AI bubble: invest in useful applications without getting locked into one provider
AI bubble: invest in useful applications without getting locked into one provider
Your team can save time with AI while you question the prices investors are paying to fund the sector. Both views can be right. For your business, the useful decision is to fund a use case that delivers a measurable service while keeping the ability to change providers.
The “AI bubble” debate often bundles together three questions: company valuations, the resilience of your chosen provider and the value of the work done for your business. Separating them lets you invest without tying your organisation’s future to one platform.
Valuation, provider resilience and business value are different decisions
A valuation reflects the price investors are willing to pay to fund or buy a company, along with their expectations for growth. It does not measure the time your sales team saves on a proposal. Equally, a tool that helps your team does not establish that its provider is worth the price the market assigns to it.
Provider resilience raises a different question: can the supplier maintain the service you rely on, on terms your business needs? Examine contractual commitments, service continuity, technical dependencies and, where your exposure warrants it, public financial information. The greater the impact of losing the service, the deeper that review should be; for a convenience tool, a replacement procedure may be enough.
Business value is tested in your own work: a request correctly handled, a usable document, a manual entry avoided. You can continue a cost-effective use case without taking a position on the sector’s stock-market future. You can also reject a subscription from a financially strong company if the work delivered is not worth the expense.
Start with the cost of a task actually completed
A subscription price or a rate per volume of text does not tell you what a completed case costs. To compare two solutions, use the same representative tasks and define an acceptable result before testing.
Cost per completed task divides spending within the tested scope by the number of accepted results. Include model calls, supporting tools, retries, human review and an appropriate share of integration and maintenance. Keep the allocation method consistent across options.
Imagine a sales team preparing replies to incoming enquiries. One solution drafts quickly but regularly misses a contractual condition; another needs fewer corrections. In this example, the lowest generation price can lead to the higher overall cost once reviews are included. Compare turnaround time, errors and team workload with the process without AI, too.
This measure complements the budget for an SME AI agent: it helps you decide whether the service deserves to grow, not just whether you can afford to set it up.
Keep your data and business context reusable
Switching providers becomes expensive when your procedures, corrections and decisions exist only in the provider’s conversations. A file backup is not enough if nobody knows which version is authoritative or why an exception was approved.
Keep source documents, approved rules, useful history and acceptance criteria in spaces your business controls. Check retrieval rights and available formats. The connections between a case, a decision and its owner matter as much as the exported text. Our Company Brain guide to business memory explains how to organise that context.
The European Commission identifies barriers to switching data processing services—including exit charges, procedures and lack of interoperability—among the issues addressed by the Data Act. Contractual terms matter; they do not replace a retrieval and reuse test with your own data. An available export is not yet a successful migration.
Test a switch before you need one
Dependency is not just about possible financial trouble. OpenAI documents API model retirements and recommended replacements: even a business staying with the same provider may need to adapt its software. Include a switching test in the life cycle of your use case.
Take a few familiar tasks, retrieve their data and run them with a different configuration. Check quality, access permissions, output formats, turnaround time and cost. You can preserve instructions and expected outcomes; you still need to test how the replacement model behaves.
For an AI Worker, also define what stops during the transition: reading, drafting, editing the CRM or sending a customer message. Replacing the model must not trigger an action that has already been carried out.
Match your fallback to the impact of an interruption
An assistant helping to prepare a publication may be adequately covered by a manual process. A workflow handling customer requests throughout the day needs a more explicit continuity plan. Keeping two providers active at all times is not necessary for every use case: it adds cost and review work.
- Occasional use: retain the source material and a usable manual procedure.
- Daily use: document exports, test an alternative and assign responsibility for the switching decision.
- Critical process: define acceptable downtime, reduced-service arrangements and restart conditions; test them with business owners.
Before expanding your investment, you should be able to answer four questions: which task justifies the expense? What can we retrieve when leaving the service? Who approves the replacement? How do we work during an interruption? Concrete answers are more useful than certainty about the market’s next phase.
Invest in a use case you can keep improving
You do not need to settle the argument between optimists and pessimists before starting. Choose a limited process, measure the result and keep your context reusable. Increase the investment when the benefit holds up after corrections, running costs and continuity requirements.
SAY Go Big Online can help you scope that first use case, compare solutions on your tasks and define the data, approvals and exit plan you should retain. The aim is a tool that helps your team today and can be improved or replaced tomorrow. Let’s discuss your first measurable AI use case.
