The cost of running a large language model has dropped roughly 90% since early 2025, according to a16z's latest analysis of AI infrastructure spending. That single data point explains more about where small business is heading than any keynote or product launch.
Eighteen months ago, building meaningful AI automation required serious upfront investment — custom integrations, expensive API calls, and specialist consultants. That's no longer the case. And the businesses that are moving now are opening a gap that will be very difficult to close.
The Numbers Are In
McKinsey's 2025 State of AI report found that 72% of organisations have adopted AI in at least one business function, up from 55% the year before. More telling: companies in the top quartile of AI adoption saw revenue growth 3.5x higher than those in the bottom quartile.
This isn't just a big-enterprise story. The Australian Small Business and Family Enterprise Ombudsman's 2025 report noted that small businesses using AI tools reported an average 23% reduction in administrative time — roughly one full day per week freed up for revenue-generating work.
The denominator is shifting. The question isn't "can we afford to adopt AI?" It's "can we afford not to?"
What's Actually Working
Let's cut through the noise. Based on what we're seeing at Heygentic — building automation systems for businesses ranging from real estate agencies to luxury resorts — these are the use cases genuinely delivering value right now:
Intelligent inbox triage. Google's Workspace AI now handles email classification and priority routing natively. For businesses on Microsoft, Copilot for Outlook does similar work. But the real value comes from custom implementations — AI agents that understand your specific workflow, extract action items, and route to the right team member. I've seen this save 5-10 hours per week of admin time for service businesses.
Automated proposal generation. Tools like Jasper and custom GPT-based workflows can take a brief or RFP and produce a first-draft proposal that's 80% there. The human reviews, adjusts positioning, and sends. An afternoon's work becomes thirty minutes.
Customer support augmentation. Not replacing support teams — augmenting them. Intercom's Fin now resolves up to 50% of support queries without human involvement, according to Intercom's published benchmarks. Zendesk's AI agents report similar numbers. The key: AI handles the routine, humans handle the complex, and both get full context.
Data reconciliation. The unglamorous work that eats hours. Invoice processing, CRM updates, inventory reconciliation. Zapier's AI features and Make.com have made this accessible without custom code. What required a developer eighteen months ago now requires a thirty-minute configuration.
The Australian Context
Australia's labour market makes this especially urgent. The ABS's latest labour force data shows unemployment sitting near historic lows at 3.7%, with persistent skills shortages across over 300 occupations according to Jobs and Skills Australia.
We're a high-wage economy where finding and retaining good people is expensive and getting harder. The average wage growth is running at 4.1% annually, well above the long-term average.
Automation doesn't mean fewer jobs in this context — it means your existing team handles more. The bookkeeper who spent three days on month-end reconciliation now spends half a day. The office manager who triaged fifty emails before lunch now handles the ten that actually need human judgment.
For Australian businesses competing globally, this is how you close the productivity gap without burning out your people.
What's Coming Next
Two shifts worth watching over the next twelve months:
First, multi-agent systems are moving from research to production. OpenAI's Swarm framework, Anthropic's tool-use architecture, and LangChain's LangGraph are all making it practical to orchestrate multiple AI agents on a single workflow. Instead of one AI doing one task, coordinated agents will handle end-to-end processes — monitoring inboxes, updating project management tools, and drafting client communications in concert.
Second, fine-tuned models trained on your business data are becoming affordable. OpenAI's fine-tuning API, Anthropic's custom model program, and open-weight models like Llama 3 running on commodity hardware mean a 50-person company can now have an AI that knows its processes, tone, and institutional knowledge. Not a generic chatbot — a junior team member who's read every document you've ever produced.
The Bottom Line
The businesses that figure this out now will have a compounding advantage. Every month of automated operations is a month of refined processes, accumulated data, and freed-up human capacity that competitors don't have.
The gap is widening. The tools are accessible. The ROI is proven. The only remaining question is timing.
Sources
- The State of AI in 2025 — McKinsey & Company
- AI Infrastructure Spending Analysis — Andreessen Horowitz
- Small Business AI Adoption Report — Australian Small Business Ombudsman
- Labour Force Statistics — Australian Bureau of Statistics
- Skills Shortage List — Jobs and Skills Australia
- Fin AI Agent Benchmarks — Intercom