Frameworks, core principles and top case studies for SaaS pricing, learnt and refined over 28+ years of SaaS-monetization experience.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
.png)
The audit tells you what to agentify. The harness is how you agentify it. And this is where the distinction between bolting AI onto a service business and truly converting it into an agentic business becomes critical.
Bolting AI onto a service business means giving your team access to ChatGPT, Claude, or Copilot and hoping they become more efficient. Some will. Most will use these tools sporadically, in ways that are idiosyncratic and inconsistent across the team. Output quality varies, client experience varies. Margin improvement is modest, perhaps 15 to 20 percent of production hours saved, which the firm can either pass on to clients as faster delivery or capture as incremental margin. This is closer to optimization of the existing model than to conversion.
Building a harness means constructing a software system that encodes your agency’s specific methodology, orchestration logic, quality standards, and institutional knowledge into an agentic delivery engine. The harness is not a collection of AI tools. It is the integration layer that makes those tools behave the way your best people would behave if they had infinite time and perfect memory.
The encoding process works in layers. Each layer corresponds to a component of the strategy firm’s existing human delivery process that must be translated into software.
The first layer is methodology. The firm’s pricing optimization engagement follows a specific sequence that the senior team has refined over hundreds of projects: it begins with a discovery phase (stakeholder interviews, data collection, hypothesis formation), moves through a research phase (competitive benchmarking, customer segmentation, willingness to pay analysis), then an analysis phase (pricing model construction, scenario modeling, sensitivity testing), and concludes with a recommendation phase (pricing architecture design, migration planning, board presentation).
This sequence, including the decision criteria for when to loop back from analysis to research, when to expand scope to include adjacent revenue levers, and when to flag that the client’s assumptions about their competitive position are wrong, becomes the orchestration logic of the harness. The workflow engine takes a client brief as input, including company profile, product portfolio, current pricing, and strategic objectives, and produces a task graph that routes work to specialized agents in the right order with the right context.
The competitive research agent knows it must complete its work before the pricing model agent can begin, because the model requires competitive reference points as inputs. The customer segmentation agent knows to flag when cluster analysis reveals a segment the client did not mention in the brief, because the methodology requires the orchestrator to pause and surface unexpected findings for human review.
The second layer is quality standards. When a junior analyst at the firm builds a market sizing model, the principal reviewing it is checking for a specific set of things: whether the data sources are credible and current, whether the assumptions are documented and defensible, whether the top down and bottom up estimates converge within an acceptable range, and whether the model handles edge cases that the analyst might not have considered. These quality criteria become verification tools in the harness.
A deterministic validation agent checks that all data sources are from the firm’s approved list, that every assumption cell in a financial model has a linked rationale, that all competitive pricing data is less than 90 days old, and that sensitivity ranges have been tested across at least three standard deviations. A model based evaluation agent reviews the strategic coherence of the analysis, asking whether the market sizing logically supports the segmentation, whether the segmentation aligns with the willingness to pay data, and whether the final pricing recommendation follows from the evidence rather than from unstated assumptions.
This layered verification process catches the same classes of errors that the principal catches, at machine speed, and surfaces only the judgment level questions for human review.
The third layer is institutional memory. This is where the strategy firm’s ten years of accumulated knowledge becomes its most powerful asset. The firm has conducted pricing engagements for dozens of B2B SaaS companies, and patterns have emerged that no public dataset captures.
Companies transitioning from founder led sales to a structured sales team almost always have pricing that is too low in the enterprise segment because the founder underpriced to win early deals. Vertical SaaS companies with less than $30 million ARR that attempt to introduce usage based pricing without a platform commitment fee see 15 to 25 percent higher churn in the first two quarters. Companies in the cybersecurity vertical can sustain 20 to 30 percent price increases during contract renewals without meaningful churn if the price increase is framed around a new compliance capability.
None of this is in a textbook. It lives in the heads of the firm’s partners and senior principals. It is the primary reason clients pay $400 per hour for their time. Encoding this knowledge into a memory system that agents can query when calibrating their analysis for a specific client context is the single highest leverage investment the firm can make in its harness.
The memory system is not a static database. It is a living knowledge graph that updates with every engagement, capturing not just what the firm recommended but what happened after the client implemented the recommendation, which strategies produced the predicted results, which ones underperformed, and why.
The fourth layer is data governance and client isolation. The firm works with companies that compete directly with each other. A pricing strategy developed for one cybersecurity vendor must never influence or contaminate the analysis for a rival. The harness must enforce strict client isolation at the data level, with separate memory partitions, separate context windows, and separate agent instances for each engagement. The permission system must also manage the boundary between what agents can do autonomously (research from public sources, plus model construction from provided data) and what requires human approval (strategic recommendations, analyses that draw on proprietary client data). This graduated trust model starts conservative and expands as the harness proves its reliability on lower stakes tasks.
When these four layers are encoded and integrated, the result is a delivery engine that can execute the production work of a strategy engagement autonomously, at machine speed, with quality that meets or exceeds the standards the firm’s partners would apply. A competitive landscape analysis that used to take an analyst team two weeks can be delivered in hours. A pricing model with full scenario analysis that used to consume forty hours of analyst time can be generated, verified, and made ready for partner review in an afternoon. The partners and principals shift from reviewing junior work product to directing the agentic system and making the judgment calls it surfaces, concentrating human effort on the activities where human value is highest.
Join companies like Zoom, DocuSign, and Twilio using our systematic pricing approach to increase revenue by 12-40% year-over-year.