Airbnb Rejects AI Vendors Over Volatile Token Costs, Legal Risks
Ai4
Enterprise buyers are dropping software vendors in favor of internal AI prototyping unless providers eliminate volatile consumption pricing and mitigate unprivileged compliance liabilities.
## Background
Published on YouTube by Ai4 on August 19, 2026, this presentation features John Polito, Deputy General Counsel for Commercial & Technology at Airbnb, Inc. The video explores how software companies and technology consultants can position themselves as genuine, good-faith partners in an enterprise landscape where every vendor solution is heavily scrutinized against internal AI capabilities and frontier LLMs.
## TL;DR
John Polito details the intense internal pushback enterprise software faces as business and engineering teams increasingly question fixed SaaS subscriptions in favor of generative AI and 'vibe coding.' He highlights major vendor missteps that derail deals—including retroactively demanding customer data training rights, imposing unpredictable token-based pricing, and generating discoverable, non-privileged compliance records. In contrast, winning vendor partners adapt by offering realistic multi-week proof-of-concept trials, aligning pricing with value realization milestones, and collaborating closely with internal legal and finance champions to frame adoption around strategic enablement rather than simplistic headcount cuts.
## Atmosphere & style
The presentation carries a candid, pragmatic, and unvarnished tone, driven by Polito's dual background in software engineering and corporate technology law. Bypassing generic AI hype and standard vendor buzzwords, the talk feels like a behind-the-scenes masterclass in enterprise procurement and legal risk management. Delivered through constructive real-world war stories and anonymized RFP case studies, it balances firm boundary-setting with genuine empathy for how vendors can help internal corporate teams succeed.
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=4s) In an era where internal stakeholders challenge software costs against frontier AI models, successful vendors must equip internal support teams (legal, procurement, finance) with clear narratives to defend external software investments.
- [01:53](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=113s) Software renewals face intense scrutiny from internal 'vibe coding' and DIY engineering skepticism, with vendors jeopardizing partnerships by demanding customer data training rights or pushing volatile pricing.
- [03:43](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=223s) Although generative AI makes internal prototyping fast and easy, scaling to enterprise-grade production still requires battle-tested vendor infrastructure with formal security, privacy, and contractual SLAs.
- [05:12](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=312s) Standard one-week synthetic POCs fail for deep infrastructure and analytics workflows; vendors must adapt evaluation timelines and datasets to realistic operational constraints.
- [07:26](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=446s) Enterprise finance requires predictable budgets, making open-ended pay-as-you-go token pricing unviable for core infrastructure compared to structured tiers and deferred billing milestones aligned with value delivery.
- [10:42](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=642s) Enterprise AI search and indexing tools present significant legal exposure regarding attorney-client privilege and confidentiality if strict need-to-know access boundaries are not preserved.
- [13:36](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=816s) Implementing interactive AI intake validation absorbs the friction of compliance enforcement upfront, converting internal support teams from bureaucratic bottlenecks into strategic enablers.
- [15:30](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=930s) Automated compliance tools that generate definitive, unprivileged declarations of legal non-compliance create severe litigation risks, making vendor willingness to use advisory phrasing a prerequisite for enterprise adoption.
- [16:49](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=1009s) High-value vendor partnerships are built on strategic empathy, framing product ROI around enablement and organizational impact rather than blunt, morale-damaging headcount reduction narratives.
## Detailed walkthrough
### [00:04](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=4s) Navigating Vendor Relationships and Internal Support at Airbnb
John Polito, Deputy General Counsel for Commercial and Technology at Airbnb, oversees the commercial legal team managing all contracts, AI governance, and general technology legal support for over 3,000 engineers and data scientists. Operating as the primary interface layer between external vendors or consultants and Airbnb's internal technology organization, Polito observes recurring patterns across incoming software licenses, proofs of concept (POCs), and service agreements. Offering the standard lawyer disclaimer that his remarks reflect personal perspectives rather than official company statements, he bypasses standard conference talking points about measuring ROI, defining KPIs, and preventing rogue autonomous agents. Instead, his focus centers on how vendors can effectively collaborate with internal support functions—such as legal, procurement, and finance—to help internal champions successfully justify and implement external technology.
The best vendor partners recognize that in-house support teams must repeatedly sell and defend software expenditures to internal stakeholders. While support teams rarely receive adequate recognition, a vendor who provides actionable strategies and clear value narratives makes their internal buyers look good. In 2026, every outside expenditure faces intense scrutiny, with internal stakeholders continually asking why a capability cannot simply be built in-house using frontier models or existing enterprise tools. Polito notes that while technology has advanced, bespoke software, specialized consulting, and seasoned domain expertise remain vital. Drawing an analogy to the six classical simple machines, he emphasizes that combining every primitive tool into a DIY internal solution rarely produces an effective result, though the vendor landscape must adapt to heightened internal cost consciousness.
[01:53](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=113s) The "Vibe Coding" Challenge and Heightened Renewal Scrutiny
The modern procurement environment is dominated by constant internal pushback questioning whether outside software can be replaced by internal DIY engineering or ad-hoc generative coding ("vibe coding"). This skepticism does not originate solely from technical teams; business and operations units frequently challenge fixed software costs, asking whether internal upskilling and AI assistants can eliminate external subscriptions. As a result, software contract renewals are subjected to unprecedented levels of examination, where buyers must continually justify why an existing tool should not be deprecated in favor of internal automation.
```mermaid
flowchart TD
accTitle: Vendor Evaluation Pressures in 2026
subgraph S1["External Pressures"]
A["Vendor Demands<br/>• Data training rights<br/>• Volatile token pricing"]
end
subgraph S2["Internal Scrutiny"]
B["Executive & Business Questions<br/>• 'Why not vibe code this internally?'<br/>• Up-skilling ROI vs. Fixed SaaS"]
end
subgraph S3["Outcome"]
C{"Renewal Decision"}
C -->|"Poor Alignment"| D["Churn / Canceled Contract"]
C -->|"Collaborative Partner"| E["Expanded Enterprise Adoption"]
end
A --> C
B --> C
```
Vendor behaviors during renewals often dictate whether a multi-year partnership survives. Polito highlights two major missteps that consistently fail in current negotiations:
- **Demanding training rights on customer data:** Long-standing vendors returning at renewal to impose revised terms allowing them to ingest enterprise data to train their proprietary foundation models.
- **Introducing volatile cost structures:** Attempting to retroactively swap predictable pricing for open-ended, consumption-based token models that destabilize internal budgeting.
[03:43](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=223s) Proofs of Concept vs. Internal Prototyping: Crossing the Production Chasm
At the inception of any new technology initiative, enterprise teams grapple with whether to run a formal vendor Proof of Concept (POC) or bypass procurement entirely by spinning up an internal prototype using generative chatbots. Prototyping offers immediate gratification—internal teams can start within hours using their real production data without waiting for infosec reviews, privacy assessments, or legal contract negotiations. However, moving an experimental prototype into enterprise-grade production remains an arduous engineering endeavor. Polito, drawing on his background as a software engineer prior to entering law, notes that while junior employees might assume rolling an internal prototype into production is trivial, mature enterprises still depend on battle-tested vendor solutions that back their reliability with explicit contractual and reputational guarantees.
| Approach | Advantages | Disadvantages & Risks |
|:--- |:--- |:--- |
| **Internal Fast Prototyping** | • Instant start (no legal/infosec delays)<br/>• Uses real internal data immediately<br/>• Low upfront capital cost | • Massive friction transitioning from prototype to production<br/>• Lacks enterprise security, governance, and scale<br/>• Unguaranteed long-term reliability |
| **Vendor Proof of Concept (POC)** | • Battle-tested infrastructure<br/>• Contractual and reputational SLAs<br/>• Clear pathway to supported production | • Heavy upfront infosec, privacy, and legal reviews<br/>• Synthetic data limitations during initial trials<br/>• Requires structured time commitment from staff |
[05:12](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=312s) The Pitfall of the One-Week Synthetic POC: A Case Study
A standard offering from software sales teams is the "no-cost, one-week proof of concept." While brief, standalone evaluations work well for modular utilities like image generation that require minimal infrastructure integration, they fail completely for deep workflow and infrastructure solutions like Contract Lifecycle Management (CLM) platforms. A one-week trial inevitably relies on synthetic data and forces internal teams to test the software during evenings and weekends on top of their full-time operational responsibilities.
Polito illustrates this dynamic through a recent competitive Request for Proposal (RFP) for an analytics-heavy enterprise platform:
- **The Inflexible Vendor:** Insisted strictly on a standard one-week POC using synthetic data. Because a single week of synthetic usage is statistically meaningless for evaluating analytics pipelines and imposed an unmanageable burden on the evaluation team, this vendor was disqualified.
- **The Adaptive Vendor:** Initially proposed the same one-week synthetic structure, but actively listened when Airbnb explained the operational limitations. They pivoted to an eight-week evaluation window, supplied pre-populated synthetic datasets, and provided dedicated training on the analytics engine. This enabled the internal team to generate authentic mock reporting for executive leadership, securing the vendor a long-term enterprise contract.
[07:26](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=446s) AI Pricing Realities: Token Volatility vs. Budget Predictability
Enterprise finance and procurement teams require cost certainty. In 2026, many vendors whose underlying technology incurs non-deterministic backend Large Language Model (LLM) API expenses attempt to pass raw consumption fees directly to clients through pay-as-you-go (pay-go) token pricing. Polito notes that pay-go pricing is virtually impossible to sell internally to corporate finance, as budgetary approvals require fixed tiers and predictable schedules. While Airbnb occasionally leverages pay-go contracts for low-risk, disposable testing tools that can be deactivated instantly if performance falters, core infrastructure requires structured, predictable spend.
```mermaid
flowchart LR
accTitle: Intelligent Gatekeeping Workflow Redesign
subgraph Phase1["Legacy Failing Workflow"]
direction TB
A1["Engineer Submits<br/>Incomplete Ticket via Slack"] --> B1["Days of Silence"]
B1 --> C1["Reviewer Rejects:<br/>'Request lacks 10 required items'"]
C1 --> D1["Frustrated Engineer<br/>& Antagonistic Dynamic"]
end
subgraph Phase2["AI-Enabled Collaborative Workflow"]
direction TB
A2["Engineer Drafts<br/>Request in Slack"] --> B2["Interactive AI Validation Layer<br/>Prompts for 9-10 required specs"]
B2 --> C2["Fully Populated Ticket Reaches<br/>Legal / Infosec Review"]
C2 --> D2["Support Approves Rapidly<br/>(Support Acts as Enabling Partner)"]
end
```
Negotiating pricing timelines is equally critical during complex enterprise rollouts that require six to nine months of deep stack integration before any business value is realized. Thoughtful vendors work with internal legal and finance leads to structure deferred billing milestones that align vendor revenue recognition with actual enterprise value realization. Furthermore, Polito debunks the sales pitch that falling token prices and rising GPU efficiencies make AI software inherently cheaper: overall enterprise costs have increased compared to two years prior, and the automated routing logic dictating backend token volume is frequently controlled by the very vendors who profit from higher consumption.
In one recent RFP negotiation, a vendor attempted to raise an agreed quote by 35% just weeks before closing, arguing that backend token inflation on a capped-cost agreement would cause them to lose money. Because the vendor was candid about their underlying cost model rather than concealing it, Polito and the finance team engaged in good-faith negotiations to establish an adjusted, reasonable pricing structure rather than killing the deal.
[10:42](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=642s) Regulatory Governance, Privilege, and Support-Aligned Partnerships
Operating effectively in an in-house legal and compliance capacity requires moving away from an adversarial mindset and treating external vendors as repeat partners. High-performing vendors distinguish themselves by taking the time to understand the specific performance metrics, P&L goals, and annual bonus targets of their internal champions, helping them craft business cases that showcase organizational impact.
Simultaneously, support teams in FP&A, procurement, and legal bear the responsibility of managing enterprise risk as probabilistic, non-deterministic generative models replace deterministic systems:
- **Regulatory Strategy:** Vendors must move beyond generic compliance statements regarding the EU AI Act or state-level AI regulations (such as California and Colorado) and provide actionable operational insights based on how peer enterprises are navigating scrutiny from European regulatory bodies like the Irish Data Protection Commission.
- **Attorney-Client Privilege and Discovery:** Enterprise data indexing and broad AI retrieval engines frequently clash with legal confidentiality and attorney-client privilege. Common law and US legal standards require strict "need-to-know" access boundaries, which unconstrained enterprise discovery tools often fail to enforce.
[13:36](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=816s) Re-Engineering Internal Workflows: From Bureaucratic Gatekeeper to Strategic Teammate
Internal support teams strive to be viewed as strategic teammates rather than corporate roadblocks. Polito details a major workflow initiative involving an automated ticketing and review intake system integrating Slack with Jira for engineering compliance reviews:
- **Initial Implementation Breakdown:** The original deployment allowed engineers to generate review tickets via Slack. However, because the system lacked automated validation, engineers routinely submitted incomplete tickets missing critical technical details (such as model serialization specifications, testing protocols, and data provenance). Reviewers—juggling operational day jobs—would take days to review submissions before returning frustrating, outright rejections. This dynamic generated friction, positioning the support teams as slow, bureaucratic police.
- **Collaborative Redesign:** Working alongside the vendor, the team re-architected the Slack intake bot into an intelligent, interactive pre-flight gatekeeper. When an engineer initiates a request, the bot immediately validates the payload against all required technical parameters, prompting the engineer in real time for missing specifications before the ticket ever reaches a human reviewer.
By automating the enforcement of compliance requirements upfront, the tool absorbs the friction of saying "no." As a result, when tickets reach the legal and infosec teams, they arrive fully formed and ready for rapid sign-off, transforming support teams into efficient enablers who unblock technical roadmaps.
[15:30](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=930s) Compliance Reporting and the Danger of Inflexible, Non-Privileged Records
John Polito shares an example from the audit and compliance space to illustrate how vendors must understand the real-world legal and regulatory risks their software can inadvertently create. He frequently encounters tools pitching demos that scan all internal policies and issue verdicts on whether the company complies with regulations across the globe. From an in-house legal perspective, this approach is fraught with danger: having non-lawyer business clients run scans that generate unprivileged business records explicitly stating, *"You, company, are violating this law in this jurisdiction,"* creates damaging evidentiary facts that can be weaponized against the enterprise in future litigation or regulatory investigations.
When Airbnb encountered a vendor with a tool they liked, they asked for a simple reporting modification to mitigate this legal exposure: instead of generating definitive declarations like *"You are out of compliance,"* they asked if the output could use softer, investigatory framing such as, *"You might want to look at this here and give it more investigation."* The vendor flatly refused to adjust their language, leading Airbnb to walk away from the deal entirely. For leaders at large public companies who have been through the regulatory wringer, vendor inflexibility around evidentiary safety is an immediate dealbreaker.
- **Legal Risk Creation:** Automated compliance audits run by non-lawyers create discoverable, non-privileged written records admitting regulatory non-compliance.
- **Litigation and Investigation Exposure:** Definitive claims of illegality generated by software can be directly used as adverse evidence in court or regulatory inquiries.
- **Constructive Framing:** Airbnb requested softened, advisory phrasing prompting further review rather than hard declarations of lawbreaking.
- **Vendor Inflexibility:** The vendor's refusal to accommodate standard risk-management phrasing cost them the entire commercial engagement.
[16:49](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=1009s) Defining Success Criteria and True Vendor Partnerships
Returning to the broader theme of successful partnerships, Polito closes by emphasizing the value of vendors who proactively help clients define realistic success metrics and internal narratives. Highlighting a favorite legal tech vendor Airbnb adopted, Polito recalls how their team came in acutely aware of the broader macroeconomic climate and tech industry layoffs. Rather than pushing an aggressive headcount-reduction angle, the vendor proactively asked how they could help shape an implementation narrative that did not boil down to *"now you only need half as many lawyers."*
That level of strategic empathy and forward-thinking reinforces trust and validates the software investment. When software providers bring genuine domain expertise and look out for the client's organizational health rather than treating them solely as a transaction, they transition from a mere software seller into a long-term partner. Polito concludes his talk by encouraging vendors to adopt this mindset and opens the floor for post-presentation audience questions.
- **Macro Context Awareness:** Strong technology partners pay attention to industry realities (such as tech-sector layoffs) when positioning ROI.
- **Narrative Framing:** Rather than pitching labor replacement, thoughtful vendors help teams craft rollout stories centered on enablement and capability expansion.
- **Partnership vs. Transaction:** Proactively protecting the client's internal standing elevates a vendor from a standard sale into a valued long-term relationship.
## Notable quotes
> [01:53](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=113s) Why not vibe code this internally? — John Polito
> [01:53](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=113s) Demanding training rights on customer data — John Polito
> [01:53](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=113s) Introducing volatile cost structures — John Polito
> [05:12](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=312s) no-cost, one-week proof of concept — John Polito
> [07:26](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=446s) Request lacks 10 required items — John Polito
> [15:30](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=930s) You, company, are violating this law in this jurisdiction — John Polito
> [15:30](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=930s) You are out of compliance — John Polito
> [15:30](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=930s) You might want to look at this here and give it more investigation. — John Polito
> [16:49](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=1009s) now you only need half as many lawyers. — John Polito
## Mentioned
John Polito, Airbnb, Inc., Ai4, Slack, Jira, EU AI Act, Irish Data Protection Commission
## Video description
John Polito - Deputy General Counsel, Commercial & Technology, Airbnb, Inc.
In 2026, clients are comparing every vendor's offerings to the capabilities of now-commonplace frontier LLMs and general-purpose AI assistants. What are some hallmarks of software companies and technology consultants who show up as genuine, good-faith potential partners, ready to help clients grapple with that evaluation? In contrast, what behaviors may not only close the door on a deal but also lead a vendor to develop a lasting, not-so-great reputation? In this short presentation, hear one technology leader's views, distilled from some (suitably anonymized) real-world examples.
According to John Polito, Deputy General Counsel at Airbnb, successful enterprise technology partnerships require vendors to actively equip internal support teams to defend software expenditures against cost scrutiny, unproven in-house generative prototyping, and legal liabilities.
Defending External Software Against In-House Vibe Coding (1:53)
Polito observes that enterprise software renewals in 2026 face intense skepticism from business and technical units questioning whether commercial tools can be replaced by internal engineering or generative "vibe coding." While assembling primitive tools rarely yields an enterprise-grade solution, internal support teams in legal, procurement, and finance must repeatedly justify external spending. These renewal negotiations break down when vendors demand retroactive rights to ingest customer data to train proprietary foundation models, or attempt to replace predictable contract pricing with open-ended, consumption-based token structures that destabilize enterprise budgets.
Escaping the Traps of Synthetic Proofs of Concept (3:43)
Enterprise teams often favor internal generative prototypes because they launch immediately with production data and bypass infosec, privacy, and legal reviews. However, Polito notes that moving an ad-hoc prototype into production remains an arduous engineering effort that lacks enterprise security, scale, and contractual service-level agreements (SLAs).
Vendors frequently undermine formal evaluations by offering rigid, one-week proofs of concept (POCs) that rely on synthetic data and burden staff with evening testing. While viable for lightweight utilities like image generation, this structure fails for deep infrastructure like Contract Lifecycle Management. In a recent enterprise analytics RFP, Airbnb disqualified an inflexible vendor that insisted on a standard one-week synthetic POC, awarding the contract instead to a competitor that adapted by providing an eight-week evaluation window, pre-populated synthetic datasets, and dedicated engine training.
Aligning AI Token Volatility With Enterprise Budgeting (7:26)
Corporate finance teams mandate cost predictability, making raw pay-as-you-go (pay-go) token pricing unworkable for core enterprise infrastructure, even if Airbnb occasionally uses it for disposable, low-risk tools. Polito disputes vendor claims that falling token prices and GPU efficiencies make AI software inherently cheaper, reporting that overall enterprise costs have risen compared to two years prior, with vendors often controlling the backend routing logic that drives consumption.
To bridge lengthy implementation cycles, thoughtful vendors negotiate deferred billing milestones that align payment schedules with the six to nine months required for deep stack integration. When backend token inflation prompted one vendor to request a 35% quote increase shortly before closing an RFP, their upfront transparency about the underlying cost model allowed Airbnb's finance and legal teams to negotiate an adjusted rate rather than abandoning the deal.
Mitigating Evidentiary Exposure and Operational Friction (15:30)
Automated compliance and governance software can create substantial legal liability if vendors disregard enterprise evidentiary risks. Polito warns that automated policy scans run by non-lawyers often generate discoverable, unprivileged records asserting that a company is violating specific regulations, which can be weaponized in litigation. Airbnb abandoned an engagement with an audit software vendor that flatly refused to soften definitive non-compliance assertions into advisory language prompting further investigation.
Vendors provide far greater value when they help internal teams eliminate procedural bottlenecks. Airbnb collaborated with a vendor to convert an intake bot connecting Slack and Jira into an interactive pre-flight gatekeeper that validates compliance parameters in real time before human review, turning support teams from bureaucratic blockers into strategic enablers. Polito also highlights the importance of narrative alignment, praising vendors who actively avoid pitching headcount reduction and instead help internal champions frame software adoption around capability expansion.
Polito concludes that long-term vendor survival depends on strategic empathy: software providers must offer predictable cost structures, mitigate legal and evidentiary liabilities, and help internal champions demonstrate measurable organizational enablement beyond ad-hoc internal experimentation.
## Background
本内容由 Ai4 于 2026 年 8 月 19 日发布在 YouTube 平台上。视频记录了 Airbnb 商业与技术副总法律顾问 John Polito 的演讲，主要探讨在 2026 年前沿大模型与通用 AI 助手普及的背景下，软件及技术供应商应如何展现出真正可信赖的合作伙伴特质，以及哪些负面行为会导致商务谈判破裂并损害供应商声誉。
## TL;DR
在 2026 年企业普遍面临“能否用 AI 内部自研代码”的灵魂拷问下，外部技术采购与存量续约正承受前所未有的严苛审视。Airbnb 商业与技术副总法律顾问 John Polito 从法务与后台支持部门的实战视角出发，剖析了企业在采购决策中的核心痛点，包括僵化的短期 POC 与灵活测试的对比、固定费用与不可控 Pay-Go 计费模式的博弈，以及合规与法律特权风险等。他指出，优秀的供应商应当具备同理心与透明度，主动协助内部团队构建价值叙事并分担合规压力，从而超越单纯的销售买卖，转型为长期的深度战略伙伴。
## Atmosphere & style
整场演讲呈现出务实、坦诚且略带幽默的叙事基调。作为前软件工程师出身的资深法务专家，讲者抛弃了空泛的 ROI 术语与宏大口号，以“战壕一线”的内部人视角直击采购博弈中的真实摩擦与潜规则。演讲节奏明快生动，结合了大量鲜活的匿名脱敏案例，既有对僵化供应商的辛辣吐槽，也有对建立真诚共赢伙伴关系的深刻洞见，令听众在轻松共鸣中感受到极强的实操指导价值。
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=4s) 企业法务与采购处于技术落地的核心接口层，优秀的供应商会主动提供论证依据，协助法务向内部业务客户证明方案价值并推进立项审批。
- [01:53](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=113s) 大模型普及引发了企业自研风潮，采购与续约面临严苛审视；若供应商此时强推数据训练权或不可控的新计费模式，极易直接导致续约失败。
- [03:43](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=223s) 虽然内部“氛围编程”能快速构建原型，但从原型走向成熟生产环境存在巨大鸿沟，企业级供应商的核心价值在于提供安全合规与稳定交付的背书保障。
- [05:11](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=311s) 针对底层基础设施系统的评估，死板的“一周免费 POC”往往流于形式；优秀供应商应因地制宜，提供更具弹性的测试周期与深度数据支持。
- [07:26](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=446s) 企业财务与法务极度偏好有明确上限的固定费用表（Fixed Schedule），排斥将不可控成本转嫁给买方的 Pay-Go 计费机制，并倾向于与实际交付进度绑定的弹性付款排期。
- [09:18](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=558s) 在 AI 基础设施与 Token 成本波动下，商务谈判双方应摒弃零和对抗思维，以长期伙伴心态保持透明沟通，共同协商合理的折中定价方案。
- [11:09](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=669s) 具有前瞻性的供应商会主动探询客户的年终绩效考核与损益表诉求，并输出跨行业其他客户在技术转型中的组织价值评估经验。
- [12:07](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=727s) 将非确定性 AI 引入企业场景带来了新型合规与律师-委托人特权保密风险，需要供应商兼备跨司法管辖区的实际监管与实操应对能力。
- [13:36](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=816s) 通过 AI 工具在需求提交前端自动校验并指出缺失信息，让系统充当“纠错坏人”，使法务与安全团队能够快速介入充当解决问题的“破局英雄”。
- [15:34](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=934s) 审计工具若直接出具确定性违规结论，会生成不受法律特权保护的商业记录并带来诉讼风险；供应商若缺乏对法律风险的敏感度与定制灵活性将被直接淘汰。
- [16:52](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=1012s) 优秀的供应商能够主动协助内部团队构建富有建设性的组织价值叙事，避免将技术引入单纯包装为裁员替代，建立深度的长期信任。
## Detailed walkthrough
### [00:04](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=4s) 演讲开场与背景：法务视角下的技术供应商合作
Airbnb 商业与技术副总法律顾问 John Polito 开启了本次分享。在 Airbnb 的法务领导层中，Polito 带领团队统筹所有的商业合同审批、AI 治理机制，并为公司内部 3,000 多名工程师和数据科学家提供法务支撑。正因如此，他的团队成为了外部技术供应商、咨询顾问与 Airbnb 庞大技术组织之间最核心的“接口层”。在开场给出一个标准的法律人免责声明（仅代表个人观点，不代表 Airbnb 官方）后，Polito 幽默地表示要跳过那些在各个会议上被反复提及的陈词滥调——比如怎么精确衡量 ROI、设定关键绩效指标（KPI），或者如何防止 AI Agent 行为失控等。
Polito 选择从一个更为一线和务实的视角切入：在企业内部，任何一份技术许可、每一个概念验证（POC）项目都必须流经他的团队。在这个位置上，他见识了无数供应商打交道时的固定套路。他首先向所有在企业后台默默奉献的法务、采购和财务同仁致敬，认为这些支持部门平时缺乏足够的认可。紧接着，他直言供应商之间存在巨大差距：优秀的供应商深知企业内部的立项与审批阻力，会主动提供翔实的依据和策略，协助法务与采购向内部业务客户“反向推销”该方案；而差的供应商则完全不理解客户的内部流程，甚至给推进合作制造障碍。
### [01:53](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=113s) “我们自己写代码不行吗？”：自研风潮与供应商续约的严苛审视
进入 2026 年，外部采购环境发生了翻天覆地的转变。如今每当企业准备产生一笔外部软件或咨询支出时，都会面临来自其他部门极其尖锐的灵魂拷问："为什么我们不能自己写代码解决？为什么不能直接拿现成的企业工具自己在内部搭建？" 这种必须时刻为外部采购寻找合规与合理性证明的压力，已经深刻影响了每一笔供应商合作与咨询交易。Polito 打了个比方：就像古典力学里有六种简单机械一样，把它们一股脑全拼在一起并不能解决所有复杂问题；因此定制化软件、专业咨询顾问以及真正具备交付能力的行业专家依然有其不可替代的生存空间，世界并没有完全被员工的“DIY 自研”所吞没，但整个市场的竞争逻辑已被彻底重塑。
在这种背景下，存量供应商的合同续约（Renewals）正面临前所未有的严苛审视。企业管理层都在盘算：之前采购的所有工具是否真的必不可少？公司对员工进行 AI 技能升级的投入，能否通过砍掉这些外部固定成本来体现成效？而让续约雪上加霜的是某些合作多年的老牌供应商的傲慢操作——他们不仅没有提供便利，反而借机提出全新的数据使用条款，要求使用 Airbnb 的企业数据来训练自己的模型，或者单方面引入极易导致成本失控的新计费模式。Polito 直言，这类行为在 2026 年的采购决策中“反响极其糟糕”，极易直接断送续约前景。
```mermaid
flowchart TD
accTitle: 采购与续约决策面临的内部审视流程
subgraph S1["内部挑战与质疑"]
A["业务提出采购/续约需求"] --> B{"内部灵魂发问:<br/>'能否用 AI/现成工具自研？'"}
B -->|"可以替代"| C["削减外部支出<br/>转向内部自研 (DIY)"]
B -->|"无法替代"| D["进入供应商合规与商务评估"]
end
subgraph S2["供应商行为分化"]
D --> E{"供应商策略判断"}
E -->|"不合理加价 / 索要数据训练权 / 计费失控"| F["严苛审查下被否决淘汰"]
E -->|"理解客户难处 / 协助内部证明价值"| G["达成合作并成功落地"]
end
```
### [03:43](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=223s) POC 与快速原型的博弈：从概念验证到真正生产落地的鸿沟
在技术愿景与工作流规划的初期，法务与技术团队经常陷入一场路线之争：究竟是走传统的概念验证（POC），还是干脆跳过流程直接做快速原型开发（Prototyping）？在内部，尤其是喜欢利用聊天机器人进行“氛围编程”（Vibe Coding）的支持团队中，直接原型的呼声极高。业务部门的逻辑很直接：自己做原型明天就能立刻动手，完全不需要排队等待信息安全审查（Infosec review）、隐私合规审查以及法务团队繁琐的合同起草过程，而且可以直接使用真实的业务数据。
然而，从一个玩具式的原型走向成熟稳定的生产环境（Production），中间隔着巨大的鸿沟。作为前软件工程师出身、如今“留着灰白胡子”的资深法务专家，Polito 调侃道：24 岁的年轻营销顾问可能会觉得软件上线轻而易举，但任何成熟的技术组织都明白从开发推进到生产发布有多么艰难。引入成熟供应商的核心价值，恰恰在于其经过实际场景检验的稳定交付能力，以及能够在合同和商业信誉上为结果提供背书与兜底保障——这正是法务团队在说服内部业务客户时最强有力的支撑点。
- **快速原型（Prototyping）的诱惑**：无需等待漫长的安全审查、隐私合规和法务盖章，能立刻接入真实数据并快速启动测试。
- **概念验证（POC）的不可替代性**：由专业供应商提供成熟、受保障的企业级能力，规避自研原型在扩展性、安全性和生产级交付上的巨大风险。
### [05:11](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=311s) 案例剖析：告别死板的“一周免费 POC”，寻找灵活的真正伙伴
针对供应商普遍推销的“一周免费概念验证（POC）”，Polito 结合实际案例进行了利弊剖析。对于像图像生成这类无需与内部重度基础设施深度绑定的单点工具，一周免费测试完全可行；但对于底层基础设施或需要嵌入员工日常工作流的系统（例如合同生命周期管理系统 CLM），一周的免费 POC 往往流于形式。因为在一周的短时间内系统根本无法接入真实数据，这导致测试人员必须在日常本职工作之外，利用晚间和周末等私人时间，使用人工合成数据进行疲于奔命的额外测试，根本无法验证工具能否真正为其减负。
在一个涉及深度数据分析模块的 RFP 竞标中，两家入围最终轮的供应商展现出了截然相反的合作态度：
| 供应商类型 | 提出的 POC 方案 | 沟通反馈与调整 | 最终合作结果 |
|:--- |:--- |:--- |:--- |
| **死板供应商 A** | 1 周免费 POC（仅限合成数据） | 拒绝根据分析模块需要时间沉淀的实际情况做出任何让步 | **直接淘汰**。短期合成数据对深度分析报表毫无评估价值。 |
| **灵活伙伴供应商 B** | 最初同样提议 1 周 POC（合成数据） | 认真倾听客户痛点，主动将测试期延长至 **8 周**，提供预填充的合成数据并配合开展分析培训 | **成功中标并持续合作至今**。协助法务向领导层完美呈现了未来的报表分析价值。 |
Polito 强调，优秀供应商的特质在于懂得“根据客户的实际现状因地制宜”，而不是拿针对初级客户的一套死板模版强加给成熟企业。
### [07:26](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=446s) 计费模式与预算痛点：固定费用 vs. 不透明的 Pay-Go 机制
在探讨预算与商业计费结构时，Polito 深入分析了当前大模型时代下企业端与供应商之间的博弈。许多供应商底层技术依赖于后端的实时 Token 消耗，由于自身成本处于动态波动中，供应商陷入了是“将不确定成本转嫁给客户”还是“自行消化”的纠结。因此，越来越多的供应商开始向企业兜售按量付费（Pay-Go）模式，试图将黑盒系统产生的 Token 消耗成本直接推给买方。但从法务和财务做预算的角度来看，不可预测的计费几乎无法通过内部审批，企业极其偏好具有明确梯度与上限的固定费用表（Fixed Schedule）。在 Airbnb 内部，Pay-Go 模式能够获批的唯一罕见例外，是针对那些大家预期大概率会失败的小工具——因为一旦工具不好用，企业可以随时停止使用并立即中断付款。
此外，对于大型底层架构系统的采购，通常需要 6 到 9 个月甚至更长的深度集成周期，企业在实施初期根本无法获取实际业务价值。优秀的供应商能够站在长期伙伴关系的视角理解这一痛点，在面对季度末或年底销售人员迫切希望确认收入（Book revenue）的需求时，双方可以通过商务谈判达成弹性的付款排期——确保在企业系统真正上线并产生业务回报之前，不承担过重的财务负担。而这种富有弹性的付款结构，恰恰是不可控的 Pay-Go 计费机制所无法支持的。
### [09:18](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=558s) Token 成本博弈与采购谈判中的透明度
关于人工智能基础设施的成本，外界常有一种乐观论调，即“无需担心，Token 价格正在持续下降”。但现实情况是，即便单 Token 价格在走低、GPU 能效在提升，整体支出依然显著高于两年前。更值得警惕的是，决定如何消耗 Token、通过何种技术路径路由请求的模型服务方，往往正是用户消耗更多 Token 时的直接受益者。
在谈及近期的 RFP（采购提案征集）经验时，法务团队往往处于谈判的核心。近期有一家供应商在完成了三至四周的系统测试、双方已敲定合同细节且即将签约的最后关头，突然提出要将两个月前的报价上调 35%，理由是“市场环境发生了变化”。法务团队对此的第一反应是“你们疯了吗？（You are out of your mind）”，因为所有预算都需要向财务团队进行严密论证，必须明确说明到底产生了什么新增价值。供应商随后坦白解释道，他们预测未来的 Token 成本将大幅上涨，而之前承诺的封顶价格（capped cost）很可能会导致他们在这笔交易中亏本。
面对这种坦诚，团队尽管无法直接接受 35% 的涨幅，但依然愿意基于透明的事实坐下来协商一个合理的折中方案。很多初入企业法务或支持部门（如采购、财务）的人容易陷入一种对抗性思维，认为外部供应商都在试图占便宜，从而将自己仅仅定位为“防范外部风险、争取最低价格”的守门人。但现实中大家都是长期博弈的重复参与者（repeat players），保持善意、假设对方并非刻意欺诈，并要求对方坦诚相待，往往是达成双赢合作的前提。
### [11:09](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=669s) 理解客户的绩效衡量标准与年终诉求
在企业组织架构中，每个人都有自己的上级和考核机制，支持团队与业务实施部门在年终都需要面对 P&L（损益表）、绩效考核（Performance Review）以及奖金评估。优秀的供应商往往能够展现出超前的同理心与前瞻性，他们会在交流中主动询问并记住客户的具体考核方式：“请告诉我们，在年底的绩效评估中，你们希望展现怎样的成果与业绩？”
更具价值的合作伙伴还会分享行业前沿的视角，告诉客户其他企业是如何顺应技术变革、重构自身团队价值评估标准的。企业法务通常只服务于一家大客户（即自身所在的公司），无法接触到海量外部样本，因此供应商如果能将跨行业客户在技术转型与绩效考核方面的演进经验反馈回来，会成为极具建设性的助力。
### [12:07](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=727s) 后台支持职能的合规挑战与非确定性风险
法务、FP&A 以及采购等后台支持部门的核心职责之一，是评估外部环境变化所带来的不确定性与风险，而内部业务部门通常并不愿直接处理这些繁琐的合规问题。随着企业将非确定性（non-deterministic）的 AI 方案引入过去由确定性（deterministic）系统主导的业务场景，整个风险版图发生了根本性变化。
优秀的供应商会与企业法务并肩作战，不仅熟知《欧盟人工智能法案》（EU AI Act）以及美国加州、科罗拉多州等地的最新监管要求，还会带来极具实操性的监管动态与应对经验。例如，供应商会主动提示某家客户在爱尔兰监管机构面前遇到的审查痛点，并提供相应的落地建议。
在法务领域，AI 的广泛接入带来了特有的法律挑战：
- **律师-委托人特权与保密性（Attorney-Client Privilege & Confidentiality）**：在普通法系国家，敏感信息的共享受到极高限制，严格遵循“有知情必要（Need-to-know）”原则，而许多具备广泛数据发现能力的 AI 访问控制系统在设计之初并未充分考虑这一法律边界。
- **合规前瞻性要求**：企业不仅需要被动满足法规文本，更需要应对多司法管辖区下实际执法尺度的动态变化。
### [13:36](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=816s) 让工具做“坏人”：优化工单流程与内部角色重塑
在企业内部推进 AI 落地时，不仅需要将 AI 视为团队成员（Teammate）而非单纯的工具，法务等支持职能同样希望被内部业务部门视为同盟者，而非阻碍项目推进的“路障（Roadblock）”。
过去通过 Slack 集成 Jira 自动创建技术工单的方案曾遭遇过严重挫折：研发人员通过 Slack 发起请求后，生成的工单往往信息缺失严重（例如缺少模型选型、序列化方案或原型测试数据等关键要素）。由于审核人员日常工作繁重，通常会在两天后才驳回并抱怨“你的工单不合格，重填一遍”，导致业务部门极度不满，支持部门也沦为了惹人厌烦的“流程警察”。
```mermaid
flowchart TD
accTitle: 工单审查流程改进前后对比
subgraph BAD["改造前的低效阻断流程"]
A["工程师提交粗糙工单"] --> B["等待 2 天人工初审"]
B --> C["法务/安全团队被迫充当'警察'打回"]
C --> D["内部关系紧张与流程延误"]
end
subgraph GOOD["改造后的 AI 实时反馈流程"]
E["工程师通过 Slack 发起需求"] --> F["AI 工具即时校验并给出指引<br/>'需要补齐这 9 项模型与架构参数'"]
F --> G["工程师完善信息后正式提审"]
G --> H["法务/安全团队高效介入<br/>充当解决问题的'破局英雄'"]
end
```
通过与供应商重构工作流，系统在工程师于 Slack 提交需求的当下便进行智能前置交互。在工单流转至信息安全（Infosec）或法务部门之前，工具会自动给出友好且具体的提示：“我们非常乐意协助推进，但目前仍缺少 9 项必要信息（如模型的具体细节、序列化策略等），请补充完整”。这种机制让自动化工具承担了指出错误的角色，而法务与安全团队则能在信息齐备后迅速接入解决实质问题，真正从“纠错的家长”转变为“攻坚的英雄”。
### [15:34](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=934s) 合规报告的法律陷阱与供应商的灵活性
在合规与审计工具的选型中，很多供应商演示时会宣称其工具能够扫描企业的所有制度，并自动出具一份报告，明确指出企业“在全球违反了哪些法规”。这种做法在法律上极其危险：它等于让非律师的业务人员创建了一份不享有法律特权保护的商业记录（unprivileged business record），直接坐实了公司违法的证据，在未来的诉讼或监管调查中将造成灾难性后果。
针对此类工具，法务团队曾向供应商提出明确的定制要求：
- 将输出结论从确定性的“贵公司存在违规（You are out of compliance）”，修改为引导性的“建议关注此处并做进一步调查（You might want to look at this here and give it more investigation）”。
然而该供应商态度极其僵化并断然拒绝调整，导致合作直接终止。对于大型上市公司而言，供应商如果缺乏对法律风险的敏感度与配合灵活性，将无法建立长期信任。
### [16:52](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=1012s) 共同构建价值叙事：从单纯工具到真正伙伴
在科技行业面临裁员潮的大背景下，技术工具的引入往往会引发团队对人力削减的焦虑。一家优秀的法律科技供应商在推介方案时，主动向法务团队提出：“我们注意到了行业内的裁员动态。在推进这套系统上线时，我们该如何帮你们打造一个更有建设性的业务叙事，而不是简单粗暴地宣传‘现在法务团队可以裁掉一半人了’？”
这种具备前瞻思维、主动协助内部客户构建组织价值叙事的供应商，超越了单纯的销售关系，真正展现了专业伙伴的深度协同价值。
## Notable quotes
> [01:53](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=113s) 为什么我们不能自己写代码解决？为什么不能直接拿现成的企业工具自己在内部搭建？ — John Polito
> [03:43](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=223s) 24 岁的年轻营销顾问可能会觉得软件上线轻而易举，但任何成熟的技术组织都明白从开发推进到生产发布有多么艰难。 — John Polito
> [05:11](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=311s) 根据客户的实际现状因地制宜 — John Polito
> [09:18](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=558s) 无需担心，Token 价格正在持续下降 — John Polito
> [09:18](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=558s) 你们疯了吗？（You are out of your mind） — John Polito
> [11:09](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=669s) 请告诉我们，在年底的绩效评估中，你们希望展现怎样的成果与业绩？ — 优秀供应商
> [13:36](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=816s) 我们非常乐意协助推进，但目前仍缺少 9 项必要信息（如模型的具体细节、序列化策略等），请补充完整 — AI 自动化工具
> [15:34](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=934s) 建议关注此处并做进一步调查（You might want to look at this here and give it more investigation） — John Polito
> [16:52](https://www.youtube.com/watch?v=bm-DW7vo2sk&t=1012s) 我们注意到了行业内的裁员动态。在推进这套系统上线时，我们该如何帮你们打造一个更有建设性的业务叙事，而不是简单粗暴地宣传‘现在法务团队可以裁掉一半人了’？ — 法律科技供应商
## Mentioned
John Polito, Airbnb, Inc., Ai4, Slack, Jira, 《欧盟人工智能法案》（EU AI Act）
## Video description
John Polito - Deputy General Counsel, Commercial & Technology, Airbnb, Inc.
In 2026, clients are comparing every vendor's offerings to the capabilities of now-commonplace frontier LLMs and general-purpose AI assistants. What are some hallmarks of software companies and technology consultants who show up as genuine, good-faith potential partners, ready to help clients grapple with that evaluation? In contrast, what behaviors may not only close the door on a deal but also lead a vendor to develop a lasting, not-so-great reputation? In this short presentation, hear one technology leader's views, distilled from some (suitably anonymized) real-world examples.
[Source](https://www.youtube.com/watch?v=bm-DW7vo2sk)
Source quotes
[1:53] John Polito: 为什么我们不能自己写代码解决？为什么不能直接拿现成的企业工具自己在内部搭建？
[3:43] John Polito: 24 岁的年轻营销顾问可能会觉得软件上线轻而易举，但任何成熟的技术组织都明白从开发推进到生产发布有多么艰难。
[5:11] John Polito: 根据客户的实际现状因地制宜
[9:18] John Polito: 无需担心，Token 价格正在持续下降
[9:18] John Polito: 你们疯了吗？（You are out of your mind）
[11:09] 优秀供应商: 请告诉我们，在年底的绩效评估中，你们希望展现怎样的成果与业绩？
[13:36] AI 自动化工具: 我们非常乐意协助推进，但目前仍缺少 9 项必要信息（如模型的具体细节、序列化策略等），请补充完整
[15:34] John Polito: 建议关注此处并做进一步调查（You might want to look at this here and give it more investigation）
[16:52] 法律科技供应商: 我们注意到了行业内的裁员动态。在推进这套系统上线时，我们该如何帮你们打造一个更有建设性的业务叙事，而不是简单粗暴地宣传‘现在法务团队可以裁掉一半人了’？
Source quotes
[1:53] John Polito: Why not vibe code this internally?
[1:53] John Polito: Demanding training rights on customer data
[1:53] John Polito: Introducing volatile cost structures
[5:12] John Polito: no-cost, one-week proof of concept
[7:26] John Polito: Request lacks 10 required items
[15:30] John Polito: You, company, are violating this law in this jurisdiction
[15:30] John Polito: You are out of compliance
[15:30] John Polito: You might want to look at this here and give it more investigation.
[16:49] John Polito: now you only need half as many lawyers.
Original transcript
[0:04] Thank you very much. Hey everybody. Uh my name
is John Polito. I'm Deputy General Counsel for
[0:09] Commercial and Technology at Airbnb. I'd like
to spend our brief time together to talk a
[0:15] little bit about some of the great things
that happened for me with vendors in 2026.
[0:19] Maybe a few less great ones. Uh bit about me
to help give context to this presentation.
[0:25] Um so at Airbnb, I'm in legal leadership. I
have the commercial legal team that handles
[0:28] all the contracts. I have the AI governance. I
have general support for over 3,000 engineers
[0:33] and data scientists. So my team and I are
pretty much the interface layer between
[0:37] uh outside vendors and consultants and our
technology organization. Uh and it wouldn't be
[0:43] a lawyer presentation without a disclaimer.
I me I'm not talking on behalf of Airbnb.
[0:47] All right. So the good thing about Thursdays
I'm going to skip all the stuff from all the
[0:50] presentations you've already heard because
I think we've all heard variations on the
[0:54] it's really important to get your ROI, you know,
measurements correctly, get your right KPIs, make
[0:58] sure your agents aren't going crazy. All of those
things are true at everywhere and I'm sure um we
[1:03] can all resonate with that. I want to talk about
something a little bit different. So where I sit,
[1:07] every single license that comes in, every proof of
concept, um all comes through my team. So I see a
[1:13] lot of the same patterns over and over again and
I got to say, I don't you know, for people who are
[1:18] are in support functions at companies like me,
um you all rock. We don't get enough, you know,
[1:23] love so thank you to all my legal and procurement
and finance friends, you're awesome. Um for folks
[1:28] who are vendors, some of you are really great
and understand, yes, we do this all the time
[1:33] internally, but you you help people to implement
or bring in or justify cost over and over again
[1:38] and so you can really give us tips for how we're
going to sell stuff to our internal clients.
[1:43] Some of you don't. Uh and [laughter] I would love
it if everybody could move to that step where they
[1:47] can help us to look good internally as we bring
in the services that people are providing. Um
[1:53] one of the big changes at least that I've seen
and I talked to my we all feel it is the world
[1:59] is different now. Uh when you're thinking about
any kind of outside spend that you have to justify
[2:03] because someone is going to ask me, "Why can't we
just write code this? Why can't we just use some
[2:08] enterprise tool and do it ourselves internally?"
Um it's not necessarily going to be the tech
[2:12] organization asking this, it might be some other
part of the company. So, that justification, that
[2:16] drumbeat is always a thing in the back of my mind
and my cross-functional partner's mind. Um but it
[2:21] has changed every single uh really vendor deal
that I work on, every single consultant deal. Um
[2:28] for those who don't uh aren't engineers, uh
there are six traditional simple machines.
[2:32] Doesn't really help to combine them all. Um so,
this is my way of saying, I think there's still a
[2:37] role for bespoke software, for bespoke consulting,
for experts, uh for people who actually know how
[2:42] to deliver product. I don't think that the world
has been replaced yet by do-it-yourselfing.
[2:46] Um but the landscape has changed. I'm not going to
go through all of these, um but I'm going to talk
[2:50] about maybe two points that I haven't heard so
far this week. One of them is um so, for folks who
[2:56] deal with contracts, we have regular renewals with
a lot of our vendors. Ev- and it's not just my
[3:02] personal experience, again this is with my peers
as well. Renewals are getting a lot of scrutiny
[3:07] right now cuz the same question, "Do you still
need all the tools you need you had coming in
[3:11] before? Or can we replace it internally? You know,
can we show that our investment in upscaling our
[3:16] workforce um has made success by cutting down
some of these fixed costs?" Um this gets even
[3:22] harder where a vendor who you've had for years and
years comes back and says, "Now I want new terms
[3:26] for how I can use your data to train my product."
Or now I want to introduce in a totally new cost
[3:31] structure that's going to make things a little
destabilized on your end. Um it's not selling.
[3:36] I got to tell you, those things are landing
pretty poorly this year in my experience. Um
[3:43] So, one of the places that my team gets
pretty heavily involved is at the beginning
[3:46] of a technology um vision, work stream, etc.
And there's a lot of back and forth now about
[3:52] whether we're going to keep doing proof of
concepts or whether we're instead we're just
[3:54] going to prototype everything. Sort of start
from the beginning, bring it in and just run.
[3:59] And there's a lot of pressure, especially
for my my support teams who like to
[4:04] use chatbots to vibe code whatever to say, let's
just prototype. Why? We can start tomorrow,
[4:09] right? We don't need to go through the
process of waiting for infosec review,
[4:12] waiting for privacy review, looking for the
legal to actually paper the deal for a POC.
[4:17] We can start right away and we can do it with our
real data. I'll go through an example in a second,
[4:22] but this idea that when you're testing something,
can you use your real data or is it extra
[4:26] extra data that's not yours on time? Actually
makes a difference. On the flip side,
[4:31] I think most mature organizations, and again,
myself and my peers are seeing this experience,
[4:36] that move from prototype to actual
production, even for internal tools, um
[4:42] it's it's a difficult road. Folks who've been
in I'm I have the gray beard, I'm old. Uh
[4:46] I used to be a software engineer before I was
a lawyer. So, uh it's not so easy to move from
[4:50] developing something to pushing it out, but
if you are a 24-year-old marketing consultant,
[4:54] you might think it's super easy. And so having to
have that conversation over and over again that
[4:58] bringing in a vendor who's got a tool who actually
is you know, proven in the field that they can
[5:02] produce some results reliably and will guarantee
it, whether contractually or reputationally or
[5:06] otherwise, um I think still has value that carries
the day when I'm talking to my internal clients.
[5:11] Um so here is my first sort of real world
examples. I'm not going to name names.
[5:18] We get a lot of requests for vendor POCs and a lot
of them are we will come in and do a no cost one
[5:24] week proof of concept for you. Um sometimes that's
exactly what we want, right? Image generation is
[5:30] a great example. Like you don't need to I
think have generally heavy infrastructure
[5:34] investment from your internal teams when you're
bringing in something like image generation.
[5:39] Something that is really at the infrastructure
layer or something that's coming in and replacing
[5:43] someone's actual regular workflow. Um I'm a lawyer
so I think about CLM technology, for example. Uh,
[5:50] one week, you know you're not using real data
if you're coming in giving me a free one-week
[5:54] POC. And what does that one week mean then? It
means that someone is testing that technology,
[5:59] and it's not their day job, and it's not helping
them reduce the work that they're already having
[6:03] to do. They're doing it extra, evening, weekend,
you know, whenever they can squeeze in some hours
[6:07] to test this technology. So, I was doing a recent
RFP. Some of the folks I really liked came in,
[6:13] we talked about POC. I got a lot of,
you know, one-week POC, synthetic data.
[6:19] Not only does that mean, you know, hey, that means
that my team, or the team that's that's doing the
[6:24] testing, actually is doing this in their off
hours, it's not helping them with their job,
[6:27] but it was a tool that had a big analytics
component. What am I going to do with one week of
[6:32] synthetically generated data to look at analytics?
It makes no sense. Um, so, we had two finalists in
[6:37] this in this instance. One would not move off of
the one-week POC. We did not use them. The second
[6:43] also offered us a one-week POC with synthetic
data. And we talked to them, we said, "Hey,
[6:46] come on. I really want to test this, but if you
want us to have one week of usage, we need a lot
[6:51] more time." We ended up with that team with
an eight-week POC, again, synthetic data, and
[6:55] they worked with us [snorts] to do some training
on analytics, to give us pre-populated synthetic
[6:59] data, so we could actually do some analysis and
show leadership, "Hey, if we have this tool,
[7:04] here's what our reporting might actually look
like." Um, that's a vendor we're using today.
[7:08] It's a great relationship. Um, they listened, they
understood the problem, and it wasn't their first
[7:12] rodeo, right? They knew that, um, you know, a
one-week POC is something you get in the door for
[7:16] someone who maybe isn't experienced. Uh, but what
you really want is someone with the flexibility
[7:20] to come in and and uh, meet you where you are. Um,
I want to talk a bit about budgeting, too. So, so,
[7:26] I mean, you are here at 3:00 listening to a lawyer
talking about finance, and you're all awesome. So,
[7:32] uh, one of the things that I'm seeing in my
role, again, leading the commercial team, is
[7:38] vendors aren't sure what they they do when they
their own technology depends on a stream of
[7:45] not yet realized costs, right? Folks who are
turning tokens on the back end. Should they pass
[7:49] that cost on to me, the consumer, or should they
eat it, right? And no surprise, when I'm trying to
[7:55] budget for the year, when I'm trying to make sure
people have authorization, I love a fixed number.
[7:59] I like a schedule that I can live with, you know,
with cheering and so forth. Um we've had more and
[8:04] more vendors come to us and say, "What would be
really great for you is a pay-go system where
[8:08] however much our black box requires us to spend
in tokens, we'll pass that cost on to you."
[8:14] Again, I can't sell it, to be honest. It's
just not a thing that makes sense from a
[8:16] finance perspective without unless it is the best
tool in the universe and perhaps not even then.
[8:22] Um on the flip side, we have brought in some
smaller tools on a pay-go system if we thought
[8:26] they weren't going to work because [clears throat]
we stopped using them. We stopped paying for them.
[8:31] And so, um that's really the rare case where
I've been able to sell that internally.
[8:35] On cost in particular, I've had and um I've had
some examples where we've had some great partners
[8:41] who understood from a relationship perspective
that if we're building a you know, we're doing a
[8:46] giant implementation, something deep in the stack,
it's going to take a while to really bring in,
[8:50] we're all in. We're not going to get value for 6
months, 9 months. It's going to take us a while
[8:55] to really build it into our systems. It's end
of the year, you want to book that revenue.
[8:59] I understand right at the quarter for the
sales people. So, we negotiate. We say, "Hey,
[9:04] we understand this dollar makes sense. We want to
negotiate this. Can we Can we wait? Can we find
[9:08] a payment structure that makes sense so that
we're not actually all the way in financially
[9:11] until we're getting the value? Um and that is
harder again to do with a pay-go system. So, uh
[9:18] Oh, this is from the incident. So, uh people
tell me, "Don't worry, token price is going
[9:22] down." To which I think many people have said,
"It's not my stick figures, but I love them."
[9:26] Um even if price per token is going down,
even if GPU efficiency is getting better,
[9:30] things still cost more than they did 2 years ago.
Um also, I will note that the technologies that
[9:34] often decide the pathways to decide how many
tokens you're using are the folks who benefit
[9:38] when you spend more tokens. So, think about that.
Um All right. So, I had a recent RFP. Uh there's,
[9:44] you know, a lot of my talking about RFPs cuz as
a lawyer I sit in this space a lot. Where we had,
[9:49] I think it was a three or four week period where
we were testing, we got through negotiation,
[9:54] we're about to close, and they said, "You know,
market conditions have changed and therefore
[9:57] we're increasing the price quote that we gave you
two months ago by 35%." And I said, "You are out
[10:02] of your mind, right? Like we justified this,
we have a finance team, we have to go through
[10:06] and really make, you know, explain what is the
value that's changed." And they said, "Well,
[10:10] we we predict that tokens are more expensive
and because we gave you a capped cost,
[10:15] uh that essentially means that we would lose
money on this, we think." And I said, "Well,
[10:20] I appreciate the transparency. I I can't
possibly give you a 35% increase. Let's talk
[10:25] about what's reasonable." So, that's
a vendor where we actually, I think,
[10:27] got to a resolution only because they were so
transparent. I mean, I knew, right? You know
[10:31] what happens when someone comes in and says,
"We've changed your price dramatically." But,
[10:35] again, it's not just that we get things when we
have a good relationship with vendors. Sometimes
[10:38] they get things from us, right? Cuz but they
have to tell us what's actually happening.
[10:42] Um and I guess for people who are new in-house,
if if you're in a in a support organization,
[10:48] it's easy to be adversarial, right? It's
easy to think everybody on the outside
[10:51] is out to get you. You're here defending the
company, you're getting the best deal you can.
[10:55] Um that's great maybe for procurement.
But for everybody else, we kind of live
[10:58] in the real world, we're all repeat
players. Um I think it helps to have
[11:01] some good faith and assume people aren't
trying to screw you all the time. Um I
[11:09] I have a boss. My boss has a boss. Uh
my clients have bosses. We all have
[11:14] ways that we show up at the end of the year
for bonuses and so forth. Um I love it when
[11:20] a vendor comes in and says, "I understand. I
remember this conversation that we had where
[11:26] this is the way that you're going to get measured
on your success." Or not me, usually, but the
[11:30] client group that's implementing. You know, what
Tell us how you show up or want to show up at the
[11:35] end of the year in your, you know, P&L and your
performance review or what have you. Um to me,
[11:39] those are the clients who are thinking ahead
or the vendors who are thinking ahead to say,
[11:43] you know, this is this is going to make a
great relationship. And I've been fortunate
[11:47] in some cases to have folks come in and
say, we've seen people make a transition
[11:51] in how they're trying to advocate for being
measured based on changes in in technology. Um
[11:57] I love that. I'm not out there talking to multiple
clients. I'm I have one big, great client,
[12:02] uh but it's really helpful, I think, to have
that kind of interaction and get that feedback.
[12:07] Um I'm going to take a second, if I may, and
just talk a little bit about what it's like
[12:11] to be in the support function. So, as I said, I'm
a lawyer. Um one of the things that we do in the,
[12:17] you know, FP&A, procurement, legal, uh other
back-end functions is really think about
[12:24] what's changing in the world to create
uncertainty, risk, etc. Our clients mostly don't
[12:30] want to deal with this, right? It's part of
our job. As we bring in non-deterministic
[12:34] solutions to things that we had deterministic
solutions for before, our landscape has changed,
[12:39] right? And so again, one of the things that some
of our good vendor partners have done this year
[12:43] is really walk with us and say, here's where,
you know, here's what we've thought about I mean
[12:49] at this point, everyone should know their strategy
for complying with the EU AI Act, with California,
[12:53] Colorado, etc. Um you know, if you're if not, we
can have a separate talk after this. I'm happy to
[12:59] help you out with this. I'm sure there are people
who are. Um but it's more, you know, let's bring
[13:02] let's get practical. Let's actually say, we've
brought in this tool, we've seen some regulatory
[13:06] pressure, we know that this client is up in
front of the Irish regulator, maybe you are, too.
[13:10] Here are some things we've seen. Super helpful.
Um one piece that has come up for me in legal in
[13:15] particular is the question about attorney-client
privilege and confidentiality. Um a lot of access
[13:21] control systems don't really think about the fact
that under most uh US and common law nations,
[13:26] you have to be careful about sharing information
not just a certain roles but people have need to
[13:30] know. Hard to do when you're sort of putting out
these broad disco- data discovery systems. Um
[13:36] I was here at AI4 last year and one of my favorite
presentations talked about companies that do well
[13:42] in implementing AI solutions think of
AI as a teammate and not just a tool.
[13:48] And I thought [snorts] I would like my internal
clients to think of me as a teammate and not just
[13:53] a tool or perhaps better said as a roadblock. Um
I've had some clients I have some vendor rather
[13:58] who have really helped me to get here and I have
an example I'm just you know, yes, everybody loves
[14:03] a tool that you can reach out through Slack and
create some, you know, get something started.
[14:07] Outstanding. Uh we had a vendor come in and give
us a solution to do, you know, connect to the Jira
[14:12] do some ticketing based on some requirements
that um were reasonably well defined. And it
[14:17] was a disaster because what happened is a bunch
of my engineering clients were creating tickets
[14:22] and the tickets were terrible, right? You
needed 10 things to do, I don't know, like a
[14:27] test against your prototyping, make sure
you had the right series serialization etc.
[14:31] We weren't getting the data. The vendor the
engineers got no feedback about the fact that
[14:35] their ticket was insufficient. The people
who were looking at the tickets they had a
[14:38] day job and so they you would get to it two days
later and say your request sucks try it again.
[14:43] Hated it hated it. We went back to the vendor and
we said I can't do this, right? Like this is a
[14:48] great idea and a terrible implementation. Um and
we workshopped to really get to a place where same
[14:54] thing you can submit your request through Slack
you want to get some kind of review and before
[14:58] it came to I think the first stop was actually
infosec and not legal but I'll say legal. Uh
[15:05] you got feedback from the tool that said, "Hey, we
would love to help you however you need these nine
[15:10] things. Like we need to actually understand
your model. We need to understand what your
[15:14] serialization strategy is etc." It made us the
good guys not the bad guys. It made someone else
[15:18] be the you know, I don't want to be everybody's
parent. I don't want to be the the police.
[15:23] Um let the tool to position us so that we
look like the heroes cuz we come in and say,
[15:27] "We can get through this roadblock. We're
the people who can give you, you know,
[15:30] the solution instead of being the people
who say, 'You filled out the form wrong.'"
[15:34] I have one more uh example that I want to share
which has to do with compliance reporting. So,
[15:39] I have a lot of interface with and love one
of my friends who do audit compliance. Um
[15:44] I have seen a lot of tools that come in and
say, "We would like to give you a demo where
[15:49] we can run against all of your um policies
and tell you whether you're complying with a
[15:54] number of regulations around the world." And
I'm like, "Okay, let me get this straight.
[15:58] So, you're coming to my non-lawyer client and you
want them to create a business record that's not
[16:04] privileged that says, 'You, company, are violating
this law in this jurisdiction.'" That's a terrible
[16:09] idea, right? It creates a fact that's bad for
litigation or for regulatory investigation. Um so,
[16:15] we had a vendor who came in and pitched for
us. We said, "We like your tool. We hate the
[16:18] way you report. Instead of saying, 'You are out of
compliance,' can you say, 'You might want to look
[16:23] at this here and give it more investigation.'"
They said no. So, we ended up not using them. Um
[16:31] when you get that kind of feedback from somebody
in my position or in our ex-position, I think it's
[16:35] worth listening to. Uh but I appreciate that
for a lot of folks, maybe they're not thinking
[16:40] because they haven't been through the ringer.
But for those of us at large public companies,
[16:43] we've got to protect ourselves and um you can't
do it with somebody who's that inflexible.
[16:49] Um I guess I just want to go back and close a
little bit on this idea of success criteria.
[16:54] One of my uh favorite vendors, and this is
actually for a legal tool, so it's one that
[16:58] we use directly, came in and said, "You know, we
we've seen the news like everybody else. We've
[17:02] seen layoffs across tech. You know, how can we
help you to build a story when you implement
[17:07] this technology that isn't now you only need
half as many lawyers." That kind of forward
[17:11] thinking just really makes me feel good about
the investment and the relationship. Makes me
[17:15] feel that someone is bringing their expertise and
not just treating us as a sale, but as a partner.
[17:20] It It was a great experience. That's
what I got. Uh happy to talk to people
[17:26] afterwards for questions for 2 minutes. Um
and thanks. Enjoy the rest of the conference.
