Riley Brown and the Zendesk Story
Interview July 20, 2026
“The arrival of AI fundamentally changed the pricing problem for SaaS; the traditional seat‑based construct was suddenly at structural risk.”
“We deliberately designed for a world where seat purchases stagnate or decline and outcome‑based revenue becomes the primary way we capture value from automation.”
Key Points
AI is an existential shock to seat‑based CX pricing, forcing platforms like Zendesk to rethink both pricing and the underlying business model around measurable outcomes rather than headcount.
Zendesk’s core innovation was moving from ticket‑based metrics to rigorously defined automated resolutions, audited by LLMs, creating a hybrid model where outcome‑based revenue can progressively replace seat‑based revenue as automation rises.
Effective pricing under disruption is inherently multi‑disciplinary: you need cross‑functional teams, explicit planning for future states you cannot yet fully observe, and live market experimentation (including parallel models) to learn where customers truly perceive value.
Credit‑based systems offer a flexible backbone for agent‑economy pricing, and Riley anticipates a shift toward currency‑denominated wallets plus the emergence of derivative markets for compute and outcome capacity.
In the agent economy, distribution becomes headless and microtransactional: MCP/API access should be free while usage is monetized. In this world, pricing must now serve two personas in tension — buying agents, which reward standardized definitions they can compare, and human buyers, who pay a premium for differentiation.
Background
I came into software pricing with an economist’s toolkit and a strong interest in strategic game play. My academic focus was economics, with a particular attraction to game theory and the theory of the firm—disciplines that frame markets as structures of incentives and strategic interaction. From the beginning, I wanted to be close to the frontier where new business models and new technologies meet, which led me to San Francisco and the early cloud software ecosystem.
After a first role at a small startup in Kansas City, I joined Zendesk, then one of the darlings of the early seat‑based SaaS world in customer experience (CX). I started as an analyst, working across strategy, planning, and marketing, and eventually moved into product. Zendesk at that time had already established itself in ticketing, but was pushing toward a broader CX platform vision that spanned support, sales, voice, chat, and the full range of customer interaction channels. To support this ambition, the company invested in a product strategy and analytics team charged with ensuring that the product roadmap and commercialization model were coherent with the platform vision.
Being embedded in this team gave me a broad vantage point on cloud software: I was learning SQL, understanding the market, segmenting buyers, and watching how a product‑led growth (PLG) motion scaled into a more sales‑led enterprise motion. Zendesk’s global footprint—from San Francisco to Singapore and Melbourne—meant that I worked with product teams across regions, seeing how a common platform vision had to adapt to local realities. It was an ideal training ground for understanding how visionary software products are actually built and commercialized.
Monetization at Zendesk and Algolia
Monetization was a core remit of our product strategy group. At the time, Zendesk’s commercial model was almost entirely seat‑based; new capabilities raised the canonical question: bundle into existing SKUs or split as add‑ons? Nearly everything was priced per seat, with exceptions for APIs and a few platform‑level components. As Zendesk grew into a multi‑product company, a more systemic pricing problem emerged: the same buyer was purchasing multiple SKUs across the portfolio. The obvious—but powerful—move was to consolidate these into suites, and to treat pricing not as a feature‑level decision but as a portfolio‑level strategy that carried a view of future capabilities as well as current ones. That realization led Zendesk to invest in a dedicated pricing team, formed out of the existing product strategy group and reporting into product.
I left briefly to work on monetization at Algolia, another San Francisco‑based company, before being invited back to Zendesk to join the founding pricing group in 2019. This was still pre‑AI‑hype. Our focus was on being more intentional about pricing systems, designing a portfolio of offerings rather than a single product, and embedding pricing recommendations into the lifecycle of new capability development. At the same time, Zendesk was building out a deal desk to support its enterprise push and cycling through phases typical of seat‑based SaaS: SKU expansion and unbundling to address diverse buyers, followed by consolidation and repackaging into new versions of the suite. For several years, the core model remained seat‑based, with some platform exceptions, and my work was about locating the company on this expansion–consolidation timeline and tuning the strategy accordingly.
The arrival of AI
The arrival of AI fundamentally changed the pricing problem for SaaS. As AI moved onto the tip of everyone’s tongue, the dominant narrative in customer service was displacement: AI would take over a large share of Tier‑1 interactions—questions about returns, policies, and other routine queries. For Zendesk and its peers—Intercom, Freshdesk, Salesforce, and others—this was an existential threat to the seat‑based revenue base. Internally, our question quickly became: how do we pivot not just the pricing model, but the business model around value in an AI world? If customers adopted AI successfully and began to reduce their human agent headcount, how could a CX platform continue to capture value commensurate with the automation it was enabling? The traditional seat‑based construct was now at structural risk; pricing had to carry a larger share of the adaptation burden.
Around 2023, we made this transition the explicit focus of the pricing team. We conducted studies with customers and non‑customers—CFOs, CIOs, heads of customer service—asking how AI would reshape their operations. Most respondents planned to adopt AI within one to three years and expected significant automation, but were not yet ready to say they would reduce budgets. We read this as temporary caution rather than a long‑term equilibrium. We deliberately planned for a world in which seat purchases would stagnate or decline, and contraction in seat counts would become a leading signal of value created by automation.
That anticipation drove a systematic exploration of outcome and usage metrics that could complement or eventually replace seats. We knew usage‑based pricing would be a key entry ticket into AI, and we had a uniquely well‑defined unit of usage: the ticket. In CX, tickets (or cases, conversations, depending on the vendor) are mutually recognized events that aggregate value: customers know what a ticket is, Zendesk knows what a ticket is, and the metric is already embedded in operational reporting. Tickets therefore satisfied a central requirement for outcome‑oriented models: a shared, measurable unit that both vendor and customer accept as bearing value.
Aligning with value
However, a pure per‑ticket pricing model did not get us close enough to the true point of value. Customers do not perceive value from a ticket simply because it exists; value is realized when the initial question is resolved. When we tested alternative formulations of the ticket metric with survey respondents, willingness to pay and acceptance were markedly higher for variants that emphasized resolution—resolved questions or resolved queries—rather than raw ticket volume. This insight led to the concept of the “automated resolution.” Initially, the label was inelegant, but it captured a more accurate outcome: not just a ticket processed, but a customer’s question resolved without human intervention.
Defining this rigorously was non‑trivial. In Zendesk’s existing product, every ticket had a “resolved” status, but this was a system state, not a guarantee of customer satisfaction. A resolved ticket simply meant the human agent no longer saw it in the interface; it did not certify that the end user’s problem was actually solved. We had historical experience with resolution‑based pricing through an earlier bot that charged per resolution defined as a user clicking “my question is resolved.” That definition was too strict: many queries are effectively resolved even if the user does not press the button, and tying billing to an explicit click introduces unnecessary volatility and under‑counting. Intercom had used a similar construct, with comparable limitations.
The breakthrough came from product leadership. The idea was to embed AI itself into the definition of an outcome. If large language models (LLMs) could evaluate tickets against a repeatable prompt, they could act as an automated auditor of resolution, checking both the system’s notion of “resolved” and the likely satisfaction of the end user.
Practically, we designed a process in which an LLM runs a reproducible prompt over tickets and returns a classification: yes, this is a ticket; yes, it is resolved in the conventional system sense; and yes, based on the content and context, this appears resolved from the end customer’s point of view. Only when all of these conditions were met would Zendesk count an “automated resolution” for pricing purposes. This additional evaluation step makes the definition less permissive for Zendesk and more protective for the customer: it constrains the vendor’s ability to claim outcomes while expanding the customer’s confidence that billed outcomes correspond to genuine value. Early customer testing showed that this stricter, AI‑validated definition of automated resolution aligned closely with how decision makers think about value, and it elicited higher acceptance than simpler ticket‑count metrics. It also became a differentiator versus other vendors who were experimenting with automated resolution or per‑outcome pricing without the AI evaluation loop.
The result was a hybrid model: we sell automated resolutions on top of seats and observe, over time, how spend and budget allocation migrate from seat‑based charges toward outcome‑based charges as AI adoption increases and automation rates rise. Conceptually, we are looking for a clear relationship between rising automation, more automated resolutions, and eventually a decline in seat‑based revenue—ideally offset by growth in outcome revenue.
Lessons for SaaS and AI agents
From this transition, I have several lessons that are broadly applicable to B2B SaaS and, increasingly, to agent‑based AI systems. First, treat pricing design under disruption as a multi‑disciplinary problem. In our case, none of the participants were “experts” in AI‑driven outcome pricing at the outset. We made progress by getting smart people in the room—regardless of whether they carried pricing titles—socializing the problem, sharing constraints and trade‑offs, and co‑creating frameworks on the fly. That collaborative model is what ultimately led to the idea of using AI as part of the outcome definition itself.
Second, build for a future you cannot yet fully observe. We planned explicitly for a world in which seat purchases would decline and labor budgets would shift toward technology. That required talking not just to large, early adopters but across the customer portfolio, including late adopters, to understand how different organizations envisioned their AI journey. The studies on planned AI adoption, expected automation, and eventual budget shifts gave the pricing team enough directional confidence to design for future states rather than current comfort.
Third, emphasize real‑time learning through market testing. In a context where information is imperfect and the landscape is moving quickly, I see value in what I describe as “move fast and break things” applied to pricing. Salesforce’s current approach—supporting multiple models in parallel and learning from customers’ actual choices—is a good example of using live experimentation to navigate uncertainty. Running multiple pricing constructs side by side, even at the cost of internal complexity, can yield richer signals about where value is perceived and how budgets are reallocating.
What is coming
After leaving Zendesk earlier in 2026, I joined a company that implemented a credit‑based model and now work as an independent consultant, having designed both outcome‑based and credit‑based hybrid pricing systems. That dual exposure informs my view of where agent‑economy pricing might go next. In my perspective, credit models serve as a powerful catch‑all for anything AI can do. A well‑structured credit system always has a place for a new AI capability: when a new use case is discovered, it can be mapped to the credit rate card, expanding the menu of priced outcomes without redesigning the entire model. For businesses undergoing transition—shifting from human labor to AI agents—as well as new AI‑native companies building their first pricing models, credits offer a flexible backbone.
Looking ahead two to three years, I expect several evolutions. One is currency as the unit of measure. Many buyers already express a preference to fund AI usage as budget in real dollars and cents rather than abstract credits. The lesson I’ve taken from CX software buyers is that they want fewer abstractions between themselves and the truest measure of value: the dollar. When every vendor maintains its own credit definitions and its own rate card, it becomes nearly impossible for the CIO or CFO, whoever holds the budget, to understand the true value they are receiving in real dollars, and smart vendors will remove that translation burden rather than force buyers to carry it across a portfolio of incompatible credit systems. Over time, I expect the credit to become an accounting abstraction beneath a currency-denominated wallet."
A second evolution is the emergence of secondary markets for compute and even outcome capacity. Once credits or wallets are funded in real currency and map to well-defined units, whether raw compute or higher-order value metrics built on top of it, those units begin to behave like a utility, and demand for trading them will follow. Buyers will want to hedge future usage, sell forward unused capacity, and manage the risk of large prepaid commitments. For platforms like Salesforce, Microsoft, Oracle, SAP, and perhaps ServiceNow, sanctioning such a market has real business benefits. It expands the market for their credits, reduces buyer risk in exactly the way futures markets are designed to do, and makes large upfront commitments easier for customers to justify.
But it is a complicated strategic decision with genuine downside. Breakage profits shrink toward zero, because what is sold is exactly what is delivered and someone will always step in to consume available capacity. Revenue recognition becomes very complex. A visible secondary price constrains the platform's own pricing power, and the platform cedes some control over who ultimately consumes its capacity and at what price. My prediction is that the large platforms will eventually follow the path Amazon took with the EC2 Reserved Instance Marketplace, where AWS lets customers resell unused reserved-instance capacity to other customers through an exchange that Amazon itself operates and takes a fee on: the platform accepts the secondary market as inevitable, brings it inside its own walls, and monetizes it rather than fighting it.
Finally, I extend the discussion into the agent economy itself. I imagine an environment where many transactions are handled entirely by agents—human principals set constraints and budgets, but agents negotiate, allocate credits, and execute at speeds and granularities humans cannot match. Microtransactions for very small amounts of money happening at high frequency are a natural fit for autonomous agents; here, human oversight is both too slow and too costly.
In this world, distribution becomes increasingly headless. The primary interface may be an API or a Model Context Protocol (MCP) server rather than a human‑oriented UI. My view is that if MCP usage ultimately drives revenue—because the underlying product or API has a consumption‑based pricing model—then access to MCP itself should be free, and the pricing model should account for usage regardless of whether it flows through UI, API, or MCP. This is consistent with practice at Value IQ, where MCP server access is free but credits burned through MCP are billable.
In this agent economy, I expect a growing standardization of AI outcome and usage metrics – and with it a slow drift of price points back toward the price of compute. Once every vendor sells the same unit, the unit commoditizes, competition shifts from definition to price, and price competes down toward marginal cost, which in AI is compute. The strategic implication is that the durable margin will belong to vendors whose actions and outcome definitions remain genuinely differentiated while staying simple enough for the customer to understand. What makes this hard is that businesses will increasingly price for two personas at once. For the human buyer, a differentiated outcome definition is precisely what fetches a higher unit price; it is the vendor's argument for why its resolution, its action, its result is worth more than the market rate. For the buying agent, the opposite holds: standardization is what allows an agent to compare offers and make well-informed commercial decisions, and a definition it cannot parse is a definition it discounts or ignores. Managing the tension between those two personas, differentiated enough to command a premium with humans, legible enough to be chosen by agents, will be one of the defining pricing problems of the next few years.