Business Model Transformation in the AI Agent Era | Rewriting Costs, Revenue, and Solutions with a 90-Day PMF Re-Check Cycle (2026 Edition)

Hello, this is Hamamoto from TIMEWELL.

This is the ninth installment of the AI-agent-first management series. The previous pieces focused on technology selection and organizational change. Today I want to step into the question that weighs most heavily on executives: where exactly do you apply AI agents so the business model itself gets rewritten? Let me put my conclusion up front. Using AI purely for operational efficiency is not just a missed opportunity, it is genuinely risky from a management perspective. The reason is simple: every competitor is doing the same thing. If you both shave 10% off costs, but your revenue structure and solution stay frozen in the old model, the gap in margin profile reverses within five years. In this article I lay out three levers, cost, revenue, and the solution itself, and then describe how to embed a 90-day product-market-fit (PMF) re-check cycle into the way the company is run.

Stopping at operational efficiency is no longer enough

Over the past year, the most common story I hear from executives is, "We rolled out a chatbot and cut inquiry volume by 30%." That is a real result, the team on the ground worked hard, and I have no intention of dismissing it. But from a management standpoint, that is only a small fraction of the available return on investment. Even if support costs drop by 30%, if the price plan is still a flat monthly fee and the service itself is unchanged, the customer's perceived value has not moved at all.

Klarna is a clear example. Their AI assistant handled roughly two-thirds of customer interactions and the equivalent workload of 700 full-time agents in 2024¹. Customer service cost per transaction fell 40% in two years, from $0.32 in Q1 2023 to $0.19 in Q1 2025². So far that is a textbook efficiency story. What is more interesting is what came next. In spring 2025 Klarna walked back its AI-only stance and shifted to a human-AI hybrid model³. They preserved CSAT, kept the option of having a human handle complex cases, and used that to rebuild a path toward premium-tier monetization. Efficiency was the entry point, but the delivery structure of the service itself was redesigned. That is where the real lesson lives.

Operational efficiency is a destination anyone can reach. The actual job of management is to redirect the time and cost freed up by AI into new revenue and new services. When Japanese companies stall on AI adoption, it is often because no one has answered the question, "OK, now what does this allow us to change?" Efficiency is not the goal. It is the starting point of business model transformation.

Rewriting cost structure from fixed to variable

The first lever an executive needs to face is cost structure. Personnel cost has long been the archetypal fixed cost: salaries are paid regardless of order volume, peaks are absorbed by overtime, and slack periods see fixed cost squeeze profit. AI agents take this structure on directly. In May 2025 Salesforce introduced Flex Credits, a usage-based model for Agentforce: $0.10 per action, with packs starting at 100,000 credits for $500⁴. The point is "pay for what you use," and a clean shift from monthly licensing to a fully variable model is now institutionally available.

CIO.com, writing for CIOs, frames AI agent budgeting bluntly: "Build in control, ownership and accountability from day one, the same way you would for any other variable operations resource"⁵. That is a refreshing framing. It treats AI agents not as an IT department tool, but as a variable-cost line on par with cost of goods sold or logistics. Putting AI run-rate next to COGS in the monthly P&L is what makes it possible to see how AI spend correlates with gross margin. In months when margin compresses, you dial usage down. In months when it expands, you push harder. That kind of management decision becomes possible for the first time.

Here is a concrete simulation. Imagine a BPO contract handling 10,000 inquiries per month, currently costing 6 million yen in labor. If AI agent run-rate is 30 yen per inquiry, that is 300,000 yen per month. Of the 5.7 million yen difference, half could fund a price reduction passed to the customer and half could fund development of a new service. If the discount lifts deal volume by 1.5x and the new service raises ARPU by 20%, revenue nearly doubles. This is not a cost-cutting story; it is a growth strategy made possible by reorganizing the cost base. When fixed cost converts to variable, the range of offensive management decisions widens dramatically. That is the heart of cost strategy in the AI era as I see it.

Rewriting revenue structure from time-based to outcome-based pricing

Once cost is variable, the next lever is revenue. The wall many executives hit here is, "I don't have the courage to change how the service is priced." I understand. A consultancy or build-shop that has been quoting on man-month rates for years cannot easily flip to "we charge when the result lands." The market, however, has already started moving. Intercom's Fin AI Agent is priced at $0.99 per resolution, and Zendesk charges $1.50 only for tickets the AI fully resolves; failures are free⁶. A "success-based" pricing model that buyers can actually accept is now viable specifically because of AI agents.

A Sequoia Capital partner told Fortune in April 2026, "Services are the new software"⁷. The labor market is roughly six times the software market. Copilots (tools used by humans) are sold to humans, but autopilots (the outcome itself) are sold against the labor budget, which is an order of magnitude larger than the software budget. I broadly agree. In Japan I expect a similar shift through 2026: tax accounting firms moving bookkeeping from monthly retainers to "50,000 yen per filing," recruiting agencies keeping success fees while bringing AI-driven candidate sourcing in-house and doubling gross margin.

Three concrete patterns. First, time-based to outcome-based: change "100 hours per month" consulting and build engagements into "X yen per KPI achievement." Second, subscription to consumption: charge SaaS not by seat but by transactions processed. Third, add new revenue streams, such as exposing internal data via API or white-labeling AI agents to other companies. Gartner forecasts that 40% of enterprise SaaS will include outcome-based components by 2030⁸. The flip side is that any company that does not adapt within five years will be pulled into pure price competition. Pricing redesign is the kind of work only an executive can really do, and it is not safe to delegate.

Rewriting the solution itself with 24/7 availability and hyper-personalization

When cost and revenue structures change, the shape of the solution itself shifts. This is the third lever, and the one executives should be most excited about. AI agents do not get tired. They run at 2 a.m. They handle English, Chinese, and Vietnamese instantly. They retain the full history of a single customer and respond with proposals more thoroughly grounded than the customer's own memory. With that as a baseline, entire service categories that were uneconomical to staff with humans suddenly open up.

Take financial advisory for mid-market companies. Previously this was viable only at Big-Four pricing. With AI agents handling baseline analysis, a "full-time CFO companion" service in the 100,000-yen-per-month range becomes realistic. Or export-control consulting for sole proprietors. As economic-security regulation tightens, a human-only delivery model would cost at least 500,000 yen per month, but with an export-control AI agent like TIMEWELL's TRAFEED, the same service can be offered for tens of thousands of yen per month. METI-aligned classification checks and counterparty screening run automatically, around the clock. This is not efficiency. This is creating, from scratch, services that reach customer segments that previously could not be served at all.

Hyper-personalization is the other key. VC funding into AI agents reached $28 billion in 2025, four times the $7 billion of the prior year⁹. The pattern in those investments is striking: vertical-specialized agents in legal, healthcare, and finance return far more than general-purpose ones. The reason is simple. General-purpose agents can only solve the lowest common denominator. Vertical agents can be tuned to the workflow of a specific company. WARP, our AI consulting offering at TIMEWELL, takes this stance: we don't simply deploy generic tools, we design agents backwards from each client's specific management problem. I expect the price gap between general-purpose and specialized to widen by more than 10x within three years. Solution design is the largest area of differentiation in the AI era.

Embedding a 90-day PMF re-check cycle into management

Reading this far you may be thinking, "There is too much on the list." The catch is that running these moves on annual or three-year planning cycles will not keep pace with AI's progression. Bessemer Venture Partners reports that companies with a structured validation process reach PMF 35% faster¹⁰. AI models themselves update on a 90-day cadence, and what they can do shifts each cycle. Management cadence should match that, on a quarterly basis.

Here is a concrete rhythm. Month 1: build a hypothesis-validation prototype in two weeks. Month 2: run it in production with five pilot customers, measuring price, delivery format, and outcome metrics. Month 3: make the scale decision. Three options only: stop, continue, or accelerate. Repeat every quarter. Importantly, the PMF re-check should itself be assisted by AI agents. Feed in inquiry logs, churn reasons, churn signals, and conversion patterns, and let the agent propose "features to cut in the next 90 days" and "features to add." An OpenAI product lead said in a 2025 interview, "PMF for AI products has its own metrics: accuracy, hallucination rate, response quality. Traditional engagement metrics aren't enough." The metrics surface has expanded. The era of relying on executive intuition alone is coming to a close.

If you have an enterprise AI foundation like ZEROCK, you can embed a mechanism that auto-generates improvement hypotheses from internal data. The board pack arrives saying, "Top three complaint categories last month, conversion potential to a new service: X%," with no human assembly required. I see that state, where AI surfaces the hypotheses needed for management decisions and humans focus on judgment and approval, as the destination of AI-agent-first management. If you cycle hypothesis, validation, and decision every 90 days, you ship four updates while a competitor debates strategy at an annual offsite. Three years of that compounds into a gap that cannot be closed.

Summary: three actions executives can run in the next 90 days

Long article, so let me close with three actions you can start tomorrow.

First, reorganize cost structure. List your top three fixed-cost lines and pick one to convert into variable-cost AI agent run-rate. Salesforce Agentforce at $0.10 per action, or a vertical agent like TRAFEED, lets you launch a pilot in the tens-of-thousands-of-yen range per month.

Second, reorganize revenue structure. Pick one existing service line and propose a switch from time-based to outcome-based pricing to one customer. Even if they decline, you walk away with information.

Third, install the PMF re-check cycle. Add a 15-minute "PMF re-check" item to your next quarterly board agenda. That alone changes the quality of decisions.

Through WARP we work alongside executives on these three moves, helping them reach the first crest of business model transformation in 90 days. AI is not a cost line for efficiency; it is a lever for rewriting the business model. Until the executive holding the lever moves, the organization will not move. What will you change in the next 90 days? I am still iterating every day, and I will share new findings in future articles.

Related reading: the series overview is in Three strategic options for AI-agent-first management, the operational redesign side is in Stop, reduce, automate: classifying work in the AI era, enterprise trends from Google Cloud Next are covered in Enterprise AI agents at Google Cloud Next 2025, and a broader management view of the AGI era is in The arrival of the AGI era.

References

Footnotes

1. OpenAI case study, "How Klarna's AI assistant does the work of 700 full-time agents," https://openai.com/index/klarna/ ↩

2. Customer Experience Dive, "Klarna credits AI for slashing customer service costs," https://www.customerexperiencedive.com/news/klarna-ai-slash-customer-service-costs/748647/ ↩

3. Bloomberg, "Klarna Turns From AI to Real Person Customer Service" (2025-05-08), https://www.bloomberg.com/news/articles/2025-05-08/klarna-turns-from-ai-to-real-person-customer-service ↩

4. Salesforce News, "Salesforce Introduces New Flexible Agentforce Pricing" (2025-05-15), https://www.salesforce.com/news/press-releases/2025/05/15/agentforce-flexible-pricing-news/ ↩

5. CIO, "How to get AI agent budgets right in 2026," https://www.cio.com/article/4099548/how-to-get-ai-agent-budgets-right-in-2026.html ↩

6. a16z Enterprise Newsletter, "AI Is Driving A Shift Towards Outcome-Based Pricing," https://a16z.com/newsletter/december-2024-enterprise-newsletter-ai-is-driving-a-shift-towards-outcome-based-pricing/ ↩

7. Fortune, "This Sequoia partner thinks AI-enabled services are the new software" (2026-04-21), https://fortune.com/2026/04/21/services-are-the-new-software-sequoia-venture-capital-julien-bek-ai-native-eye-on-ai/ ↩

8. Stormy AI, "The Shift to Outcome-Based Pricing: A 2026 GTM Playbook," https://stormy.ai/blog/outcome-based-pricing-2026-gtm-playbook ↩

9. Sequoia Capital, "Insights from AI Ascent 2025," https://inferencebysequoia.substack.com/p/insights-from-ai-ascent-2025 ↩

10. Bessemer Venture Partners, "Mastering product-market fit: A detailed playbook for AI founders," https://www.bvp.com/atlas/mastering-product-market-fit-a-detailed-playbook-for-ai-founders ↩