The venture capital firm Andreessen Horowitz, usually abbreviated to a16z in the industry, is for the first time recording actual consumer spending in the US for its top 100 list of AI applications. The result shifts the picture of the mass market: usage and willingness to pay are far apart.
According to the data, almost half of US consumers use AI offerings. But only 4.5 per cent pay for a subscription. The rest remain on free versions or use AI where it is already embedded in other products.
The picture becomes even clearer at the top: the top one per cent of payers spend an average of 900 dollars per month. This money flows primarily into professional tools for developing and automating – that is, into software that writes or checks code or performs recurring work steps without manual intervention.
Broad usage, narrow monetisation
The figures initially describe the US consumer market and cannot be transferred one-to-one to companies in the German-speaking region. The pattern is nevertheless revealing because it corrects a common assumption: high usage figures are not proof that an AI product creates affordable value for users.
A chatbot that occasionally formulates an email or answers a question is convenient. But the benefit is rarely so clearly measurable that people are prepared to pay for it regularly. It is different with tools that intervene in a specific workflow and save time or error costs there – there, a willingness to pay apparently emerges that also supports three-digit monthly amounts.
Where AI creates measurable value today
The fact that developer and automation tools in particular attract the highest spending has structural reasons. In software development, the conditions for AI are particularly favourable:
- Clear tasks: Writing a function, adding tests, narrowing down a bug – these are clearly delimited work packages with a recognisable result.
- Fast verifiability: Code can be compiled, tested and reviewed. A wrong suggestion usually becomes apparent early.
- High repetition: Much in development and operations is repetitive and therefore easy to automate.
- Expensive working time: Where specialist time is scarce, a tool pays for itself even with moderate savings.
The same criteria serve as a test grid for other areas. Processes with clear input, a defined result and a way to check them are much better suited to AI support than diffuse, highly context-dependent tasks.
Consequences for tool budgets
Many companies currently spread their AI funds widely: a subscription here, a pilot project there, plus licences that are barely used after the initial enthusiasm. The data suggest a different prioritisation – less scattering, but targeted provision where the leverage is greatest.
A pragmatic approach for budget planning:
- Record usage: Which AI licences exist, who actually works with them regularly? There is often a gap between purchased and actively used seats.
- Identify bottlenecks: Where do waiting times, rework or manual transfers between systems arise? There is usually more potential there than in general text work.
- Equip teams instead of distributing evenly: Development, support and operational process teams tend to benefit more than areas with mainly advisory work.
- Measure results: Compare throughput times, error rates or ticket volumes before and after introduction – otherwise the benefit remains a claim.
- Adjust: Consistently cancel tools without demonstrable impact and reallocate the funds.
What the figures do not say
The survey measures consumer spending, not productivity. The fact that a small group pays high amounts initially proves only a strong willingness to pay for professional applications. It is not proof that each of these expenses also pays off – and certainly not a guarantee that the same tools will have a comparable effect in another company.
Likewise, the low subscription rate does not imply that AI is unimportant in everyday life. A considerable part of usage runs through free offerings or through functions already integrated into existing software. For companies, this means: the question is not only "which AI subscription", but also which AI functions remain unused in already licensed systems.
From tool to process
The real difference between occasional use and measurable benefit rarely lies in the model, but in the embedding. A tool only unfolds its effect once it is connected to the existing data, systems and approval paths – including the question of who checks results and who is accountable in case of doubt.
This applies to development environments as well as to automation in content management, in editorial work or in order processing. Without this integration, AI remains an additional window alongside the actual work, and it is precisely such additional windows that tend to be cancelled first.
For web and software projects, this means: AI budgets should land where tasks are clearly delimited, results verifiable and processes recurring – typically in development, operations and process automation. Instead of distributing many licences widely, it is worth integrating a few tools cleanly into existing systems and approval processes and measuring their impact against concrete metrics.
Those who proceed this way recognise early which investment pays off – and which only costs attention.
Sources
- Only a few users pay for AI, but those who do pay a lot — The Decoder