Most conversations about AI and sustainability talk past each other. One side says AI will solve the climate crisis. The other says it’s accelerating it. Madalina Buzdugan isn’t on either side. She’s asking the question most people skip: what does sustainable AI for small business actually look like in practice, starting this week, with the tools you already have?
Madalina spent 15 years in startups and scale-ups, building corporate sustainability programs from scratch and learning what it means to make the business case for something people haven’t prioritized yet. When generative AI hit the market, she recognized the dynamic immediately: fast adoption, thin structure, and a conversation that dumped accountability on individual users instead of the systems and companies driving the growth.
That’s the gap she works in now. Not telling businesses to stop using AI. Helping them slow down long enough to use it on purpose.
Why sustainable AI for small business starts with your values, not your tools
Madalina’s first move with most clients isn’t a tool audit. It’s a conversation about what they actually care about.
If your organization has a value like “treat customers as friends,” that value already tells you something important about AI. It tells you that fully automating your customer support flow probably doesn’t fit. You don’t need a policy document to tell you that. You need to know your values clearly enough that the answer is obvious.
That sounds simple, but most businesses skip it. They buy licenses, distribute access, and ask people to produce more output, whatever that output is. No training, no policy, no conversation about what matters. When the questions come back from employees, and they will, nobody knows how to answer them because nobody defined success before the rollout started.
That’s not an AI problem. It’s a change management problem that AI is making visible.
Madalina frames the core question this way: “What is it that you’re trying to adopt? What does success look like? And what type of impact are you comfortable with?” Three questions. Most companies haven’t answered any of them.
What the environmental cost of AI actually looks like (and what we still don’t know)
The numbers floating around about AI energy and water consumption are mostly unverifiable. Madalina is clear about this. The major AI companies have not published the data needed to confirm the figures that circulate online, and in some cases have actively lobbied against disclosure requirements.
That doesn’t mean the impact is zero. It means the conversation about individual guilt, “I’m a bad person because I used Claude today,” is the wrong frame. That’s the same dynamic that played out with carbon footprints, where a BP marketing campaign successfully shifted accountability from large companies to individual consumers. The lens belongs on company-level decisions, data center transparency, and regulation, not on whether you should feel bad for using AI to write a meeting agenda.
What individuals and small businesses can do is stay informed, ask suppliers better questions, and make intentional choices about which tools they use and when. That’s not nothing.
Two small things Madalina recommends: a browser extension called AI Impact Tracker (works primarily with ChatGPT) and AI Watch, which tracks energy use and converts it to everyday comparisons so teams can have a real conversation about impact. Neither one exists to create guilt. Both exist to create information.
How to use less without doing less
This is where Madalina gets practical, and where the conversation usually turns for the businesses she works with.
The first thing she recommends is reconsidering which model you’re using. Most everyday tasks don’t require a large language model. Small language models like Green PT, Viro AI, Taura, Smol, or Gemma will handle the majority of daily use cases with less computing power. It’s not a sacrifice in quality for most tasks. It’s a more deliberate choice.
The second thing she talks about is batch prompting: instead of opening your AI tool and prompting it ten separate times for ten different tasks, take a few minutes to consolidate. Write one prompt that addresses all ten. The model fires up once instead of ten times. Less computing load, and often better output because you’ve had to think through what you actually need before you ask.
The third thing, and the one most teams overlook, is auditing which AI features are already running inside the tools you own. Many teams are running three or four note-taking tools simultaneously across the same meetings, creating five different transcripts of the same conversation and massive cognitive overhead on the back end. Before buying another AI subscription, check what’s already in your Google Workspace, your Zoom account, or the software you’re already paying for.
What the pushback on AI adoption is actually telling you
Roughly 29% of employees are actively working against AI adoption at their organizations. Another 44% of Gen Z workers are doing the same. Madalina doesn’t read these numbers as evidence that people are difficult or resistant to change. She reads them as evidence that most rollouts skipped the conversation.
People pushing back on AI are often asking reasonable questions that nobody has answered: Where is my data going? Will this replace me? Why are we doing this? What does success look like? If those questions don’t have answers, the pushback is the logical response.
The fix isn’t faster onboarding or better demos. It’s the conversation that should have happened before the licenses were purchased. What do you care about? What does this tool need to do for your team? What’s off-limits?
That conversation is also what makes an AI policy manageable to write. When you start from values, the policy writes itself.
Use This Today
Before you open any AI tool for your next task, ask yourself three questions:
- Who am I actually trying to reach or serve with this output?
- Do I already have something in place that could handle this?
- Do I really need a large language model, or is there a lighter option?
This takes about two minutes. It doesn’t add friction. It adds intention. And over time, it adds up to a very different relationship with AI in your business.
One more thing: if you don’t know what your AI footprint looks like, AI Impact Tracker or AI Watch can give you a starting point. Run it for a month, bring the data to your team, and have the conversation. The answer might be that your current usage is completely worth it. That’s a valid outcome. The point is to know, not to assume.