Business Growth

AI and Pricing: Dynamic and Algorithmic Pricing for Services, and the Fairness Constraints That Govern It

AI pricing arrived in services faster than the norms governing it. The economic evidence is real: studies of German retail fuel markets found margins rose roughly 28 percent in local duopolies after both competitors adopted algorithmic pricing, and OECD work now tracks algorithmic pricing as a live competition concern across G7 jurisdictions (OECD, 2025). The constraint evidence is equally real and much older. Kahneman, Knetsch, and Thaler (1986) showed communities hold firm fairness norms: raising prices to exploit demand shifts is judged unfair even when economically rational, a finding Uber's surge pricing relearned publicly. For service founders, the opportunity is decision-grade pricing intelligence; the boundary is a fairness frontier that algorithms cross at the brand's expense.

Joshua Agonya Pi'Rwot

By Joshua Agonya Pi'Rwot

Founder, Business Growth Accelerator

Executive summary

Algorithmic pricing lifted margins 28 percent in studied duopoly markets, yet fairness research shows customers punish exploitative prices. This deep dive maps AI pricing for services and the constraints that bound it.

Section 1

The five challenges at a glance

The gap between what AI pricing can do and what customers will accept is the defining tension of this decade's commercial models. Algorithms optimize for short-run revenue; relationships price over years. The table summarizes the five challenges service firms face as pricing automation spreads: fairness backlash when prices track demand rather than cost, opacity that breeds distrust, regulatory exposure as algorithmic collusion and surveillance-pricing scrutiny intensifies, personalization that quietly damages loyalty (Gartner's 2025 survey found personalization can triple purchase regret at key journey points), and the data-poverty problem of small firms trying to train pricing models on thin deal histories. The analyses that follow cover the fairness evidence, the regulatory frontier, and the practical architecture for using AI as pricing intelligence rather than autonomous price-setter.

Section 2

Challenge 1: Fairness norms are a hard constraint, not a communications problem

The foundational evidence predates the technology by four decades. Kahneman, Knetsch, and Thaler (American Economic Review, 1986) surveyed community standards of price fairness and found a consistent structure they called dual entitlement: customers grant firms the right to protect profits when costs rise, but judge it unfair to exploit shifts in demand by raising prices. Raising the price of snow shovels after a blizzard was judged unfair by 82 percent of respondents, even though it is textbook market clearing. The norms have teeth: customers who perceive unfairness punish firms beyond the transaction, through churn, negative word of mouth, and willingness to pay premiums to competitors perceived as fair. Uber's surge pricing became the modern demonstration: studies of rider perception found majorities viewing surge as unfair, with opacity and rapid fluctuation, prices changing every few minutes, amplifying the reaction even when surge demonstrably increased driver supply (Psychology Today, 2015; fairness-perception studies of Uber usage). The implication for AI pricing in services is precise: algorithms that raise prices because demand allows it will be detected and punished, while prices that move with documented cost, scope, or capacity logic remain within the entitlement. The framing is not cosmetic; it determines which side of the norm the increase lands on. A service firm can charge more for rush delivery (the customer buys scarce capacity) where it cannot charge more because its calendar is full (the firm exploits the customer's timing). Same economics, opposite fairness verdicts.

Section 3

Challenge 2: The regulatory frontier is moving toward algorithmic pricing fast

What was a behavioral risk is becoming a legal one. The OECD's 2025 review of algorithmic pricing across G7 jurisdictions catalogs the concerns now on regulators' desks: tacit collusion among competing algorithms, discriminatory and exploitative pricing, and hub-and-spoke structures where many competitors price through one shared tool (OECD, 2025). The empirical base is unusually strong for a policy debate: research on German retail fuel found stations adopting AI pricing raised margins around 5 percent on average, and roughly 28 percent in local duopolies once both rivals adopted, with no comparable change in monopoly markets, a pattern consistent with algorithmic softening of competition rather than efficiency (NBER, 2024). Experimental work has pushed further: LLM-based pricing agents in simulated oligopolies converge on collusive pricing quickly and consistently, without being instructed to coordinate (Fish, Gonczarowski, and Shorrer, 2024). Legislative responses have begun, including proposed US statutes targeting pricing algorithms that facilitate coordination and Department of Justice statements of interest in private algorithmic-pricing litigation. For service founders the exposure is more concrete than it sounds: industry-standard pricing tools that benchmark rates across competing firms in one vertical are structurally similar to the hub-and-spoke pattern regulators are examining, and surveillance pricing, setting individualized prices from behavioral data, has drawn formal regulatory inquiry. The practical rules: never feed competitively sensitive pricing data into shared tools that recommend prices back to competitors; document the cost and value logic behind algorithmic price moves; and keep individualized pricing tied to scope differences, not to inferred willingness to pay.

Section 4

Challenge 3: Personalized pricing quietly damages the relationships services depend on

The most seductive AI pricing application, charging each customer their personal maximum, is the one the evidence warns against most clearly. Gartner's 2025 survey of 1,464 B2B buyers and consumers found a personalization paradox: customers experiencing personalization were 1.8 times more likely to pay a premium, but 53 percent reported negative experiences, and those customers were 3.2 times more likely to regret their purchase and 44 percent less likely to buy again (Gartner, 2025). Personalization created time pressure and information overload at exactly the journey points where buyers needed confidence. Price personalization compounds this with a fairness violation: dual-entitlement research implies that two clients paying different prices for identical scope will, upon discovery, read the difference as exploitation, and in B2B services discovery is routine, procurement teams benchmark, executives change companies, agencies share decks. The damage lands on the asset service firms live on: multi-year trust. The economically meaningful alternative is versioned pricing rather than personalized pricing: let prices differ because scope, speed, seniority, and terms differ, dimensions customers can see and choose, rather than because an algorithm inferred one buyer's desperation. AI is extremely useful here in the legitimate direction: estimating which version a prospect values, predicting deal-specific willingness to pay to inform which tier to lead with, and flagging when a proposed discount is unjustified by any scope difference. The distinction to enforce: AI may inform what you offer, and may never silently change what identical buyers pay.

Section 5

Innovative solutions

The productive frontier for service firms is AI as pricing intelligence, not autonomous repricing. Four patterns are emerging. First, willingness-to-pay estimation at proposal time: models trained on the firm's own deal history (win/loss, client attributes, scope, final price) flag when a draft quote sits below the winnable range; even simple models beat founder intuition once a hundred-plus historical deals exist, and below that threshold, structured rules outperform models, the honest answer to the thin-data problem (NBER, 2024, documents how data scale conditions algorithmic pricing performance). Second, dynamic capacity pricing with published rules: rush fees, peak-season scheduling premiums, and last-minute booking pricing, all framed as buying scarce capacity, which sits on the acceptable side of dual entitlement because the trigger is visible and chosen by the buyer (Kahneman, Knetsch, and Thaler, 1986). Third, quote standardization: AI review of outgoing proposals against the firm's pricing policy catches the casual concessions that build pocket-price leaks, functioning as an automated deal desk. Fourth, continuous repricing intelligence for retainers: models that track delivered scope against contracted scope and flag agreements drifting underwater, turning the annual repricing conversation from guesswork into documentation. Across all four, the governing architecture is the same: the algorithm recommends, a human approves, the rules are explainable to a client in one paragraph, and every automated price move can show its cost-or-scope justification. Firms running this stack get most of the measured margin upside of algorithmic pricing while staying inside the fairness frontier that pure optimization crosses.

Section 6

Solution framework

Govern AI pricing with a three-zone policy. Green zone, automate freely: internal intelligence that never touches a customer-facing price without human review, willingness-to-pay scoring, underpricing alerts, retainer drift detection, proposal policy checks, competitive rate research from public sources. Yellow zone, automate with published rules: customer-facing dynamic pricing whose triggers are visible, chosen, and cost-or-capacity justified, rush fees, peak scheduling premiums, volume tiers, early-commitment discounts. The published-rule requirement is the fairness control: dual-entitlement research shows buyers accept price differences they can see the logic of and select into (Kahneman, Knetsch, and Thaler, 1986), and surge-pricing studies show opacity, not movement itself, drives the strongest backlash. Red zone, prohibited: individualized prices for identical scope based on inferred willingness to pay; demand-triggered increases without cost or capacity justification; and feeding pricing strategy into shared industry tools that recommend prices across competitors, the hub-and-spoke pattern in regulatory crosshairs (OECD, 2025). Operationally: appoint one pricing owner; require every algorithmic recommendation to carry its justification in plain language; review a sample of AI-influenced quotes monthly for drift; and write the client-facing explanation before deploying any yellow-zone mechanism, if the explanation sounds exploitative in one paragraph, the mechanism is red zone regardless of its economics. Revisit the policy semi-annually, because the regulatory perimeter is moving: what is reputational risk this year is statute risk soon.

Section 7

Evidence-based action plan

Days 1-20: build the data foundation. Consolidate deal history, every proposal for two-plus years with scope, list price, concessions, outcome, and client attributes, into one table; this is the training set for every later capability and most firms discover the data quality problem here, not the modeling problem. Simultaneously, write the three-zone policy and circulate it to everyone who prices anything. Days 21-45: deploy green-zone intelligence. Start with the highest-yield, lowest-risk use: an underpricing flag comparing each draft quote to historical winnable ranges, plus a retainer-drift report comparing delivered versus contracted scope across the book. Tools matter less than the discipline; a spreadsheet model beats no model. Days 46-75: launch one yellow-zone mechanism with published rules, typically a rush-fee schedule (25-50 percent premiums for compressed timelines) framed explicitly as priority capacity purchase, the framing dual-entitlement research validates (Kahneman, Knetsch, and Thaler, 1986). Announce it prospectively; never apply it retroactively. Days 76-90: instrument and review. Metrics: realized price versus list (should rise), win rate (should hold within a few points), rush-fee attach rate, client complaints referencing pricing (target zero fairness complaints), and concession frequency (should fall as the automated deal desk catches drift). Quarterly thereafter: re-run the model on fresh deals, audit a quote sample for red-zone leakage, and scan the regulatory landscape, OECD and national enforcement activity around algorithmic pricing is updating fast enough that annual review is no longer sufficient (OECD, 2025). The end state is a firm where pricing decisions are informed by machines, justified by economics, and owned by humans. For adjacent evidence in this pillar, see [The Pocket Price Waterfall for Service Businesses: Where Agency Margin Really Leaks](/blog/growth-pocket-price-waterfall-services) and [Pricing New Services: What the Evidence Says About Penetration, Skimming, and Launch Price](/blog/growth-pricing-new-services-launch).

FAQ

Direct answers for operators.

Can a small service business actually use AI pricing with limited deal data?

Yes, but start with intelligence, not automation. Below roughly a hundred historical deals, structured pricing rules outperform trained models, so use AI for proposal review, rate research, and retainer-drift detection while you accumulate clean deal data. Once the history exists, simple willingness-to-pay models flagging underpriced quotes deliver most of the value with none of the customer-facing risk of automated repricing.

Is dynamic pricing fair for services?

It depends on the trigger. Fairness research (Kahneman, Knetsch, and Thaler, 1986) shows buyers accept price differences justified by cost, scope, or visibly scarce capacity, rush fees and peak premiums the client chooses, but judge demand-exploiting increases as unfair and punish them. Publish the rules, let clients select into premiums, and never raise prices simply because an algorithm detects you can.

What is the regulatory risk in using pricing algorithms?

Three exposures are live: algorithmic collusion, where competing firms using the same pricing tool effectively coordinate (the hub-and-spoke pattern OECD flagged in 2025); surveillance pricing based on personal behavioral data, now under formal inquiry; and discrimination claims when individualized prices correlate with protected characteristics. Mitigations: never share pricing strategy with cross-competitor tools, document cost-or-scope justifications, and keep humans approving customer-facing prices.

Should we charge different clients different prices using AI willingness-to-pay estimates?

Not for identical scope. Gartner's 2025 research found personalization tripling purchase regret at key journey points, and fairness norms mean discovered price discrimination reads as exploitation, fatal in B2B where procurement benchmarks routinely. Use willingness-to-pay estimates to decide which tier or version to lead with, and let prices differ only along dimensions the client can see and choose: scope, speed, seniority, and terms.

Joshua Agonya Pi'Rwot

Written by

Joshua Agonya Pi'Rwot

Founder, Business Growth Accelerator · Country Director, AVODA Group Uganda · EMBA

Joshua helps service-business operators turn scattered marketing into a clear path from first attention to booked call. He is Founder of Business Growth Accelerator and Country Director of AVODA Group Uganda.