Businesses
Four Business Tasks Companies Should Not Fully Hand Over to AI Yet
09 Sept 2026
AI is becoming embedded across business workflows, but contracts, customer service, sales outreach and meeting follow-up still expose companies to legal, reputational and operational risks when human review disappears.
Businesses are moving quickly to automate more of their everyday work with artificial intelligence. The challenge is no longer whether AI can perform a task, but whether it can perform it reliably enough for companies to remove people from the process.
Recent research suggests that many organisations are still struggling to convert widespread AI use into consistent enterprise value. McKinsey reported in July 2026 that most companies remain early in their AI transformation, with individual productivity gains often failing to translate into broader organisational impact without redesigned workflows, governance and oversight.
George El-Hage, Founder and CEO of Wave Connect, argues that some business processes remain particularly risky to automate end-to-end. He identifies contract editing, customer calls, cold sales emails and automatic task creation from meeting notes as areas where AI can assist, but where human judgment should remain part of the workflow.
1. Editing Contracts Without Legal Review
Generative AI is increasingly used to review agreements, identify clauses and suggest redlines. That can shorten the first stage of contract review, but automatically accepting those changes creates a different level of risk.
El-Hage warns that AI systems can propose unnecessary rewrites, misinterpret the commercial importance of clauses or weaken protections while attempting to make language appear more balanced. Contract negotiations often depend on context, risk appetite, regulatory requirements and the relationship between the parties, factors that cannot always be inferred from the wording alone.
The wider legal environment reflects the same concern. The European Commission has established an Expert Group on AI Contracting to examine practical risks around automated contracting and develop guidance for businesses, while the EU AI Act explicitly recognises human oversight as an important control for higher-risk AI applications.
For companies, the safer model is therefore AI-assisted contract review rather than autonomous legal decision-making. AI can surface unusual clauses, compare language against internal playbooks and prepare suggested edits, while a qualified person remains responsible for the final decision.
2. Replacing Customer Service Calls With AI
Voice agents have improved rapidly, but customer tolerance remains a significant constraint.
A 2026 study by AnswerConnect involving 6,000 consumers across the US, UK and Canada found that 31% said they would hang up if connected to an AI agent, up from 29% in late 2025. The same research found that 85% preferred speaking to a real person, while 59% reported frustration with AI customer-service systems.
Other research shows that a poor AI experience can affect the wider customer relationship. Clutch found that 67% of surveyed consumers had considered leaving or had stopped doing business with a company following a poor AI-support experience, while 81% felt AI systems had at some point prevented them from reaching a human agent.
This does not mean companies should abandon voice AI. A better approach is to use automation for predictable queries while giving customers a clear and immediate path to a person when the conversation becomes complex, sensitive or frustrating.
3. Letting AI Write Cold Sales Emails End-to-End
Sales outreach is another area where greater automation does not necessarily produce better outcomes.
Wave Connect highlights data from a six-month test covering approximately 74,000 emails across nine client accounts. Human-written campaigns achieved a 3.4% positive reply rate, compared with 2.1% for AI-written emails, a decline of roughly 38%. The test kept prospect lists, infrastructure and sending conditions comparable while changing whether the copy was produced by a person or AI.
El-Hage argues that prospects are increasingly familiar with the patterns associated with automated outreach.
“Many businesses believe that AI sending emails out will bring more clients, because it’s easier to reach people,” he said. “But the common practice now is just to delete an email if it reads like AI.”
That does not make AI useless in sales. It can research prospects, identify relevant signals, structure drafts and generate variations quickly. The problem emerges when every recipient receives language produced from similar prompts, structures and templates.
Human editing remains valuable because relevance and credibility often matter more than volume.
4. Automatically Turning Meeting Notes Into Tasks
AI meeting assistants can now record conversations, create summaries and generate action items within seconds. Automating the next step — automatically assigning those actions to employees — creates additional risk.
A conversation does not always distinguish clearly between an idea, a suggestion and an agreed responsibility. Without enough organisational context, an AI system can turn speculative comments into tasks, assign work to the wrong person or miss conditions attached to a decision.
There is also a data-governance issue. AI notetakers can capture confidential discussions involving employees, finances, legal matters, intellectual property or corporate strategy.
The Associated Press reported in July 2026 that meeting assistants can turn sensitive workplace conversations into searchable datasets, with privacy specialists advising businesses to understand how vendors store transcripts, recordings and metadata. Legal specialists have similarly warned that these tools can create risks involving privilege, cybersecurity and regulatory compliance, particularly when used during sensitive meetings.
A more controlled workflow is to allow AI to suggest action items before participants review and approve them.
The Better Business Model Is Human-AI Collaboration
The four examples point to the same operating principle.
Businesses do not necessarily need to choose between human work and automation. In many cases, the stronger model is to place AI earlier in the workflow while keeping people at the points where judgment, accountability and context matter most.
| Business task | Useful role for AI | Human role still needed |
|---|---|---|
| Contract review | Identify clauses, compare language, suggest edits | Assess legal and commercial consequences |
| Customer service | Handle routine questions and initial triage | Resolve complex or sensitive cases |
| Sales outreach | Research prospects and produce drafts | Personalise messaging and protect brand voice |
| Meeting follow-up | Summarise discussion and suggest actions | Confirm decisions, ownership and sensitive information |
The cost of automation should therefore be measured against more than time saved. Customer trust, legal exposure, security, sales performance and error correction also belong in the calculation.
For business leaders, the important question in 2026 is increasingly not “Can AI do this?” but “Where should the human checkpoint remain?”
About Wave Connect
Wave Connect is an event lead-capture and networking technology company. Its products include digital business cards, NFC tools and AI-powered badge scanning designed to help businesses collect and manage leads from events. Founder and CEO George El-Hage provides commentary on business automation and the practical use of AI across sales and operational workflows.
Data Sources
- McKinsey & Company 2026 — From adoption to impact: Three horizons of AI transformation: AI adoption is expanding, but many organisations remain early in converting individual use into enterprise-wide value.
- European Commission — Automated contracting and AI: outlines emerging legal questions and the Commission’s work on AI contracting guidance.
- EU AI Act — Article 14, Human Oversight: establishes human-oversight requirements for high-risk AI systems.
- AnswerConnect, 2026 AI Customer Service Attitudes Report: survey of 6,000 consumers found 31% would hang up on an AI agent and 85% preferred a human.
- Clutch, 2026 AI Customer Support Report: 67% of consumers had considered or stopped doing business with a company following a poor AI-support experience.
- 74,000-email human vs AI outreach test: human-written emails recorded a 3.4% positive reply rate versus 2.1% for AI copy.
- The Associated Press, July 2026 — AI meeting notetakers: examines confidentiality, data retention and privacy risks around automated meeting assistants.
- Polsinelli, 2026 — Hidden Risks of AI Notetakers: covers potential legal, privacy, security and compliance exposure.
Share
Sara Srifi
Sara is a Software Engineering and Business student with a passion for astronomy, cultural studies, and human-centered storytelling. She explores the quiet intersections between science, identity, and imagination, reflecting on how space, art, and society shape the way we understand ourselves and the world around us. Her writing draws on curiosity and lived experience to bridge disciplines and spark dialogue across cultures.
