governance, security, and genuine need, and it is the question MIT’s research and every mature framework point to as the executive’s true starting point (MIT Project NANDA, 2025). Most organizations get stuck chasing capabilities and accumulating tools that never deliver enterprise value, precisely because they started with can instead of should.
Once you have a list of what AI should do for you, build a vision statement around it. Then—and this is the part most leaders underfund—build a change-management plan. Change management is the hardest part of this entire journey, and without it you will accumulate AI technical debt that is brutally difficult to reverse while your intellectual property leaks out of every corner of the business. Individual AI literacy is rising in leaps and bounds, but institutional use of AI is still in its infancy. Left to their own devices, most individuals behave as a blend of innovators and early adopters: fearless to create, share, and move the goalposts to produce an impressive personal result. That instinct rarely translates into the bottom-line impact that a business can use to justify its investment, because people think in terms of output. For AI adoption to be profitable, the strategy must focus on outcomes.
From Output to Outcomes: A Digital Workforce Problem
This is a radical change in nearly everyone’s job except leadership. Most of our best producers identify themselves and their value by the output they create, relying on leadership to convert that output into outcomes. It is the unwritten contract between leadership and the boots on the ground. Asking a top producer to suddenly own outcomes—without retraining and support—breaks that contract.
That is why AI adoption is not a technology problem. It is an organizational change-management problem, a digital workforce problem. The risk of getting it wrong is concrete and personal. The half-life of a technical skill—how long it stays useful before becoming obsolete—has collapsed from roughly eight years before 2020 to a projected two years by 2030 (Gartner, 2023). When organizations delegate cognitive work to machines without preserving deliberate human learning pathways, they risk building a hollow workforce that cannot function when the technology fails or the context shifts. In my own work guiding these transformations, I am tracking attrition of roughly 30 percent of top producers when leadership fails to help them make the leap from output to outcome (E. J. Schwartz, personal observation, 2026). That number is unacceptable, and it is avoidable. The World Economic Forum (2025) estimates that a majority of workers will need reskilling this decade, and organizations that invest in it are markedly more likely to realize positive outcomes from AI (Gartner, 2023). The lesson the market has already learned the hard way is that “right-sizing” belongs at the end of the journey, not the beginning.
Why ERPs Are not Going Away: Context = Knowledge + Data
There has been a year of talk that AI threatens traditional operational systems—ERPs, CRMs, financial platforms. I beg to differ. Nothing on earth offers context to an AI better than an operational system of record, and context is the whole game for agentic AI. Start with a truism: AI knows almost everything about your industry and how businesses in it generally operate—but it knows nothing about how you run your business. Every time an AI must reach into its model to supply a piece of information it was not given, it invites a hallucination. This is not a fringe risk; it is structural. Generative models optimize for fluent, coherent output, and that objective can conflict with factual grounding, so the model fills gaps with plausible inventions (Frontiers in Artificial Intelligence, 2025). The proven mitigation is to stop making it guess and ground the model in your own verified information through retrieval-augmented generation and similar techniques—which condition the model’s output on retrieved enterprise knowledge rather than its trained priors (Microsoft, 2025). At first, a model that guesses well feels like a blessing—intuitive, even magical. In truth it is a gap that must be filled.
Here is the distinction most data scientists will not draw for you. They will tell you, correctly, that you must get your data in order. But that is only half the truth. Context = Knowledge + Data. Focus only on data and you build half your context and leave the other half up for grabs. Knowledge is what your business does with that data. Knowing you hold 96 ninety-gallon containers in inventory is data. Knowing that you must deliver one to a new customer to service that container weekly on a route—and that your business only services these container types on Tuesdays and Thursdays—is knowledge. AI Stewards must learn to supply context, not just data. This is exactly why ERPs, CRMs, and financial systems are not going away: most of them are both data systems and process systems, and they are the foundation on which you build your agentic enterprise. Lean on them to feed your strategy.
The AI Steward: A New Skill for a New Workforce
To operationalize context, I have introduced a role I call the AI Steward. A steward may carry a domain specialization—finance, operations, sales—but the role exists to strengthen three muscles. The first is agentic communication: the ability to issue commands to agents that are clear, specific, and actionable. The second is providing context, as described above—closing the gap between what the model knows about your industry and what it must know about your business. The third is dashboard monitoring. Just as a manufacturing operator watches a PLC-driven dashboard to keep sampling on target and within conformance limits, business is becoming an operating system. Every agentic action should carry associated metrics that fall within a red, amber, or green band confirming the agent’s outputs are inside expected ranges. When they are not, the steward digs into the observability and traceability trail, finds why the output drifted, fixes either the prompt or the context, and has the agent clean up after itself. The steward is thereby elevated above output creation and made responsible—and accountable—for outcomes.
This is precisely why your best producers make your best stewards. The skills that distinguish great professionals in this era are not the expiring technical ones, but rather the durable human ones: systems thinking, communication, curiosity, and the willingness to unlearn and relearn. Those are the traits a steward runs on. But unless we help these producers transform, the data says we will lose roughly a third of them—a loss no organization can afford when the human capital in question is this scarce and this hard to replace.
Governance You Can Enforce: Observability and Traceability
You cannot steward what you cannot see. Every AI-ready organization needs a governance plan that includes enforcement, observability, and traceability. When an agent does something, you need to be able to validate it, understand why it did what it did, detect when its actions fall outside predictable and desirable ranges, and know how to train it so the same error becomes less likely. This is the heart of the AI Governance & Risk Framework (AIGRF), and it aligns directly with the U.S. National Institute of Standards and Technology’s AI Risk Management Framework, whose GOVERN function establishes accountability, defines ownership across the AI lifecycle, and matches human oversight to the level of risk (NIST, 2023). It also tracks the European Union’s AI Act, which codifies risk-based oversight and human accountability into law (European Commission, 2024). Meaningful governance is not a one-time setup; it is a permanent operating responsibility, and human-in-the-loop oversight is real only when the human retains genuine authority to intervene (NIST, 2023).
Buy, Get, Build: The TCO Trap
With governance and context in place, leaders face a tempting question: should we build it ourselves? AI has dramatically lowered the bar to creating software—but it has done almost nothing to change the total cost of ownership (TCO). This is the most expensive misunderstanding I see.
Before AI, the act of writing the software accounted for only a small slice of TCO. The overwhelming majority—by industry estimates, the maintenance, support, defect resolution, roadmap evolution, professional services, and configuration that follow—consumes somewhere between 55 and 80 percent of a system’s lifecycle cost (Vention, 2024). In my framing, development was roughly 12 percent of TCO and everything else was the other 88 percent. Now suppose AI halves the cost to build. You have shaved perhaps 6 percent off the total. A commercial software vendor spreads that remaining 88 percent across thousands of customers. When you build in-house, you spread it across exactly one customer: you.
So, unless software is your core competency, the rule is: buy first, get your vendor to add second, and build only what is unique to your differentiator or strategic advantage. MIT’s research is blunt on this point—buying from specialized vendors and partnering succeeded about two-thirds of the time, while internal builds succeeded at roughly a third of that rate (MIT Project NANDA, 2025, as reported by Mauran, 2025). If you insist on owning the full cost of ownership for enterprise software, that ownership had better be the reason you are winning in the market. Otherwise, you are fighting an asymmetric war against a competitor who simply buys what you are struggling to build. The smarter move is to let AI deliver better analytics, automation, and extensible mini-apps on top of your vendor’s systems—deferring most of the TCO because what you build sits on a foundation someone else maintains.
Use Cases: Low-Hanging Fruit, Then Reinvention
With governance and context available, you build use cases—and there are two ways to look at it: total transformation from the outset, or minor automations first, then reinvention. The low-hanging-fruit path requires that you have modeled your current processes well enough to delegate low-value-but-necessary work to agents while squeezing high-value, high-judgment work toward your humans. Meeting transcription is the classic example. Someone in every meeting was the designated note-taker; they transcribed, distributed, collected corrections, and chased follow-ups into a task system—necessary work, but low value. An AI that does it frees up hours of labor and validation. The ROI is in what your people now do with that time.
A more strategic example is customer success. Customers who feel heard and stay in active communication with a vendor are far less likely to churn; those who never hear from you—even when they have a problem—almost certainly defect to a competitor. The research is unambiguous: proactive support meaningfully lifts retention, and because retention compounds, even a 5-percentage-point improvement can swing profitability dramatically (Gallo, 2014). Crucially, 91 percent of unhappy customers never complain—they simply leave (Nalpeiron, 2024).
Let AI listen to the calls, produce the after-call summary with actions, schedule the next follow-up, identify who internally must address each issue, and assign the meetings and tasks automatically. Your customer-success agent now spends more time on more calls with more customers—reducing churn and protecting net revenue retention.
Targeting low-hanging fruit like this across the organization is not transformative on its own, but it generates enough momentum to pay for your AI tooling, demonstrate real application to your board, and give your people valuable sandbox time to learn how to work with agents. It is a great first step—and if your culture permits, one I recommend running while you build out the deeper transformation.
The Long Game: Person-Plus-AI and Asymmetric Growth
The long game is transforming the organization entirely. AI Stewards gain breadth while their agents handle the deep vertical specializations, which lets you grow exponentially and asymmetrically against your input costs. Following the customer-success example, a fully agentic-enabled department might absorb two or three times as many customers before you need to hire and train another steward. Just as the spreadsheet collapsed floors of financial analysts into a single person, properly configured agents under a person-plus-AI strategy will reduce the need for specialists across your company.
But do not release those people yet. The person-plus-AI strategy demands that you first grow your business to the limits of your now-enhanced operating system, and “right-size” only as a Provider of Last Resort (POLR) option. If you have grown as far as you possibly can, then right-size. Until then, downsizing is a finite game while asymmetric growth is an infinite one. The organizations that ran at AI too fast, targeting headcount reduction first, have already learned this the hard way. Klarna replaced roughly 700 customer-service roles with AI, then reversed course in 2025 after service quality fell—its CEO conceding that the company had “focused too much on efficiency and cost” and that “the result was lower quality, and that’s not sustainable” (Masterson, 2025). The broader trend is so pronounced that Gartner expects half of the organizations planning AI-driven service-headcount cuts to abandon those plans by 2027 (Masterson, 2025).
There is also a hard supply-side reason to hold your talent. It will be four to six years before the first university graduates an AI Steward built for the business of tomorrow. Until then, the only way to get stewards is to grow your own—a significant investment you do not want to train and then hand to a competitor. Human capital has never been this valuable.
The payoff for getting this right is decisive. With most of the market still failing to convert AI pilots into returns (MIT Project NANDA, 2025), the few who build a true agentic operating model will simply operate faster than everyone else. While your competition is still hunting for a slot on everyone’s calendar to discuss the RFP, you have already responded, won, and dispatched the service vehicle. The final stage of the agentic enterprise runs iteratively at speeds business has never seen.
The Future Is Led by Humans
So, reflect on your culture. Then settle your vision for AI, how you will govern it, how you will give it context, and how your people will transform from producers of output into producers of outcomes. That is how you compress an organization from mountains of middle management into a lean structure: senior leaders driving strategy, AI Stewards tactically commanding orchestrator agents, and those orchestrators directing armies of sub-agents whose output finally translates into outcomes that fulfill your strategy.
The future is not without humans. It is led by them, inspired by them, and driven by them. You should no more hand an agent command of your business than a farmer would hand leadership of the farm to his tractor.
The work of becoming AI-ready is not glamorous—clean data, honest culture, enforceable governance, a transformed workforce. But it is the only path that turns AI from a liability you finance into an advantage you compound. The debt is optional. So is the win. Choose deliberately. | WA
Evan J. Schwartz is Chief Innovation Officer at AMCS Group, driving AI strategy across 80 countries in resource-intensive industries. He teaches at Jacksonville University and created The Customer Journey Framework, which draws on 35+ years of experience deploying enterprise-level digital solutions. His book “People, Places, and Things: A Framework for a Pain-Free ERP Implementation” became an Amazon bestseller. For more information, visit .
Author’s Note: The estimate that approximately 30 percent of top producers are lost during AI transformation when leadership fails to support their shift from output to outcome reflects the author’s direct observations across enterprise AI implementations and is offered as a practitioner’s data point rather than a published statistic. The 12/88 percent characterization of software total cost of ownership is the author’s framing of a well-documented industry pattern in which post-deployment maintenance and support dominate lifecycle cost (see Vention, 2024).
References
European Commission. (2024). Regulation (EU) 2024/1689 of the European Parliament and of the Council laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). EUR-Lex. https://eur-lex.europa.eu/eli/reg/2024/1689/oj
Frontiers in Artificial Intelligence. (2025). Survey and analysis of hallucinations in large language models: Attribution to prompting strategies or model behavior. Frontiers in Artificial Intelligence, 8, Article 1622292. www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2025.1622292/full
Gallo, A. (2014, October 29). The value of keeping the right customers. Harvard Business Review. https://hbr.org/2014/10/the-value-of-keeping-the-right-customers
Gartner. (2023). Reskilling and upskilling the workforce. Gartner for HR. www.gartner.com/en/human-resources/trends/reskilling-the-workforce-infographic
Masterson, V. (2025). Klarna reverses course on AI customer support and resumes human hiring. CX Today. www.cxtoday.com/contact-center/klarna-redeploys-staff-to-customer-service-pulls-from-marketing-engineering-and-legal-teams/
Mauran, C. (2025, August 18). MIT report: 95% of generative AI pilots at companies are failing. Fortune. https://fortune.com/2025/08/18/mit-report-95-percent-generative-ai-pilots-at-companies-failing-cfo/
Microsoft. (2025). Best practices for mitigating hallucinations in large language models (LLMs). Microsoft Azure AI Foundry Blog. https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/best-practices-for-mitigating-hallucinations-in-large-language-models-llms/4403129
MIT Project NANDA. (2025). The GenAI divide: State of AI in business 2025. MIT Media Lab. https://mitsmr.com/ai
Nalpeiron. (2024). How proactive customer success can curb churn. https://nalpeiron.com/blog/how-proactive-customer-success-can-curb-churn
National Institute of Standards and Technology. (2023). Artificial intelligence risk management framework (AI RMF 1.0) (NIST AI 100-1). U.S. Department of Commerce. https://doi.org/10.6028/NIST.AI.100-1
Vention. (2024). Software maintenance costs: 2024 benchmark overview. https://ventionteams.com/enterprise/software-maintenance-costs
World Economic Forum. (2025). The future of jobs report 2025. https://www.weforum.org/publications/the-future-of-jobs-report-2025/