Most conversations about AI in business swing between two extremes. There’s the keynote version, where autonomous agents run entire departments by next quarter. Then there’s what most teams are actually doing: using ChatGPT to draft emails and wondering what comes next.
The reality sits between those two, and it’s moving fast. This page covers where AI in business stands in 2026, what’s delivering results, where things go wrong, and what the shift toward AI agents means for operators. The articles linked below go deeper on specific topics.
Where Does AI in Business Stand in 2026?
88% of companies now use AI regularly in at least one business function. That’s up from 55% in 2023 (McKinsey Global Survey, 2025). The adoption curve is steep. But adoption and value are different things.
Only 6% of companies qualify as “AI high performers,” reporting real value and at least 5% of EBIT tied to AI. The other 82% are spending, testing, and not yet seeing bottom-line returns.
Global AI spending is forecast to hit $2.52 trillion in 2026, up 44% year-over-year (Gartner, 2026). Deloitte calls this the AI ROI paradox. 85% of companies increased their AI investment last year. 91% plan to increase again. Yet most don’t expect payback for two to four years (Deloitte, 2025).
The question has shifted. It’s no longer “should we adopt AI?” It’s “why aren’t we getting more from what we already adopted?”
What’s Actually Working?
The companies generating real returns share a few patterns. They started with specific problems, not a mandate to “use AI.” They tied AI to workflows with measurable outcomes. They let teams learn before scaling.
Content and creative tools
Most businesses start here. AI writing assistants, image generators, and video tools have matured enough to save genuine time. But the value compounds when these tools are part of a workflow, not used as standalone generators. A content brief that pulls competitive data and suggests structure automatically is more useful than a blank chat window.
Marketing automation
This category has delivered the fastest measurable ROI across the businesses I work with. Tools that connect data across channels, optimise ad spend, and automate creative testing produce results you can track in weeks. For e-commerce and D2C brands, this often pays for itself first. More on AI in marketing: Marketing & Growth.
Customer service
Another area where AI has crossed a threshold. Modern tools handle tier-one queries accurately, escalate complex issues with full context, and work across WhatsApp, web chat, and email simultaneously.
Where things break down
Any project that begins with “we need to use AI” and no clear picture of what problem it should solve. The pattern I see repeated is teams adopting five tools in three months, measuring nothing, and ending up with fragmented workflows and confused staff.
AI Agents: From Experiment to Infrastructure
AI agents can take actions autonomously. Instead of asking AI to “draft a follow-up email,” an agent monitors your pipeline, identifies stalled deals, drafts a personalised message based on account history, and sends it on a schedule.
62% of companies are testing agents, but only 23% have scaled them in even one function (McKinsey, 2025). Gartner predicts 40% of enterprise apps will have task-specific AI agents by 2026. That’s up from less than 5% in 2025 (Gartner, 2025).
Deloitte’s data tells a more grounded story. 30% of companies are exploring agentic AI, 38% piloting, 14% ready to deploy, and just 11% running it in production (Deloitte, 2025).
Commerce is one area where agents are moving from concept to infrastructure. Visa, Mastercard, Stripe, and Shopify are building frameworks for AI agents that browse, compare, and purchase on behalf of consumers. I wrote about where this is heading: The AI Commerce War: OpenAI and Google.
For most businesses today, the practical move is AI-assisted workflows where humans stay in the loop for high-stakes decisions. Fully autonomous agents handling complex processes without oversight remain early-stage.
Why Do Most AI Projects Fail?
AI projects fail at roughly 80%. That’s nearly double the rate of non-AI IT projects (RAND Corporation, 2024). A 2025 MIT study found that 95% of generative AI pilots produce no measurable P&L impact (MIT NANDA, 2025). Many of those pilots were exploratory by design. But the pattern is clear: most AI efforts stall before producing real returns.
The data problem
The biggest bottleneck is data. 63% of companies either lack or aren’t sure they have the right data practices for AI (Gartner, 2025). Gartner projects that through 2026, 60% of AI projects without AI-ready data will be dropped.
If customer data lives across your CRM, email tool, finance software, and three spreadsheets, no AI tool will give you useful insights. A one-time data cleanup sounds unglamorous but it decides whether your AI investments produce results or waste money.
The governance gap
The second most common failure is missing governance. You don’t need a 50-page policy. You need clear answers to four questions: who approves new AI tools? What data can and can’t be used? How are outputs checked before reaching customers? What happens when something goes wrong?
For more on governance and AI literacy at the board level: Every Director Needs AI Fluency. For governance frameworks: Leadership & Governance.
How Is AI Adoption Different in Asia Pacific?
APAC is ahead of the global average. 78% of workers in the region use AI weekly, compared to 72% worldwide. Among frontline staff, the gap is wider: 70% in APAC use generative AI regularly versus 51% globally (BCG, 2025).
The governance gap is the concern. 58% of APAC workers say they would use AI even without company approval. 35% would actively bypass restrictions. Shadow AI, where employees adopt tools the company hasn’t vetted, is the biggest unmanaged risk in the region right now.
Privacy rules add another layer. Singapore’s PDPA, Thailand’s PDPA, Malaysia’s PDPA, and Indonesia’s PDP Law each have distinct rules, all getting stricter. Cross-border data transfers become a real challenge when teams use AI tools hosted outside the region.
Practical steps: know where your data is processed, make sure your privacy policy covers AI data use, keep a register of every AI tool and the data it touches, and accept that your team is likely already using tools you haven’t approved. Build a policy that works with this reality, not against it.
What Should You Do First?
If you’re building your first AI workflows, five steps that work:
-
Pick one workflow. Find a repetitive, measurable task with clear inputs and outputs. Campaign reporting, lead scoring, email triage, content brief generation. Something you can track before and after.
-
Get your data in order. Centralise what matters. Clean up duplicates and inconsistent formatting. This step determines whether everything that follows works or wastes money.
-
Buy before you build. MIT research shows purchasing from specialised vendors succeeds about 67% of the time, while building internally succeeds roughly 22% (MIT NANDA, 2025). Most businesses should start with proven tools, not custom development.
-
Keep humans in the loop. AI outputs are polished enough to be dangerous: confident text with inaccuracies, impressive analyses built on flawed assumptions. Build review checkpoints into every workflow before anything reaches customers.
-
Expand from what works. Once your first workflow delivers, apply what you learned to the next one. AI strategy is a capability you build, not a one-time project.
I wrote about building AI tools hands-on: Building an AI Tool With No Code.
Frequently Asked Questions
What is the best way to start using AI in a business?
Pick one repetitive, measurable workflow like report generation, lead scoring, or email triage. Test free-tier AI tools on real problems before committing budget. The companies that get value from AI start with a specific problem and expand from there, rather than trying to roll out AI across the entire organisation at once.
How much does it cost to implement AI in a business?
Costs range from zero (free tiers of ChatGPT, Claude, Gemini) to millions for enterprise deployments. Most small and mid-sized businesses start at $50 to $500 per month in tool subscriptions. Start small, measure results, and scale spending only where you can prove returns.
What are AI agents and how are they different from regular AI tools?
Standard AI tools respond to prompts: you ask a question, you get an answer. AI agents take autonomous actions across multi-step workflows, like monitoring a sales pipeline, drafting personalised follow-ups, and scheduling them without manual input. 62% of organisations are experimenting with agents, but only 11% have them in production (McKinsey and Deloitte, 2025).
Why do most AI projects fail?
Data quality is the primary cause. 63% of organisations lack proper data management for AI (Gartner, 2025). Other common reasons: starting with technology instead of a defined business problem, no governance framework, implementing too many tools simultaneously, and letting AI output reach customers without human review.
Should I build custom AI solutions or buy off-the-shelf tools?
Buy first. MIT research shows purchasing from specialised vendors succeeds about 67% of the time, while building internally succeeds roughly 22%. Custom AI makes sense only when you have a unique data advantage or a workflow no existing tool can handle. Start with proven platforms and build capability gradually.
How is AI adoption different in Southeast Asia?
APAC leads globally: 78% of workers use AI weekly versus 72% worldwide (BCG, 2025). The risk is governance. 58% of APAC workers use AI without company approval, creating shadow AI exposure. Privacy regulations across Singapore, Thailand, Malaysia, and Indonesia each have distinct requirements for AI data handling.
What data do I need before implementing AI?
Clean, centralised data across your core systems: CRM, marketing tools, financial software. Duplicates, inconsistent naming, and scattered spreadsheets will poison AI outputs. You don't need big data. Even small, well-organised datasets power meaningful AI applications. A one-time data audit covering accuracy, completeness, and consistency is usually enough to start.
I build and advise businesses on practical AI implementation. If you’re evaluating AI tools or building an AI strategy, let’s talk.
