Artificial intelligence is reshaping the competitive landscape of the information services and business services sectors. The uniqueness of data assets, control over workflows, and monetization capabilities will determine which enterprises emerge victorious in this transformation.
According to information from the Zhui Feng trading desk, Goldman Sachs released a research report after completing its third annual Silicon Valley AI field trip. Discussions with AI companies, venture capital firms, and academic researchers from Stanford University and UC Berkeley reinforced four core judgments: the value of proprietary information will continue to rise; AI applications are extending from information delivery to workflow execution; physical services and education platforms are more resilient than business models reliant on large-scale white-collar labor; and monetization paths will expand from seat-based subscriptions to usage-based, transaction-based, product premium, and outcome-based pricing models.
Research Background: First-Hand Silicon Valley Insights Form the Basis for Analysis
Goldman Sachs held its third annual Silicon Valley AI field trip on August 18-19. Participants included AI companies such as Daloopa, Clio, Harvey, Corgi, Moody's, Vercel, and ClickHouse, as well as venture capital firms including Lightspeed Venture Partners, Kleiner Perkins, Menlo Ventures, and Index Ventures. Dedicated sessions were also held with researchers from Stanford University and UC Berkeley/UCSF.
Key topics examined during the trip included: the accelerating evolution of model capabilities, the distribution of workloads between open and frontier models, the transition of AI agents from assistive tools to workflow executors, and sources of durable competitive differentiation across dimensions such as data, domain expertise, customer context, and workflow ownership. Additionally, the research covered AI monetization, verification and accountability, cybersecurity, physical AI, and infrastructure requirements for agent-native workloads.
Theme One: Proprietary Data Builds the Clearest Information Moat
Goldman Sachs believes that model capabilities continue to improve, yet they still rely on external information they cannot generate themselves. This benefits enterprises that possess unique, continuously updated datasets that competitors find difficult to replicate. Meanwhile, the ability to embed this information into customer AI agents and applications can expand usage beyond traditional user interfaces. The risk is that new interfaces may diminish the value of existing retrieval and display products, shifting the value center to the application layer that controls the customer experience.
The positive implications are clearest for MCO. Moody's demonstrated during the research how insurance materials submitted via PDFs, spreadsheets, and free-form documents can be converted into standardized underwriting data within seconds, compared to the hours or even days typically required for manual processing. The platform can also parse non-standard policy language and complete catastrophe loss analysis and reserve calculations through a natural language interface. These workflows rely heavily on MCO's proprietary data, entity relationship graphs, catastrophe models, and domain expertise, making them difficult for general-purpose applications to replicate. Smart APIs, MCP servers, and agent products provide pathways for customers to access this intelligence within their own workflows, constituting incremental subscription and usage-based revenue sources.
In contrast, FDS faces greater pressure. Daloopa demonstrated how public company disclosure documents can be used to automate financial data collection and model maintenance, information that is equally accessible to FDS and other data providers. The Daloopa platform maps disclosure information to customer-customized Excel models and provides source links for each data point. Since FDS's core financial data largely derives from the aggregation and standardization of public or third-party sources, its data is highly replicable, and the value of workstation features may consequently come under pressure.
SPGI possesses differentiated assets including Market Intelligence, ratings, commodity benchmarks, indices, private markets, and specialized pricing, with multiple monetization paths in machine-consumption pricing and new workflow products. Goldman Sachs assigns a "positive/mixed" implication, with the key variable being how AI transforms user engagement patterns with Capital IQ Pro. IT (Gartner) faces greater content risk—Harvey and Simile both demonstrated that AI can transform standardized research content into reusable agent workflows, enabling enterprises to generate new analysis without commissioning individual research projects. The scarcity of standardized research and training materials may therefore decline, leading Goldman Sachs to assign a "negative/mixed" implication.
Theme Two: Workflow Control Determines Value Attribution
Goldman Sachs observed a key evolution in AI during the research: Harvey and Clio demonstrated how AI has advanced from retrieving information to drafting documents, coordinating tasks, and completing more complex professional workflows. Whoever controls the platform where customers complete their work may capture economic value that extends beyond a single data input provider.
TRI possesses a strong workflow position, combining authoritative Westlaw legal content with CoCounsel to extend into legal research, drafting, and document review workflows. Clio's $1 billion acquisition of vLex in 2025, securing a major legal content library spanning over 110 countries and more than 1 billion legal documents, confirms the strategic value and construction difficulty of global legal content assets. However, if Harvey or Clio controls the customer interface and workflows, TRI could be reduced to an underlying content provider—even if demand for its content remains strong. Goldman Sachs assigns TRI a "positive/mixed" implication.
FICO and EFX have the clearest positive implications. Stanford researchers noted during the field trip that the primary obstacle preventing enterprise agents from entering production environments is not model capability but unresolved accountability mechanisms—processes that are well-defined, verifiable, and revocable will be deployed first. This framework aligns closely with FICO's capabilities in decision governance. EFX's The Work Number, relying on employer-contributed employment and income records, is the most difficult to replicate differentiated asset. As AI agents evolve from generating recommendations to initiating decisions, verification needs will occur earlier and more frequently.
VRSK, however, faces some moat pressure. Corgi demonstrated how an AI-native commercial insurance company can automate underwriting processes without heavily relying on VRSK data. Corgi noted that most U.S. insurance pricing methodologies, filed rates, and loss ratio data are already publicly disclosed through state regulators, allowing AI-native companies to train on this information. VRSK retains advantages including proprietary claims history data, catastrophe models, and deep embedding in underwriter workflows, but Corgi challenges the assumption that all traditional insurance data becomes harder to replace in the AI era. Goldman Sachs assigns VRSK a "mixed/positive" implication.
ULS (UL Solutions) presents longer-term option value. The Stanford research session cited practices in safety-critical industries such as aviation, raising the potential demand for an "AI agent version of SOC 2," which aligns closely with ULS's expertise in testing, inspection, and certification. However, uncertainty remains regarding standard-setting, market demand, and the timeline for revenue contribution.
Theme Three: Physical Services and Education Platforms Show Greater Resilience
The risk of AI replacing white-collar work places direct pressure on labor-driven business models, but Goldman Sachs believes physical services and education platforms are in a more favorable position.
In the staffing sector, RHI faces the greatest impact. Daloopa's demonstration directly revealed RHI's exposure in accounting, finance, administrative, and junior technical roles. AI automation of data collection, model maintenance, and structured knowledge processing will reduce demand for these positions, and client insourcing of hiring may further compress staffing volumes. Emerging roles in data engineering, cybersecurity, and model implementation may struggle to close the gap, leading Goldman Sachs to assign a "negative/mixed" implication. MAN (Manpower) gains relative protection through a higher proportion of blue-collar positions, but its Experis brand still faces pressure from automation of routine IT support and analytical work.
In education, MH has a clear AI revenue opportunity. More than 100 million paid course licenses form the foundation for distributing AI-driven course content. Early adoption of AI products in assessment, learning assistance, clinical simulation, and teacher tools has already reached 7.5 million users. MH has identified fixed markups, question-based pricing, and token consumption as potential monetization models, with a potential licensing premium of 5% to 15%.
In the physical services sector, ADT, CTAS, and ROL all received positive implications. ADT can apply AI to monitoring and event triage, supporting premium service tiers, customer retention, and video analytics products. CTAS can strengthen cross-selling and service response capabilities through route data and customer history. ROL is expected to leverage property-level pest activity data and Stanford world model research frameworks to explore predictive detection and premium monitoring products. The common characteristic of these three companies is that AI enhances the value-added of services without replacing the core physical activities customers purchase.
Theme Four: Monetization Paths—Diversified Pricing Logic Beyond Seats
Goldman Sachs identified three primary revenue paths and emphasized that AI monetization requires combining differentiated assets with quantifiable customer value, rather than simply converting existing seat revenue into new pricing units.
Usage and transaction-based pricing logic holds that data requests, verification events, and automated decision counts generated by AI agent applications will far exceed those of manual software operation. MCO and SPGI can monetize machine access to proprietary information. FICO and EFX should benefit from increased frequency of automated decisions and verifications. The main risk is that usage growth remains bundled within existing enterprise contracts rather than creating incremental spending.
For product premiums and outcome-based pricing, TRI can charge for broader legal workflows through CoCounsel. MH has confirmed three models: fixed markups, question-based pricing, and token consumption. ADT can develop premium monitoring and video analytics products. The durability of these models may exceed pure token billing because customers can directly link spending to completed tasks or improved outcomes.
In terms of infrastructure demand, IRM has a more direct revenue opportunity. Vercel and ClickHouse noted during the field trip that agent applications generate request volumes, token consumption, and telemetry data far exceeding traditional software, directly supporting data center leasing and hyperscale asset lifecycle service demand. Since customers already pay for infrastructure and information processing, IRM does not need to establish new AI pricing models; execution capability and capital investment intensity are the more critical constraint variables.
Regarding revenue disruption risk, FDS, IT, RHI, and PBI face the timing gap risk where existing revenue comes under pressure before new product monetization materializes. Goldman Sachs believes all four companies may launch competitive AI products, but the potential timing mismatch between disruption and monetization supports a more cautious investment implication.