Weill, Sebastian, Woerner & Benedict (MIT CISR) — Business Models in the AI Era
We share our predictions for how business models will evolve in the AI era, introducing a framework for AI-driven, real-time models. — MIT CISR’s own summary
TL;DR
A short MIT CISR research briefing (No. XXV-10, 16 October 2025) by Peter Weill, Ina M. Sebastian, Stephanie L. Woerner and Gayan Benedict. It redraws CISR’s 2018 Digital Business Models framework for AI. The prediction is that business models over the next five-plus years become outcome-oriented and enabled by autonomous AI. The new framework keeps a 2×2 shape but replaces both axes:
- Vertical: from knowledge of your end customer to action on behalf of customers. A firm either assists customers toward an outcome or represents them, achieving the outcome for them within guardrails.
- Horizontal: from business design (value chain or ecosystem) to business execution. Structured execution starts from a specified process that AI runs. Adaptive execution specifies only the outcome and lets advanced AI construct the process.
The four AI-era models are Existing+, Customer Proxy, Modular Curator and Orchestrator. The evidence for the old framework is longitudinal: 2,378 company observations from 2013 to 2025. The new framework is a prediction. Four researchers drafted it in 2025, one of them a former CIO, and tested it in senior executive meetings. The briefing reports no data on the new models.
The digital business models, 2013–2025
CISR placed each company in one of four digital business models in survey waves in 2013 (101 and 93 companies in two surveys), 2019 (1,311), 2022 (721) and 2025 (152). The questions were the same each time, and the framework comes from Weill & Woerner’s What’s Your Digital Business Model? (HBR Press, 2018).
| Model | What it is | 2013 | 2025 |
|---|---|---|---|
| Supplier | Sells via other companies (manufacturer; insurance via an agent) | 46% | 15% |
| Omnichannel | Combines digital and physical channels (retailer, bank) | 24% | 4% |
| Modular Producer | Plug-and-play product or service (payments provider) | 18% | 23% |
| Ecosystem Driver | Go-to destination in a customer domain, connecting customers with providers | 12% | 58% |
The share of companies that lead or take part in a digital ecosystem rose from 30% to 81% over the same period. In 2025, Ecosystem Driver was the only model with revenue growth above the industry average, by six percentage points. A customer domain is CISR’s term for a customer’s end-to-end need in an area such as home, mobility, energy or secure supply chain (Weill & Woerner, CISR briefing, September 2023). The briefing gives no significance tests and no figures for the other three models’ growth or profit.
The AI-era framework
Figure 1 of the briefing, transcribed from the image:
| Structured execution | Adaptive execution | |
|---|---|---|
| Represent the customer | Customer Proxy: the company achieves an outcome set by the customer with predefined processes. Revenue: achieving customer outcomes. Employees: advocate for the customer. | Orchestrator: the company achieves an outcome set by the customer through adaptive, AI-mediated company collaborations. Revenue: share of customer value. Employees: monitor and be accountable for the design, ethics and outcomes of the business model. |
| Assist the customer | Existing+: the company delivers products and services with predefined processes, augmented by AI. Revenue: selling products and services. Employees: make decisions, improve processes, do most of the work. | Modular Curator: the company delivers solo or bundled services created through adaptive, AI-mediated company collaborations. Revenue: share of customer service bundles. Employees: ensure the company’s products combine with other companies’ in real time. |
The figure carries two rows the body text leaves out: source of revenue and role of employees. Read across the quadrants, revenue moves from selling things to being paid for outcomes, and employees move from doing most of the work to being accountable for a system that does it. Each quadrant also gets a financial-services example in the body: AI-personalised advice (Existing+), AI managing a portfolio within set parameters (Customer Proxy), a bundle of investment, insurance and credit from several providers (Modular Curator), and a fully managed wealth solution that runs without customer input (Orchestrator).
Two kinds of human role. The briefing separates human in the loop (employees review and approve every decision before execution) from human at the helm (employees set goals, align incentives, define constraints and guardrails, and monitor outcomes). Structured execution has both. Adaptive execution keeps only the helm. This is a sharper division than most of the agent-oversight-and-delegation corpus draws, and it makes the oversight design a property of the business model rather than of the individual agent.
An Orchestrator example. Amazon’s beta Buy for Me feature buys products Amazon does not sell from other brands’ websites on the customer’s behalf. It is offered as an emerging Orchestrator. The same company is in court to keep outside shopping agents off its own site, as The Verge (Oct 2026) reports; the briefing predates that dispute. The Orchestrator quadrant assumes that other firms will let the orchestrating agent in, and the agentic-web sources show that access is now being negotiated, refused or litigated. The briefing does not discuss access at all.
Case: One New Zealand
One NZ, a telecommunications provider, is moving from traditional telco to “AI-driven company”. It deployed 15 AI use cases in 2024 that cleared a margin hurdle and aims for 50 by the end of 2025, 30 of them launched or under way. The case is placed across all four quadrants:
- Existing+ (where most initiatives sit). Knowledge agents resolve over 60% of FAQs and other queries from consumer prepaid customers and 40% from enterprise customers (in beta). Marketing agents (an SEO optimiser, a non-visual content agent) build customer audience segments 60% faster.
- Customer Proxy (moving toward). Service agents act with the customer’s authorisation: upgrading plans, raising disconnection requests, opening support tickets, still with humans in the loop and at the helm.
- Modular Curator (gearing up). During a major weather event, task-based agents checked power failures and cell status, estimated battery capacity and time until generator support was needed, forecast demand and recommended actions to decision-makers. The briefing says this took minutes instead of hours.
- Orchestrator (future). The same agents would optimise responses to power outages on their own, and marketing agents would create, deploy and adapt campaigns. The marketing team would specify goals, constraints and guardrails and then monitor outcomes.
The figures are the company’s own, reported without method. The case places most of today’s real activity in the least transformative quadrant. It also shows a firm placing different initiatives in different quadrants at once, which matches CISR’s footnoted caveat that companies can operate several business models even though the survey assigns each to one.
What the briefing asks leaders
Three questions: where on Figure 1 the company can create most value in the next three years; whether there is an advanced AI-enabled initiative in the company to learn from and scale; and what new capabilities it must build. The recommended use is as a shared language: circulate it to a group of senior leaders and discuss where the company stands. This is CISR’s usual move. The briefing is a framework for executive conversation, not a test of a hypothesis.
Dynamic-capabilities reading
strategic-renewal/business-model: the whole briefing. It names the value-creation and value-capture logics AI makes possible (outcome-based revenue, share of customer value, share of service bundles) and tracks a twelve-year shift from transactional Supplier models to multi-sided Ecosystem Drivers. This is the W&W renewal outcome described as a choice of quadrant.digital-sensing/digital-scenario-planning: the framework is a scenario of business models five-plus years out, built by a small expert group and tested with executives. Its three closing questions ask leaders to interpret that scenario and turn it into a strategy for their own firm.digital-transforming/navigating-innovation-ecosystems: the ecosystem share rising from 30% to 81%, and the two adaptive quadrants defined by AI-mediated company collaborations. Modular Curator and Orchestrator only work if the firm’s AI can combine its products with other companies’ in real time.
Neighbour sources
The closest neighbour is CISR’s own Enterprise AI Maturity model (Woerner, Weill, Sebastian, Káganer, via MIT Sloan, Oct 2025), released two weeks earlier by largely the same authors. The maturity model describes what a firm can do with AI. This briefing describes how it makes money with it. Stage 4, “creating and selling AI-augmented business services”, is the capability the adaptive quadrants presuppose. The two frameworks are not cross-referenced in the briefing.
Outcome-based revenue has several sources in the corpus, each from a different side:
- Nishar & Nohria (HBR, 2026), from the buyer’s side: Buy Outcomes as a firm-boundary choice.
- Warren (YC, 2026), for start-ups: the AI-native services company sells the result, a Customer Proxy.
- Sternfels (HBR IdeaCast, 2026), an incumbent’s measured version: outcome-underwriting at about a third of McKinsey’s revenue.
- Catlin (McKinsey Podcast, 2026), on why it matters: efficiency bolted onto the existing model (Existing+) gets competed away.
On the Orchestrator side, Collison (Browserbase Navigate, Oct 2026) makes the market-level case that AI discovery becomes the new aggregation point for commerce. Buy for Me is one merchant trying to be that aggregator.
Linked entities and concepts
- Entities: MIT CISR, Peter Weill, Ina Sebastian, Stephanie Woerner, Amazon
- Concepts: enterprise-ai-adoption, agentic-web, agent-oversight-and-delegation, dynamic-capabilities
- Synthesis: organizational-frameworks-for-ai-adoption, where the framework fills the business-model layer the stack lacked.
- Dangling (single-source mention, deferred per the second-source promotion rule): Gayan Benedict (Industry Research Fellow, MIT CISR); One New Zealand Group Ltd.
Scope and reliability
What was read. The full briefing as published on the CISR web page, which shows the whole text without login, with its four footnotes and Figure 1 (an image, transcribed and checked against it). The audio transcript and the separate Talking Points and Framework pages were not fetched.
Two kinds of claim, with different support. The 2013–2025 prevalence shifts and the Ecosystem Driver growth premium rest on survey data. The waves vary greatly in size (152 companies in 2025 against 1,311 in 2019), and the briefing does not say whether the companies are the same across waves. The AI-era framework is a prediction by four researchers, tested in executive meetings. Nothing in the briefing measures how many firms sit in each new quadrant or how they perform.
Vendor-free, but member-funded. MIT CISR is a member-funded consortium, and the briefings are written for member organisations. One NZ’s figures are company-reported. The briefing is a framework for discussion, and this page treats it that way.