AI may cut the human hours behind familiar consulting deliverables, but the forecast here is a larger shift toward harder problems that were previously impractical. Experimental evidence shows task-dependent gains and failures, so framing, validation and accountability still need experts. Pricing grows more complex because human effort, model and compute spend, and calendar time are separate variables; none alone determines delivery cost or client value. Hourly billing persists alongside long-established fixed-price arrangements.
Damir Trokić asked on LinkedIn: if a consulting assignment that costs 100 hours at SEK 2,000, or SEK 200,000, could hypothetically be delivered in 20 hours, what happens to the invoice? It was an illustration, not a measurement.
I see two answers. The simple one is that we will still sell time, since in my view that is what consulting is, but we will spend it on significantly harder problems, things few can do today. The difficult one is that pricing will probably become more complex, because model choice, compute and calendar time start to count alongside human hours.
The frontier moves before the hours shrink
Consider a randomized experiment with 758 BCG consultants, published in Organization Science in March 2026. On tasks within GPT-4's capabilities, consultants with AI completed 12.2% more tasks, 25.1% faster, with better quality. On a task outside them, they were less likely to answer correctly.
So AI does not turn every 100-hour engagement into 20. It changes where expert attention goes. Framing the problem, validating the result and owning the recommendation still take people. There is counterevidence too: in a small METR study, experienced open-source developers using early-2025 AI tools took 19% longer while believing they were faster. METR later said newer speedups were likely but could not measure them reliably.
My forecast is that time freed from routine analysis goes into investigations once impractical within a budget or deadline. The evidence makes that plausible, not certain.
Three variables instead of one
Human hours, machine spend and calendar time are separate variables, and model pricing already reflects that. OpenAI charges different rates by model and token type, and its Flex processing is cheaper in exchange for slower responses and occasional unavailability.
Picture a hypothetical case. A manufacturer needs a recommendation before a supplier contract renews. AI cuts the manual document comparison, so the team can test twenty scenarios instead of three. Specialists still define constraints, chase contradictory records and own the recommendation. Faster processing costs more than the delay-tolerant option, and the client's approvals may remain the bottleneck.
Human hours fall, while total delivery cost could rise or fall. Client value rises only if the extra scenarios, or the earlier answer, change the decision.
In *Interstellar*, Cooper's crew lands on Miller's planet, where gravity near a black hole makes one hour there equal seven years back on Earth. The landing costs the team a few hours of effort. Romilly, waiting in orbit, ages 23 years.
That scene is a good way to read a consulting invoice. Hours worked and time elapsed are different clocks, and the cost of a mission depends on which one you spend. AI adds a third dial: how much machine capacity you put behind each hour. A fair price has to say which clocks the client is actually buying.
A price with more than one clock
None of this means hourly billing is dead, and fixed or outcome-linked pricing is not new. US federal procurement rules describe firm-fixed-price contracts with performance or delivery incentives. What changes is how many variables a fair price must reflect.
Clients will ask whether they are paying for expert judgment, model capacity or speed. I think the firms that do well will explain each plainly and price the combination that helps the client decide.
The bigger story, as I see it, is consulting taking on harder problems at a price that has to read more than one clock.
