Buy, Build, Deploy: Using M&A to Drive AI Transformation - Prathik Naidu, Long Lake

Acquiring and owning established services businesses, then transforming them with AI from the inside, is a more effective strategy for deploying AI into the ...

Buy, Build, Deploy: Using M&A to Drive AI Transformation

Abstract

This synthesis examines a structural argument about AI deployment: that the dominant barriers to realizing value from large language models and agentic systems are organizational rather than technical, and that ownership of the operating business - rather than vendorship to it - resolves those barriers most directly. Three deployment strategies are compared: selling AI tools into enterprises from the outside, building AI-native firms from scratch, and acquiring established services businesses for internal transformation. The analysis identifies five mechanisms through which ownership confers advantage - access, data, incentive alignment, trust, and scale - quantifies the value-capture asymmetry facing external vendors, and characterizes the economics of the approximately $70 trillion global services economy, where high marginal delivery costs suppress growth. A J-curve model of P&L transformation is presented alongside acquisition evaluation criteria and the empirical case of an applied AI holding company, Long Lake.

1. Introduction

The persistent gap between frontier model capability and realized economic impact is among the most consequential open problems in applied artificial intelligence. Benchmark performance has advanced rapidly, yet enterprise deployment of that capability remains slow, uneven, and often shallow. This pattern suggests that the binding constraint on AI value creation lies downstream of the model layer - in organizational access, incentive structures, and the mechanics of how value is captured once created.

This paper synthesizes an argument that the constraint is structural, arising from organizational complexity and value-capture asymmetry, and that ownership-based deployment - in which an operator acquires an established services business and transforms it internally - addresses these constraints more completely than alternative strategies. As one formulation states, "Building on the outside... makes deploying AI really hard. And importantly, it's not the model."

Three terms anchor the analysis. Outside-in deployment denotes the conventional vendor model of selling AI software to third-party enterprises. AI-native construction denotes founding a new firm designed around AI from inception, without legacy infrastructure or customer base. Ownership through acquisition denotes purchasing an incumbent operating business and redeploying AI capability within it, with full control of the resulting profit and loss (P&L) statement.

The central research question is: which deployment architecture maximizes both the probability of successful AI adoption and the share of value captured by the deploying party? Section 2 establishes the economic background. Section 3 compares the three strategies and details the mechanisms of ownership advantage. Section 4 extracts technical and operational insights, including the J-curve model. Sections 5 and 6 discuss implications and conclude.

2. Background and Related Work

Two economic frameworks underpin this analysis. The first is the marginal cost framework distinguishing software from services businesses. A typical software business retains approximately $80 of every $100 in incremental revenue, owing to near-zero marginal cost of distribution. A typical services business retains approximately $20, because labor costs scale proportionally with revenue. This differential constitutes "a pretty huge 80% marginal tax on growth," and helps explain a documented behavioral pattern: many services founders have deliberately declined to scale their operations, recognizing that growth under this cost structure consumes rather than generates operating leverage.

The second framework concerns value accrual under vendor relationships. Consider an AI vendor holding a $1 million annual contract that generates $5 million in annual EBITDA improvement for its client. At a 10x valuation multiple, this produces approximately $50 million in enterprise value - accruing to the client, not the vendor. The asymmetry is roughly 50:1 in favor of the entity that owns the P&L: "The largest impact actually shows up in the P&L of the business that you actually worked with." Together these frameworks establish why both services economics and vendor economics create structural headwinds for outside-in AI deployment, motivating consideration of ownership as an alternative architecture.

3. Core Analysis

3.1 The Limits of Outside-In and Build-from-Scratch Strategies

Outside-in deployment faces compounding friction at every stage of the enterprise sales and adoption cycle: security review, internal coordination, legacy system integration, unclear ownership of adoption between vendor and client, and ambiguous monetization models (tokens versus seats, price sensitivity, ROI measurement). These frictions are amplified by a deeper structural issue - organizations are more complex than their org charts suggest, with decisions entrenched in webs of experienced personnel whose tacit knowledge, exceptions, and edge cases are not captured in CRMs or standard operating procedures (SOPs). Without access to this tacit layer, vendors cannot deliver full value, regardless of model quality.

Building an AI-native firm from scratch addresses some of these issues - it allows the builder to own the entire workflow, sell the work itself, and capture resulting outcomes and P&L. However, this path introduces its own constraints: the absence of existing customers, brand, or distribution channels; difficulty hiring niche expertise in specialized industries; and client preference for firms with decades of trust and track record over unproven AI-native entrants, particularly for critical or regulated work.

3.2 The Ownership Advantage: Five Mechanisms

Acquisition-based ownership is argued to resolve the limitations of both prior strategies through five interlocking mechanisms. Access allows technology teams to embed directly with experienced staff, observing full workflows rather than reconstructed approximations. Data derives from live business operations, supplying real tasks, edge cases, and ground-truth outcomes against which agents can be improved continuously. Incentive alignment follows from ownership itself: removing competing revenue goals between vendor and client collapses the principal-agent problem, such that "you only win when your team and your business wins." Trust is inherited rather than built, as acquired businesses bring decades of brand equity and customer relationships. Scale follows from the existing customer base, expert workforce, and cash flow already present at acquisition, enabling faster compounding than a cold start would allow. These five factors are described as forming a compounding flywheel linking agents, people, and live operations.

3.3 Market Opportunity and Acquisition Criteria

The addressable context for this strategy is the global services economy, estimated at approximately $70 trillion, substantially larger than the global software market. Because AI can reduce the cost of service delivery while improving quality, it shifts service-business economics closer to software economics - partially closing the 80% marginal cost tax described in Section 2.

Not all services businesses are equally attractive acquisition targets. As AI makes intelligence cheap, the premise is that what remains valuable is trust, judgment, and accountability for outcomes. Evaluation criteria therefore center on established customer trust, the presence of human judgment in service delivery, and regulatory or accountability requirements that favor incumbents. The stated preference is for owning end-to-end outcomes - for example, handling an urgent leak - rather than point tools such as sending a newsletter. Long Lake, the applied AI holding company examined here, reviews approximately 5,000 deals annually but applies a tight filter for strategic and partnership fit.

4. Technical Insights

Several implementation-relevant findings emerge from this analysis. First, the J-curve model characterizes P&L transformation following acquisition: an initial investment phase involves building an AI platform, capturing workflows, and partnering closely with existing teams; a mid-phase sees increased usage, unlocked capacity, and realized time savings; a final phase requires redeploying that capacity toward measurable business outcomes such as growth, Net Promoter Score (NPS), and revenue per employee. Capacity gains that are not redeployed toward outcome metrics do not, by this framework, constitute completed transformation.

Second, platform and playbook reuse across acquisitions is presented as a compounding mechanism: each subsequent acquisition benefits from faster time-to-deploy and faster time-to-P&L-impact, as technology infrastructure and operational playbooks developed in prior integrations are reapplied. This implies that the marginal cost of integrating the Nth business is lower than the first, a dynamic analogous to platform economies of scale in software.

Third, the economics quantified in Section 2 - an 80% marginal cost tax on services growth versus 80% retention in software - functions as a diagnostic screen: acquisition targets are evaluated partly on the degree to which AI-driven labor decoupling can shift their cost structure toward the software profile. Limitations include the dependence of this model on access to capital for acquisitions (Long Lake has raised $7.5 billion) and on the ability to replicate integration playbooks across heterogeneous industries, from residential services to architecture, HR, and corporate travel (illustrated by the $6.3 billion take-private acquisition of Amex GBT).

5. Discussion

The analysis suggests that the primary obstacle to AI value realization in services industries is not model capability but organizational architecture - specifically, who holds the P&L and who has access to tacit operational knowledge. This reframes the AI deployment question from "which tool performs best" to "which ownership structure captures the most value from a given level of model capability." This has implications for how the AI industry allocates capital: rather than funding additional vendor competition, capital may be more productively directed toward acquiring and transforming incumbent operating businesses.

Open questions remain regarding the generalizability of the ownership thesis. The model depends on substantial upfront capital for acquisition, which constrains the number of entities that can pursue this strategy at scale. It also assumes that integration playbooks transfer across industries with meaningfully different regulatory, labor, and customer-trust profiles - an assumption only partially tested across the 40+ businesses and industries cited. Future inquiry might examine whether the five-mechanism advantage (access, data, incentive alignment, trust, scale) holds uniformly across sectors, or whether certain mechanisms dominate in specific verticals.

6. Conclusion

This synthesis has argued that ownership through acquisition offers a structurally superior path to AI deployment relative to outside-in vendorship or AI-native construction from scratch, by resolving access, data, incentive, trust, and scale constraints simultaneously. The practical takeaway for technical and investment professionals is that AI value creation in services industries may depend less on model or product superiority and more on organizational control of the operating business. For practitioners, this suggests evaluating AI go-to-market strategy not solely on technical differentiation but on capital structure and ownership position relative to the P&L being transformed.

Sources

• Buy, Build, Deploy: Using M&A to Drive AI Transformation - Prathik Naidu, Long Lake - Original Creator (YouTube)
• Analysis and summary by Sean Weldon using AI-assisted research tools

About the Author

Sean Weldon is an AI engineer and systems architect specializing in autonomous systems, agentic workflows, and applied machine learning. He builds production AI systems that automate complex business operations.

LinkedIn | Website | GitHub