Agency Outcome-Based Pricing: What Must Be Defined

From Selling Hours to Selling Outcomes

Outcome-based agency pricing ties some compensation to a defined client result rather than to labor time or a fixed list of deliverables. It can align incentives when the agency can influence the result and both parties can measure it. It becomes dangerous when "outcome" is vague, attribution is weak, or the agency absorbs risk created by the client's product, inventory, pricing, approval delays, data quality, or website.

The right question is not whether hours are obsolete. It is whether the parties can describe the causal chain from the agency's work to the result well enough to allocate responsibility.

Automation exposes the weakness in selling time

In episode 103 of Venture Step, RJ Talyor describes marketing work that once required a team and now takes a much smaller group far less time. Automated bidding reduced manual media work. Generative systems now reduce parts of copy and image production.

If an agency sells hours, faster execution can reduce billable inventory even when the client receives the same or greater value. If it sells a fixed number of emails or images, abundant generation can make the deliverable feel interchangeable. That creates pressure to price strategy or outcomes.

The pressure is real, but it does not prove that every service should move to performance fees. Research, brand positioning, crisis work, organizational change, and early product discovery often create value that is delayed or difficult to isolate. A hybrid model may fit better than pretending every contribution has a clean revenue event.

Five pricing models allocate different risks

Hourly pricing pays for verified effort. It protects the provider when scope is uncertain, but it asks the client to carry efficiency risk.

Project pricing pays for a defined body of work. It rewards efficient delivery, but scope changes can produce conflict.

A retainer pays for access, capacity, or an ongoing program. It can support continuity, though the parties still need a shared view of what good service means.

Value-based pricing connects the fee to the importance of the problem or expected economic value. It is not automatically contingent on the measured result.

Outcome-based pricing makes some payment depend on an observed result. It transfers more financial risk to the provider and creates a stronger need for definitions, data access, control, and dispute rules.

The comparison is not a ranking. Each model works only when it matches the uncertainty and control in the work.

The outcome needs a full definition

"Increase revenue" is not a contractual measure. A workable outcome needs a population, metric, data source, baseline, comparison, time window, attribution rule, exclusions, and payment formula.

Suppose an email agency is paid on incremental revenue per recipient. The parties still need to decide whether revenue means ordered, paid, shipped, or retained revenue. They need to define returns, discounts, taxes, currency, canceled orders, cross-device purchases, and the attribution window.

They also need a counterfactual. What would have happened without the agency's intervention?

Microsoft's research on online controlled experiments explains why randomization is valuable for causal claims. A concurrent holdout can reveal whether a campaign caused additional behavior rather than merely receiving credit for purchases that would have happened anyway.

The dependency map should come before the fee

Marketing outcomes sit inside a system. Creative quality may affect attention and clicks. Revenue also depends on audience selection, deliverability, offer, price, inventory, site speed, checkout, product appeal, competition, seasonality, fulfillment, and returns.

flowchart LR
    A["Agency-controlled work"] --> B["Campaign exposure"]
    B --> C["Customer response"]
    C --> D["Site and checkout"]
    D --> E["Order and retained revenue"]
    F["Client approvals, inventory, price, product, and operations"] --> B
    F --> D
    F --> E
    G["Seasonality, competition, and market conditions"] --> C
    G --> E

The diagram makes a pricing principle visible: responsibility should follow control. Current UK government guidance on risk allocation and pricing says performance measures should be objective and suppliers should be accountable for results they can influence. Although the guidance is written for public contracting, the logic transfers to agency work.

If the client changes prices, pauses campaigns, runs out of stock, delays approval, or rebuilds the checkout during the measurement window, the payment model needs a response. Otherwise the contract quietly makes the agency insure the client's operations.

A baseline must survive ordinary business change

A weak baseline compares this month with last month. That can confuse agency impact with holidays, promotions, list growth, product launches, or changes in demand.

A stronger design may use a randomized holdout, matched market, phased rollout, or another credible comparison. The design should be selected before results are visible. It should also record sample size, stopping rules, assignment failures, and data loss.

The 2026 UK Green Book describes additionality as the outcome caused by an intervention beyond what would have happened under business as usual. Marketing teams use different tools, but the same distinction matters. Attributed revenue is not necessarily incremental revenue.

Incentives can change behavior in unwanted ways

A payment formula tells the provider what to optimize. A narrow metric can produce narrow behavior.

If compensation depends only on open rate, a system may learn sensational subject lines. If it depends on last-click revenue, it may favor messages sent close to a purchase while undervaluing earlier work. If it depends on immediate conversion, it may increase discounts, message frequency, or pressure in ways that damage margin, unsubscribes, trust, or long-term value.

Outcome contracts in other sectors show the same risk. The OECD's review of payment-by-results employment services discusses risk transfer and the possibility of providers favoring easier cases. In marketing, the equivalent could be optimizing for customers who were already likely to buy while neglecting harder but strategically important groups.

A balanced measure should include the desired result and guardrails. Revenue might sit beside margin, unsubscribes, spam complaints, return rate, customer complaints, and brand or legal incidents.

Client obligations belong in the model

Outcome pricing is a two-party operating system. The client may need to provide timely approvals, clean data, adequate inventory, functional tracking, product availability, agreed offers, and access to the relevant systems.

The provider may need to document the intervention, preserve experiment assignments, disclose material changes, maintain security, report exceptions, and avoid optimizing outside the agreed boundary.

These are not administrative details. They are the conditions under which the outcome can be interpreted.

A hybrid model often handles uncertainty better

A practical first agreement may combine a base fee with a bounded performance component. The base can cover discovery, setup, governance, fixed capacity, and work whose value is not immediately measurable. The variable component can reward an agreed incremental result.

Caps and floors can prevent one noisy period from threatening either party. A learning period can establish data quality and a baseline before compensation changes. The parties can also define a review window rather than treating one formula as permanent.

This approach does not eliminate disagreement. It makes the disagreement inspectable.

A defensible pilot has an exit

Before moving to outcome pricing, write the causal chain in one page. Name every material dependency and who controls it. Define the metric and counterfactual. Freeze the measurement plan before launch. Set guardrails and data-quality rules. Then define how either party can stop or reset the arrangement when the assumptions fail.

The UK government's evidence review of payment-by-results contracts reaches a useful general boundary: there is no single best model, and external factors can materially affect delivery. Marketing contracts deserve the same honesty.

The goal is not to remove hours from the vocabulary. It is to stop confusing activity with impact while refusing to call a number an outcome before the causal and commercial terms exist.

For the measurement side of this decision, continue to [[How to Evaluate Predictive AI Marketing Claims]]. For a wider discussion of brand promises and delivery, [[E022 Content Plan|episode 22]] is the stronger companion. [[E119 Content Plan|Episode 119]] adds a founder's account of using AI for practical operating work without surrendering judgment.

Sources and editorial notes

This analysis uses the E103 transcript for Talyor's account of agency economics. The contracting principles are a Venture Step synthesis informed by current UK government and OECD guidance and Microsoft experimentation research. It is not a contract template or legal advice. A lawyer, finance lead, measurement specialist, and the people who control the underlying systems should review any live pricing arrangement.

AI assisted with research organization, structure, drafting, and validation. Dalton Anderson remains the attributed author and final editorial authority. The transcript and linked public sources control factual claims. Publication remains unauthorized.