On 5 October, Deutsche Telekom used its AI Investor Day to set an ambition: roughly €2.5 billion in gross indirect-cost savings outside the United States by around 2030, compared with 2023. This combines AI and automation, excludes model-usage costs, and remains a company target. It establishes neither realised profit nor a number of jobs that will disappear.

Yet the announcement brings an important question into focus. When a company needs less time to serve us, whose life gets better? My argument is that every AI savings account needs a companion account: what happened to customers' time, workers' prospects and the quality of the service?

Consider an ordinary billing dispute. A useful assistant finds the mistake, explains it clearly and arranges a correction. The customer escapes a tedious call; an employee escapes repetitive searching. Everyone can benefit.

But an automated system might instead end the conversation without resolving the dispute. The company records a cheaper interaction while the customer repeats the story elsewhere. Someone has saved labour by making another person perform it. That is an illustrative risk, not a finding about Telekom's service.

The distinction matters most for people with the least spare time or bargaining power. Someone using a second language, struggling to read a bill or relying on an unreliable connection may need several attempts to be understood. A fluent answer provides little agency if there is no usable route to challenge it.

There is a strong case for adopting AI. A study of 5,172 customer-support agents, in its revised 2024 version, found that assistance increased issues resolved per hour by 15% on average. Less experienced workers benefited substantially. This was evidence from one firm's assisted workforce, not a promise that every business can replace its staff.

That result suggests a valuable possibility: expertise can become easier to acquire. A new employee could spend less time searching manuals and more time understanding a customer's problem. The gain might be confidence, better judgment and a less frustrating working day.

It could also become a higher quota. If software handles routine cases, the remaining employees may face a continuous queue of difficult conversations. Preserving headcount would not, by itself, establish that work had improved. Nor would completing a training course establish that someone had a credible next job.

The ILO's 2025 assessment distinguishes exposure of tasks from elimination of whole occupations. Its analysis points towards substantial transformation because human input remains necessary. Exposure is not observed unemployment. What employers do with the hours released remains a decision.

Customers might receive lower prices. Workers might receive higher pay, paid learning or shorter hours. Businesses might invest more, or retain larger margins. Technology suppliers might take recurring fees. Competition, contracts and bargaining power influence that division; the tool itself promises none of it.

Telekom's financial presentation says some savings will support digital transformation and a different mix of fibre deployment in Germany. That is a plausible public benefit, though the intention still needs delivery. Its exclusion of model-usage costs also supplies a useful discipline: gross efficiency is only the opening calculation.

For India, the bargain crosses borders. The Economic Survey 2025–26 warns that AI could weaken the IT sector's established advantages if adaptation lags. That is a policy warning, not evidence that this announcement will cancel Indian contracts.

The mechanism is nevertheless clear. Where a service provider earns by supplying working hours, reducing those hours can shrink its business even when the client's productivity rises. A provider paid for reliably resolving problems has a different opportunity. Moving towards that model requires domain knowledge, responsibility for errors and employees with authority to handle exceptions. Buying software alone cannot supply those capabilities.

Three futures follow. In a better one, assistance improves service, savings fund investment and paid transitions, and people retain accessible human help. In an uneven one, routine requests become easier while complex cases wait longer, with personal attention increasingly sold at a premium. In a harmful one, closure statistics conceal unresolved problems and employees carry responsibility without discretion. These are conditional pathways, not forecasts.

Does your saving survive the cost of integration, checking and redress? Are customers actually getting a remedy, or simply making fewer successful contacts? Which workers gain skills and opportunities, and which inherit the hardest work?

Publish repeat-contact rates, resolution quality and customer time alongside cost reductions. Test those results across languages and service channels. Give employees paid learning time and track where they move afterwards. Preserve a reachable person for disputed outcomes.

AI can make service cheaper and life easier. The meaningful dividend is time people can use, with problems resolved and choices intact. Count that before declaring the saving a success.