Jim Rowan，美国驻德勤的AI主管-访谈系列

面谈

Jim Rowan，美国驻德勤的AI主管-访谈系列

在 Google 上的首选来源中添加 Unite.AI

Jim Rowan是德勤公司的AI主管，也是德勤咨询有限公司的负责人，帮助各组织将AI的抱负转化为商业成果。 他的职业生涯跨越了企业技术，分析，数据，和数字化转型，拥有跨越生命科学，保健和电信的经验。 他在德勤公司的战略与分析以及AI & Data实践中担任过领导角色，他的工作包括培养人才和能力组织需要采用AI.

德勤公司是一个提供咨询、审计和保证、税收和咨询服务的专业服务组织。 它的AI工作带来共同战略，数据，技术，人才和治理，帮助企业在业务中采用并推广AI. 在数字商业中，这项工作包括为AI agents准备产品信息和企业系统，将客户的经验连接到不同渠道，以及确定自动化互动和交易所需的身份、授权和信任控制。

在德勤公司工作了20多年，从这个角度来说，你什么时候明白AI agents不仅会改变公司的内部运作方式，而且还会成为企业与其客户之间的中介？

在过去几年里，我亲眼看到AI在帮助各组织将业务目标转化为过去似乎难以想象的可衡量的结果的同时，取得了第一手的进步。 我对AI的理解基于我的数据、过程和技术背景。 我开始建立德勤公司的能力，专注于AI的成功所需人才

在目前的情况下，各组织正在利用AI agents以多种方式改变其数字商业战略：提高品牌知名度，利用个性化引擎加强对话，以及协调跨越物理、数字和社会接触点的一致的全渠道经验。 这个inflection点的出现正值品牌正在导航一个动荡的市场，为代理商创造了机会，帮助他们以更大的速度，相关性和一致性应对他们的客户。

您已经讨论过公司是否有必要为“代理搜索优化”做准备。 您如何定义这个概念，如何优化AI agent 与传统的 SEO 或更新的方法优化 AI 答引擎的内容？

传统上，对消费者注意力的在线竞争是由搜索引擎优化(SEO)定义的，并假设在查询的另一端，人类正在阅读，评价和决定购买。 随着消费者转向使用AI驱动工具进行研究，基因引擎优化(GEO)一直是推出。 Agentic Search Optimization是GEO的一种渐进式迭代，它考虑AI如何不再仅仅推荐产品或服务，而是对购买决定采取行动。 这种转变意味着品牌不仅必须优化基因AI，它主要为终端用户提供建议，还必须优化代理AI，它是为执行而构建的，可以协调系统，执行工作流程和触发交易。

当AI agent被要求研究、比较和短名单产品或服务时，你期望它最重视什么信号？结构化的产品数据，定价，审查，第三方覆盖，品牌权威，或一些全新的信号是否会变得更加重要？

AI agent的考虑设定基于能见度和相关性，并将显示对产品、服务、定价和数据可以访问、机器可读和可立即核实的品牌的偏好。 跨平台的一致性，评级/审查以及品牌权威仍然重要，但这些信号也必须转化为数据agent可以解析通过。 个人背景也很重要，因为代表客户工作的agents需要可靠地访问此人的喜好、权限和事先互动，以产生最相关的建议。 除了信号本身外，各组织采取行动使其数据可读性、可理解性、可信赖性和可量化的agent比较性，其速度将是该品牌的区别。

一项挑战是，最终可能会有成千上万的消费者和企业agents建立在不同的模型上，并与不同的数据来源相连。 公司如何在分散的生态系统中保持准确的代表性和可发现性，而不试图优化每个个体的agent?

组织的重点应当是创建一个可信赖的，机器可读的基础，任何agent都可以访问和解释，而不是优化单个agents. 这意味着通过权威来源公布关于产品、定价和政策的可核查、可靠和最新的信息。 agents将开始代表消费者行事，进一步需要能够验证身份、确认资格和保证交易安全的系统。 在目前零碎的agent生态系统中，让信息最容易理解和获取的品牌将上升到搜索的顶端。 然而，公司应该避免围绕单一协议构建，因为这些协议仍在出现，尚不清楚哪些协议将变得占主导地位，因此设计互操作性的重要性以及随着生态系统的发展迅速通过或转换协议的能力。

如果商业从企业对消费者演变为企业对agent企业，在技术层面上有什么变化？公司是否需要重新考虑网站，API，产品目录，身份系统，或其他基础设施，以便agents能直接与它们互动？

是的，各组织需要重新考虑和重新构思其目前的系统，以建立能够实现B2B代理商业的正确的技术架构。 这不会在一夜之间发生，今天，发现和支付能力的发展速度超过了决策阶段，在决策阶段，企业仍在培养信心，让agents在既有的护栏内作出选择。 任何首席信息干事、首席信息干事或首席信息干事的重点都应是提供可扩展、低常态的一体化平台，并辅之以强大的技术和集成者伙伴关系。 领导人应该花时间确定哪些系统可以连接到agents，找出被封锁的系统，并在封闭环境中测试简单的任务，而不是对系统进行全面的检修。 保护拥有交易的人类也很重要，因此公司应创造灵活的道路，使客户能够轻松地在agent和人主导的互动之间移动，而不是削弱人类追求自动化的经验。 随着时间的推移，各组织可以逐步建立一个管理多个agents、升级或集成的平台。 安装用于agent访问的遗留系统，并部署实时监测的高级安保。

传统数字营销围绕排名，印象，点击，网站流量进行了大幅优化。 如果agents在人们访问网站之前越来越多地进行研究，那么公司应该用什么新的衡量标准来了解它们在代理买主行程中是否真正可见和具有竞争力？

从历史上看，衡量成功的重点是吸引消费者的注意力，并最终吸引他们的信任。 由于agents承担了更多的购买行程，各组织需要新的方法来衡量其品牌是否显眼，是否相关，最终是否被选中。 这包括将品牌的agent开发与性能测量联系起来，评分agent是否表面该品牌，确认符合条件的报价并成功完成交易。 品牌应关注其买家与竞争者，以确定投资或经验变化，跟踪客户如何发现产品，他们离开何处或停留何处。 这些衡量标准不仅可以帮助品牌区分其价值命题，还可以通过可量化的结果证明ROI，增强客户的忠诚度和业务优异度。

AI agents可以总结一个公司，与竞争者进行比较，并有可能在公司没有直接控制其代表性的情况下提出建议。 品牌应如何监测不准确或过时的信息，如何建立agents能够可靠使用的权威性真相来源？

与任何其他行业一样，对产品数据进行审计对于保持跨渠道的代理性发现至关重要。 品牌还应当考虑将模型背景协议(MCP)作为关键促进器，这创造了一种控制环境，使数据，工具和背景安全地与外部购物agents和平台共享。 将结构化的数据和MCP结合，可以实现agent对agent的相互作用，提高互操作性，帮助准确地代表品牌信息，并确保其得到适当的优先考虑。 同样，由于协议仍在出现，而且尚不清楚哪些协议会变得最受欢迎，公司应建立灵活性，以便随着生态系统的发展，它们可以迅速与新标准连接、相互转换和采用。

agents从单纯推荐产品转向采取行动，如要求报价、谈判条件或购买，身份、授权和信任基础设施将变得多么重要？企业如何区分合法客户agent与其不应信任的自动化系统？

Agent至agent商业带来了新的风险，因为没有人类的参与，交易可以迅速进行；因此，在整个过程中，人类干预和移徙至关重要。 代理商业交易需要新型的预防欺诈措施、战略性财务控制和专门为机器启动的活动建立的目标明确的报告框架。

在近期内，各组织应将AI agents视为雇员，规定责任、有管理的准入和明确的问责制。 他们还需要协议来核查合法的agents，授权支付和执行交易阈值，API许可和其他运行的护栏。

广告和有偿安置也有可能跟踪消费者进入代理生态系统。 当AI agent代表某人推荐产品时，你如何看待商业激励机制的演变，以及人们相信这些建议需要什么样的透明度？

美国广告商预计到2029年将在AI搜索广告上花费约260亿美元。 随着企业决定通过自有渠道提供的促销和定价是否在代理环境下得到尊重，商业激励势必需要演变。 这种演化的部分内容要求agents如何与定价和促销互动的明确规则。 没有护栏，agents就可以根据交易时可用的定价进行采购，可能引发意想不到的促销和定价等级，并造成差值泄漏。 企业应执行符合全球隐私和数据标准的政策，并与每个客户的人力控制、预先核准的参数保持一致。

尽管自动化程度不断提高，但你们强调，人的判断仍将是采购决定的核心。 你从哪里看到决定之间的界限。.....消费者可以安心地委托给agents 和人们仍然想要的决定。..

本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。

Interviews

Jim Rowan, US Head of AI at Deloitte – Interview Series

Add Unite.AI to your preferred sources on Google

Jim Rowan is US Head of AI at Deloitte and a principal in Deloitte Consulting LLP, helping organizations translate AI ambitions into business outcomes. His career spans enterprise technology, analytics, data, and digital transformation, with experience across life sciences, health care, and telecommunications. He has held leadership roles in Deloitte’s Strategy & Analytics and AI & Data practices, and his work includes developing the talent and capabilities organizations need to adopt AI.

Deloitte is a professional services organization providing consulting, audit and assurance, tax, and advisory services. Its AI work brings together strategy, data, technology, talent, and governance to help businesses adopt and scale AI across their operations. In digital commerce, that work includes preparing product information and enterprise systems for AI agents, connecting customer experiences across channels, and establishing the identity, authorization, and trust controls needed for automated interactions and transactions.

You’ve spent more than two decades at Deloitte working across enterprise technology, analytics, data, and now AI. From that vantage point, when did it become clear to you that AI agents would not only transform how companies operate internally, but also become an intermediary between businesses and their customers?

Over the past few years, I’ve seen AI advance firsthand while helping organizations translate business goals into measurable outcomes that once seemed unimaginable. My background in data, processes and technology has been foundational to my understanding of AI. I started building Deloitte’s capabilities focused on the talent needed to be successful with AI, such as our AI Academy that creates curriculum for both our employees and clients.

In the current landscape, organizations are using AI agents to transform their digital commerce strategies in multiple ways: increasing brand visibility, using personalization engines to enhance conversation, and orchestrating consistent omnichannel experiences across physical, digital and social touchpoints. This inflection point comes at a time when brands are navigating a volatile market, creating an opportunity for agentic commerce to help them respond with greater speed, relevance and consistency for their customers.

You’ve discussed the need for companies to prepare for “Agentic Search Optimization.” How do you define that concept, and how is optimizing for an AI agent fundamentally different from traditional SEO or the newer approaches to optimizing content for AI answer engines?

Traditionally, online competition for consumer attention has been defined by Search Engine Optimization (SEO) and assumes that at the other end of a query, a human is reading, evaluating and deciding their purchase. As consumers shift towards using AI-powered tools to conduct research, Generative Engine Optimization (GEO) has been introduced. Agentic Search Optimization is a progressive iteration of GEO that considers how AI no longer just recommends products or services but acts on those purchasing decisions as well. That shift means brands must optimize not only for generative AI, which primarily provides recommendations to the end user, but also for agentic AI, which is built for execution and can coordinate systems, execute workflows and trigger transactions.

When an AI agent is asked to research, compare, and shortlist products or services, what signals do you expect it to value most? Will structured product data, pricing, reviews, third-party coverage, brand authority, or some entirely new set of signals become more important?

An AI agent’s consideration set is developed based on visibility and relevancy and will show preference to brands whose products, services, pricing and data are accessible, machine-readable and instantly verifiable. Consistency across platforms, ratings/reviews and brand authority still matter, but these signals must also translate into data an agent can parse through. Personal context will matter too as agents working on customers’ behalf will need reliable access to that person’s preferences, permissions and prior interactions to generate the most relevant recommendations. Aside from the signals themselves, the speed at which organizations can act on making their data readable, understandable, trustworthy and quantifiable in terms an agent can compare, will be a differentiator for the brand.

One challenge is that there may eventually be thousands of consumer and enterprise agents built on different models and connected to different data sources. How can a company remain accurately represented and discoverable across that fragmented ecosystem without trying to optimize for every individual agent?

Organizations focus should be on creating a trusted, machine-readable foundation that any agent can access and interpret rather than optimizing for individual agents. This means publishing verifiable, reliable, current information about products, pricing and policies through authoritative sources. Discoverability alone likely won’t be enough, as agents will begin acting on consumers’ behalf, furthering the need for systems that allow identity verification, eligibility confirmation and secure transactions. Across a now fragmented agent ecosystem, the brand that makes their information the easiest to understand and obtain will rise to the top of search. However, companies should avoid building around a single protocol because these are still emerging and it’s not yet clear which will become dominant thus the importance of designing for interoperability and the ability to adopt or switch protocols quickly as the ecosystem evolves.

If commerce evolves from business-to-consumer toward business-to-agent-to-business, what changes at the technical level? Will companies need to rethink websites, APIs, product catalogs, identity systems, or other infrastructure so that agents can interact with them directly?

Yes, organizations will need to rethink and reimagine their current systems to build the right technical architecture capable of achieving B2B agentic commerce. It won’t happen overnight; it will take time due to the level of work and degree of change involved. Today, discovery and payment capabilities are advancing faster than the decision-making stage where businesses are still developing the confidence to let agents make choices within established guardrails. The focus for any CIO, CTO or CISO should be delivering scalable, low-latency integration platforms supported by strong technology and integrator partnerships. Rather than a complete system overhaul, leaders should spend time determining which systems can connect to agents, identifying ones that are blocked and testing simple tasks in contained environments. Preservation of humans owning the transaction is important too so companies should create flexible paths in which the customer can move easily between agent-led and person-led interactions rather than weakening the human experience in the pursuit of automation. Over time, organizations can progress to building a platform that manages multiple agents, upgrading or connecting legacy systems for agent access, and deploying advanced security with real-time monitoring.

Traditional digital marketing has been heavily optimized around rankings, impressions, clicks, and website traffic. If agents increasingly conduct the research before a person ever visits a website, what new metrics should companies use to understand whether they are actually visible and competitive in the agentic buyer journey?

Historically, measurement of success focused on capturing consumers’ attention and ultimately their trust. As agents assume more of the buying journey, organizations will need new ways to measure whether their brands are visible, relevant and ultimately selected. This includes linking branded agent development to performance measurement, scoring whether an agent surfaces the brand, recognizing eligible offers and successfully completing a transaction. Brands should focus on their buyers vs. their competitors to determine investments or experience changes, tracking how customers discover products, where they leave or where they stay. These metrics can not only help brands differentiate their value propositions but can also prove ROI with quantifiable outcomes that reinforce customer loyalty and operational excellence.

AI agents can summarize a company, compare it with competitors, and potentially make recommendations without the company having direct control over how it is represented. How should brands monitor for inaccurate or outdated information and establish an authoritative source of truth that agents can reliably use?

Like any other industry, auditing product data is critical to maintaining agentic discoverability across channels. Brands should also consider Model Context Protocol (MCP) as a key enabler, which creates a controlled environment where data, tools, and context are shared securely with external shopping agents and platforms. Combining structured data and MCP can enable agent-to-agent interactions, improve interoperability and help represent brand information accurately and ensure it is prioritized appropriately. Again, because protocols are still emerging and it is not yet clear which will become most popular, companies should build for flexibility so they can connect to, switch between and adopt new standards quickly as the ecosystem evolves.

As agents move from simply recommending products toward taking actions such as requesting quotes, negotiating terms, or making purchases, how important will identity, authorization, and trust infrastructure become? How does a business distinguish a legitimate customer agent from an automated system it should not trust?

Agent-to-agent commerce presents new risks because transactions can happen quickly without human involvement; therefore, making human intervention and migration throughout the process critical. Agentic commerce transactions require new types of fraud prevention measures, strategic financial controls and targeted reporting frameworks built specifically for machine-initiated activity.

In the near term, organizations should treat their AI agents like employees, with defined responsibilities, managed access, and clear accountability. They will also need protocols to verify legitimate agents, authorize payments and enforce transaction thresholds, API permissions and other operational guardrails.

There is also the possibility that advertising and paid placement will follow consumers into the agentic ecosystem. How do you see commercial incentives evolving when an AI agent is recommending products on someone’s behalf, and what transparency will be necessary for people to trust those recommendations?

We already see capital being reallocated to agentic commerce, with U.S. advertisers projected to spend about $26 billion on AI search ads by 2029. It’s inevitable that commercial incentives will need to evolve as enterprises decide whether promotions and pricing offered through owned channels are honored in agentic environments. Part of this evolution requires explicit rules for how agents interact with pricing and promotions. Without guardrails in place, agents could execute purchases based on pricing available at the moment of transaction, potentially triggering unintended promotion and pricing tiers and creating margin leakage. Enterprises should implement policies that align with global privacy and data standards, and align with human-controlled, preapproved parameters for each customer.

Despite increasing automation, you’ve emphasized that human judgment will remain central to purchasing decisions. Where do you see the boundary forming between decisions consumers will comfortably delegate to agents and decisions where people will still want t

Auto-fetched and translated by AI Atlas. All rights belong to the original author.