What else in tech will burst like the Metaverse and NFTs did?

The metaverse and NFTs became convenient symbols of excessive technology hype. But it would be a mistake to conclude that every new technology is a bubble. A more useful management question is: what measurable change does this technology create in our process, for the customer or in our risk profile?

Four reality checks

1. The behaviour test. Does a customer or employee actually change what they do, or do they only try the demonstration once?

2. The economics test. Does the technology reduce process cost, shorten cycle time, improve quality or increase revenue? Include licences, integration, training and maintenance in the calculation.

3. The data test. Do you have the right to use the data, is its quality sufficient and can the output be verified?

4. The risk test. Who is accountable for an error? Can a person stop the system, trace a decision and correct the damage?

The NIST Generative AI risk profile helps organizations identify risks specific to generative AI and suitable management actions: NIST AI RMF Generative AI Profile. In Europe, the EU AI Act provides additional context. It entered into force on 1 August 2024, while most provisions apply from 2 August 2026 with exceptions: European Commission — AI Act.

Where hype risk is highest today

Promises of fully autonomous marketing. AI can generate variants and analyse data, but it does not assume responsibility for positioning, rights, factual accuracy or brand risk.

“Magic” tracking. Server-side tracking can improve control and resilience of signals, but it does not capture 100% of reality and does not remove consent requirements. Avoid universal claims of a 20–30% ROAS uplift without a defined test.

Content without editorial work. Large-scale generation reduces unit cost, but it does not prove usefulness. Google warns that publishing many pages without added value can violate its scaled content abuse policy: Guidance on generative AI content.

Where value is more likely

The most practical projects are often narrow: call summaries with human approval, proposal drafts built from verified data, customer-request classification, product-data quality control or anomaly detection. Such a process is likely to create value only when it has a baseline metric, a clear owner and an error-review workflow.

Start with a one-page process card, not a licence purchase: current time and cost, desired result, acceptable error, data source, accountable person and a 30-day test. Stop the experiment if the result does not improve. Scale it only when the evidence is positive.