How to price a service when AI cuts your hours

The Problem: Two Ways to Price After AI

A job that took 20 hours now takes 10. There are two obvious ways to price it, and each has a flaw.

Price by the hour. The project price falls to half. The buyer keeps the whole saving, and being good with AI earns nothing extra. Keep the old project price. The saving is yours, but at the same price the buyer sees little difference between you and anyone else, whether or not they use AI. Your only edge is speed, and only if the buyer cares about speed. And the first one to cut the price wins the comparison.

Cutting the project price to match your new hours is the same as billing by the hour, so the buyer still keeps the whole saving. Most buyers care about the result more than the hours: Ron Baker, who argues for value pricing, says no client buys time (IFAC), and in one small-business survey only about 16% wanted hourly billing only (NAB).

The better question is who the buyer is, what you can prove, and what the alternative costs them. Poker players ask the same thing when they choose a bet size.

A Bet Size Is a Price

You do not need to play poker for this part. In poker, the size of a bet is the price the other player pays to stay in the hand. A small bet is cheap to call, so more players stay in. A big bet is expensive, so fewer do, and each one that stays pays more.

No single size is right. Good players change it with the situation, and one analysis found that forcing a player to use only one size meant they could no longer bet all their hands profitably (Brokos, GTO Wizard). A service has the same property: one price for every buyer leaves money on the table with some and loses the sale with others.

Three Things That Set the Size

In poker	The pricing version
How strong your hand is. Strong hands and bluffs tend to bet big; middling hands tend to bet small (Brokos, GTO Wizard).	What you can prove. If you can show a result the buyer cannot get elsewhere, charge more. If you cannot, charge less and win more often. Poker allows a bluff. Pricing does not: never imply a result you do not have.
Who you are playing. The same bet gets different answers from different opponents, so the best size changes with them (Brokos, GTO Wizard).	Who the buyer is. A buyer who compares on price gets the lean option, and one who buys on outcome gets the full one.
How many players are in the hand. With several opponents, bets get smaller and bluffs rarer (GTO Wizard).	How many vendors the buyer is comparing. Against several, price tighter or give the buyer a reason you are not interchangeable.

Where the analogy stops. Poker is zero-sum: what one player wins, another loses. A sale is useful only if the buyer ends up better off than the price. Read every idea here as "the buyer's gain is larger than the price".

What the Buyer Will Pay, and What Holds You Back

Poker players separate two kinds of bet size. The balanced size is hard to beat whoever sits opposite. The exploitative size is chosen to profit from one opponent's habits. If a player keeps calling however big the bet, a bigger bet wins more from them. One Upswing analysis calls these players "inelastic": how often they call barely changes with the size, so bigger bets earn more (Upswing). The aim of a value bet is to be paid as much as possible while still being paid often enough (Little, PokerCoaching).

The pricing version is the same: if you believe a buyer will pay a lot, you can quote more. Even in poker, the sources do not agree on how far to go. One 2026 analysis found that against a player who calls too much, betting thinner on more hands did better than waiting for a monster and betting big (Brokos, GTO Wizard).

Three things hold the size back.

Limit	In poker	In pricing
Your image	How the table sees you changes how they answer. An aggressive image gets more calls, a tight image gets more folds (Little, PokerCoaching).	Your reputation and what your price says about you. A quote far above what you usually charge, or what others charge, changes how the buyer reads you.
History	Past hands with the same opponent are the input. One coach advises waiting for several consistent observations before a big adjustment (Little, PokerCoaching).	What you charged this buyer before, and what they know others pay. A guess about a buyer's budget is not an observation.
Being exploited back	Every step away from the balanced size gives a sharp opponent something to use against you (Polk, Upswing).	A price set too high loses the sale. A price set too low is hard to raise later with the same buyer.

Price as a Position and a Filter

Many consultants raise their price to narrow the clients they work with, and to be seen as affordable or as premium. Here is what is known.

• A price is a clue to quality when buyers cannot judge before buying. A review of 36 studies found price and perceived quality positively related, but the effect was small and came mostly from lab studies on products (Rao and Monroe). Economic theory says the signal works when quality cannot be checked first and repeat business matters (Milgrom and Roberts).
• A higher price raises expectations. If the service meets them, buyers like it more. If it falls short, they rate it worse than the same service at a low price, in three wine experiments and one field experiment (Gneezy, Gneezy and Lauga). Premium positioning needs proof and delivery to carry it.
• A price that attracts buyers by being high is a narrow case. Economic models of luxury goods need others to see what you bought (Bagwell and Bernheim). A private service cannot rely on the price tag alone.
• Raising the price to attract better clients is practitioner opinion. Consultants such as Alan Weiss argue that higher, value-based fees bring more committed clients (getAbstract). I found no study of consultants or freelancers that measured it.

Position	What the price says	What has to be true
Affordable	Good value, an easy yes.	A narrow, repeatable scope, so the low price does not read as low quality.
Premium	A specialist for higher stakes.	Proof and delivery that meet the higher expectation the price creates.

To test it, raise the price on your next three quotes and note who says yes. Treat the result as a trial, not as proof.

What AI Actually Changes: the Evidence

The saving is the starting point of the pricing question, and the studies disagree about how large it is.

Study	What it found	Caveat
Noy and Zhang	Short professional writing tasks took about 40% less time, and rated quality rose about 18%.	Short tasks. The authors note real-world gains are smaller once checking is included.
Brynjolfsson, Li and Raymond	Support agents resolved 14% more issues per hour, 34% more for novices, with minimal effect for experienced agents.	One company, one tool.
Peng et al.	Developers finished one coding task 55.8% faster.	One task, chosen to suit the tool.
Dell'Acqua et al.	758 consultants: faster and higher-rated work on tasks inside the AI frontier, less likely to be correct on a task outside it.	Controlled tasks, not client projects.
METR	16 experienced developers on their own repositories took 19% longer with AI. They had expected a speedup, and still believed they had one afterwards.	Early-2025 tools, small sample. A 2026 update on newer tools found small speedups whose intervals include zero.

Together they say: the saving ranges from negative to large, it depends on the task, it tends to be biggest for less experienced people, and in the one study that asked, developers misjudged their own speedup. That matters for pricing: a project price set from an assumed speedup loses money on the job where the speedup does not appear.

What Has Happened to Prices So Far

• Commodity freelance work. After ChatGPT's release, writing freelancers on Upwork saw monthly jobs fall about 2% and earnings about 5.2% (Hui, Reshef and Zhou). Job posts in automation-prone categories fell relative to other work (Demirci, Hannane and Zhu). The damage so far is mainly to work that is easy to substitute. The studies measured jobs and earnings, not hourly rates.
• What small-business buyers prefer. In a 2018 survey of 750 Australian SMEs, 54% preferred a fixed fee and about 16% hourly only (NAB). It covers accounting clients and is from 2018. It is the closest evidence found.
• Pricing per result. Intercom prices its AI support agent at $0.99 per outcome, where an outcome is a confirmed resolution, a customer who stops asking for help, or a completed workflow (Intercom). It shows one AI vendor charging per result.

What the search did not find: hourly rates of AI-skilled against other freelancers, a speed premium for consulting or freelance work, and small-business willingness to pay for consulting. Any claim of "AI users earn a premium" or "AI users are priced out" would go beyond the evidence.

Where the AI Saving Should Go

A saving can be used three ways: keep it as margin, pass some of it on to win volume, or spend it on more scope or faster delivery. The factors above decide the mix. Take a hypothetical job: 20 hours at $100 an hour is $2,000. With AI it takes 10 hours, a 50% cut that is a little larger than the biggest in the studies above (about 40% less time on short writing tasks), so the hourly price comes to $1,000. If the buyer's best alternative costs $2,000, any price between $1,000 and $2,000 is open, and the factors above decide where.

The situation	Where to set the price
Repeatable work, cheap alternatives, buyers who compare	Toward the low end. Package it at a fixed price. The saving may buy volume; test it.
The buyer is paying for judgment, and a wrong answer is costly	Toward the high end. The study with novices and experienced agents suggests AI narrows the gap on routine work, which makes judgment the part worth pricing. This is an inference, not a finding.
The buyer needs it fast	Offer a fast tier next to the standard one. No study found a speed premium for consulting, so test it with a real offer.
You cannot yet say what the work takes	Measure first: log the hours of three or four projects. Sizing from an assumed speedup is a guess, and the one study that asked found developers guessed wrong.

Where This Does Not Fit

• The poker side is an analogy for thinking about the factors, and it gives no formula for a price.
• The table in the last section is a hypothesis built from the studies above. No study tested it on small-business consulting.
• If buyers can easily do the work themselves with AI, the buyer's alternative is cheap and the high end of the range is out of reach.
• Fixed prices put the scope risk on you. Write a narrow scope into every fixed-price offer.

Sources

• Andrew Brokos, Overchoice: Making Sense of Multiple Sizings, GTO Wizard Blog, 16 January 2024. Source - Medium-strength hands mostly bet small; the bigger the bet, the more it concentrates the opponent's range on strong hands. Forcing one size costs value.
• Tombos21, 10 Tips for Multiway Pots in Poker, GTO Wizard Blog, 13 February 2023. Source - Against several opponents, use smaller sizes and bluff rarely.
• Shakked Noy and Whitney Zhang, Experimental evidence on the productivity effects of generative artificial intelligence, Science, 2023 (MIT summary). Source - Short writing tasks took about 40% less time with ChatGPT, and rated quality rose about 18%.
• Erik Brynjolfsson, Danielle Li, Lindsey Raymond, Generative AI at Work, NBER Working Paper 31161. Source - An AI assistant raised issues resolved per hour by 14% on average and 34% for novices, with minimal effect for experienced agents.
• Sida Peng et al., The Impact of AI on Developer Productivity: Evidence from GitHub Copilot, arXiv 2302.06590, 2023. Source - Developers with Copilot finished one coding task 55.8% faster.
• Fabrizio Dell'Acqua et al., Navigating the Jagged Technological Frontier, Harvard Business School Working Paper, 2023 (SSRN 4573321). Source - In a study with 758 BCG consultants, AI users were faster and rated higher on tasks inside the AI frontier, and less likely to be correct on a task outside it.
• METR, Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity, 10 July 2025. Source - Experienced developers took 19% longer with AI on their own repositories, and believed afterwards that it had sped them up. A 2026 update on newer tools found small speedups whose intervals include zero.
• Xiang Hui, Oren Reshef, Luofeng Zhou, The Short-Term Effects of Generative AI on Employment: Evidence from an Online Labor Market (Washington University Olin summary). Source - After ChatGPT, writing freelancers on Upwork saw monthly jobs fall about 2% and monthly earnings about 5.2%. Short-term, one platform.
• Ozge Demirci, Jonas Hannane, Xinrong Zhu, Who Is AI Replacing? The Impact of Generative AI on Online Freelancing Platforms, Management Science, 2025 (DIW listing). Source - Job posts for automation-prone work such as writing and coding fell relative to manual-intensive work after ChatGPT.
• Accountants Daily, NAB finds 84 per cent of clients don't want time-based billing (2018 survey of 750 Australian SMEs). Source - In the survey, 54% of small businesses preferred a fixed fee and about 16% hourly only. Accounting clients in Australia, 2018.
• Intercom, Pricing (Fin priced per outcome), checked 19 October 2026. Source - The AI support agent Fin is priced at $0.99 per outcome, a confirmed resolution, a customer who stops asking for help, or a completed workflow.
• IFAC, Tomorrow's Firm and the Role of Value Pricing (Ron Baker), 2013. Source - An interview on value pricing in which Ron Baker says no client buys time.
• Dan B., Inelastic Poker Ranges: Bet Sizing vs Bad Players, Upswing Poker, 4 December 2023. Source - Against players whose calling barely changes with the size of the bet, use bigger bets and a value-heavy strategy.
• Jonathan Little, Value Betting in Poker, PokerCoaching, 2 April 2024. Source - The aim of a value bet is to be called for as much as possible while still being called often enough; an aggressive image gets more calls, a tight image more folds.
• Jonathan Little, How to Mix GTO and Exploitative Poker, PokerCoaching, 8 January 2025. Source - Use exploits as targeted tools, wait for several consistent observations before a big adjustment, and return to balanced play once the opponent adapts.
• Doug Polk, GTO vs Exploitative Play, Upswing Poker, 27 January 2017. Source - Departing from balanced play to exploit an opponent opens the door for them to exploit you back.
• Andrew Brokos, Exploiting Profiles: Episode I, The Calling Station, GTO Wizard Blog, 27 April 2026. Source - Against a player who calls too much, value bet more thinly; waiting for a monster and betting big was not the most profitable plan.
• Akshay Rao and Kent Monroe, The Effect of Price, Brand Name, and Store Name on Buyers' Perceptions of Product Quality: An Integrative Review, Journal of Marketing Research, 1989 (MSI listing). Source - A review of 36 studies found price and perceived quality positively related, with a small effect, mostly in lab studies on consumer products.
• Paul Milgrom and John Roberts, Price and Advertising Signals of Product Quality, Journal of Political Economy, 1986. Source - A theory model in which price can signal the quality of a product buyers cannot judge before buying, with repeat purchase central to the signal.
• Ayelet Gneezy, Uri Gneezy, Dominique Lauga, A Reference-Dependent Model of the Price-Quality Heuristic, Journal of Marketing Research, 2014. Source - A higher price sets a higher reference point: if quality meets it, buyers like the product more; if it falls short, they rate it worse than the same product at a low price. Three wine experiments and one field experiment.
• Laurie Simon Bagwell and B. Douglas Bernheim, Veblen Effects in a Theory of Conspicuous Consumption, American Economic Review, 1996. Source - A theory model in which luxury brands sell at higher prices to people who want to display wealth. A narrow case for visible status goods.
• getAbstract, summary of Alan Weiss, Value-Based Fees. Source - A consultant's opinion that higher fees signal quality and bring more committed clients. Practitioner opinion, not measured.

Related Reading

• The GEAR Model - when to hire full time, and when to buy the work by the day.
• The AI Tax - what AI adds to the cost of the work, not only what it removes.
• AI and the Fractional Operator - how AI changes the work of interim and fractional roles.