← All insights ←

Which working hours will clients still pay for?

By · Published: 5 October 2026

StrategyFuture of WorkBoard of DirectorsAI Adoption
Which working hours will clients still pay for?

At the end of September, the McKinsey Global Institute published a figure that will soon appear in many presentations. By 2035, around 54 per cent of today’s working hours in the United States could be automated. Such forecasts should be treated with caution; nine years is a long time. Whether the final figure is 40, 50 or 60 per cent is of secondary importance, because the change ahead will be large.

Many read this as a question about the labour market. For me, that reading falls short. I also read it as a question about the business model of service firms, and therefore as a question for boards and executive teams. If more than half of all working hours can be automated, a business model that sells its services by the hour comes under pressure.

In October 2025, the KOF Swiss Economic Institute at ETH Zurich showed that the first effects are already measurable in Switzerland. Since the launch of ChatGPT, the number of unemployed jobseekers in occupations strongly exposed to generative AI has risen by up to 27 per cent more than in occupations with little exposure. The effect was most pronounced among younger jobseekers. By 2024/25, job advertisements in these fields had fallen to between 60 and 70 per cent of their earlier level.

The figure behind the figure

A second figure from the same McKinsey study is more revealing. Of every 100 hours worked today, around 54 could be automated by 2035. According to McKinsey’s calculation, however, only about 21 hours would actually disappear. The remaining 33 hours would be absorbed. The study gives four reasons for this. Automation reduces overtime. Lower costs usually increase demand. New tasks arise, such as reviewing AI output. Regulation and contracts slow the reduction down.

At first glance this sounds reassuring. For an individual company, however, the average is of little help, because the study models an entire economy, that of the United States. Whether a company gains or loses depends on how rigorously it questions its lines of business and the future behaviour of its clients. From this follows another question: how will services be priced in future? Firms that sell hours of work that machines take over will lose revenue, even if they serve the same number of clients or more. Companies that are willing to question their lines of business and their pricing model therefore have a better chance of emerging as winners from this change.

Four criteria for price pressure

In September, in an article for the Swiss Board Forum, I described four criteria under which a sector quickly comes under pressure:

  1. The service consists mainly of processing information.
  2. The client will be able to produce the result or carry out the task on their own.
  3. The service is billed by the hour.
  4. The activity is not protected by an official licence, a liability regime or a signature requirement.

According to McKinsey, office and administrative work is the most heavily affected. For boards and executive teams, a different question matters more: which companies sell precisely these hours as their product? They include back-office providers, parts of the accounting and bookkeeping sector, translation agencies, and parts of the consulting and legal sectors.

The billable hour as a weak point

The classic hourly model has an inherent weakness. If the work is done more efficiently, revenue falls. As long as efficiency gains came slowly, this hardly mattered. I expect the pace to increase sharply with AI over the next 36 months.

Little of this is visible at present. In a survey conducted by the Association of Corporate Counsel and Everlaw in the summer of 2025, almost 60 per cent of in-house counsel had not yet seen any noticeable savings from the use of AI by their external law firms. More than 60 per cent considered it too early to say. At the same time, some companies are now writing into their outside-counsel guidelines that simple tasks should be handled with AI. The time saved should show up on the invoice.

The audit sector itself shows what this looks like in practice. In 2024, KPMG International asked its own auditor, Grant Thornton, to pass on the efficiency gains from AI. According to the Financial Times, the fee fell by around 14 per cent (measured in US dollars). The case makes my fourth criterion more precise. Statutory exclusivity protects against clients providing the service themselves; it does not protect against price pressure.

In June 2025, PwC confirmed that it had lowered the prices of individual services. According to Dan Priest, Chief AI Officer of PwC, clients had asked for their fair share of the AI efficiencies. The discussion has reached Switzerland as well. In April 2026, Bilanz reported on the same development at Swiss law firms under the title ‘KI & Stundenabrechnung: Die Anwaltskanzleien geraten unter Druck’ (AI and hourly billing: law firms come under pressure).

This is exactly the dynamic I described in my Swiss Board Forum article. Once clients establish, through their own experience or by observing the market, that the work has become cheaper, price pressure rises. It arrives with a delay. Those who mistake this delay for an all-clear lose the time they need to rebuild.

When clients use AI themselves

The second criterion hits harder still. Chegg, a US provider of study aids for students, reported a 30 per cent drop in revenue for the first quarter of 2025 and cut 22 per cent of its jobs. The CEO cited Google’s AI summaries and free ChatGPT subscriptions for students. Some clients simply stopped buying the service.

A price reduction is of little help in this case. What is needed is a different service.

The robot in your own business

In February 2026 I wrote: ‘If you work like a robot, you will be replaced by a robot.’ At that time I was writing about internal processes: manual workarounds, Excel as an interface, copy and paste between systems, and documents that are printed and scanned again.

For service providers, this pattern has a second side. Part of the hours clients are billed for pays for precisely these workarounds, and often for inefficiencies as well. They pay for switching between paper and digital formats, rounds of coordination and duplicate data entry at the provider. As long as nobody sees this and questions it, it is revenue. As soon as clients can compare, either by using AI themselves or through a provider that has already made the move, it becomes a risk.

An example from my own practice: until recently, my accountant recorded and posted accounts payable and receivable. Every month, the firm received the invoices, the credit card receipts and the bank statements. Posting was purely manual work, and most entries were repetitive. Today I use a workflow engine I built myself with a locally run AI model, without a cloud-based language model. It collects the receipts from the various channels, identifies and analyses them, and reconciles them with the bank and credit card statements. It shows me which receipts are missing and proposes the journal entries for standard transactions. I am left with missing receipts and exceptions. My accountant imports the transactions and checks them. I no longer have to search for receipts every month, and my accountant no longer has to record anything by hand.

What boards and executive teams should clarify now

The four criteria work well as a simple test for your own lines of business. What share of revenue meets all four criteria? Which part of it is protected by liability, by an official licence or by statutory exclusivity, and for how much longer? What do we sell once the hour loses its importance as a unit of value?

In my experience, the answers lead to a small number of levers. The first is rebundling the service. The second consists of activities that start where a system reaches its limits and a person has to check, supervise or assess the result. This is what I call Human in Command. It includes detecting and correcting systematic errors, personal responsibility for the result and judgement in borderline cases. Only then does the question of the pricing model arise. Hours keep their place, for example where the scope is open-ended. Alongside them come fixed fees or outcome-based prices. Those who merely replace hours with fixed fees and leave the service unchanged buy time, but not a future.

Over the coming months, studies of this kind will be discussed mainly from the perspective of the labour market. I read them as a question about our own business model: how much revenue depends on hours that a system can take over? Firms that wait to ask this question until clients do the maths will be driven by events. It is better to examine your own business model critically today. As a client of my accountant, I have already done the maths. The trigger was not primarily the cost. I see administrative tasks as an inefficiency with little added value, and I wanted to be rid of them. The price has not changed so far, because we already work with a fixed fee. My accountant now does less work for the same fee. For the moment, as noted above, this buys the firm time. The discussion about price will follow.

Transparency note

Generative AI was used in a supporting role for research and editorial preparation of this article. The conceptual design, the selection and weighting of the content, the professional assessment, the final wording of the text and the responsibility for all statements and conclusions rest with me.

Sources