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

Prathik Naidu, Long Lake | AI Engineer NYC 2026 | 18:51

Source: https://www.youtube.com/watch?v=J7n3WJnFTto
Channel: AI Engineer (https://www.youtube.com/@aiDotEngineer). Summarised by AIE Talks.
Page: https://aietalks.com/talks/buy-build-deploy-using-m-a-to-drive-ai-transformation
Published: 2026-10-19
Tags: deployment, enterprise, product-strategy, team-adoption

## TL;DR
- Deploying AI from outside a business is difficult because vendors lack access to tacit knowledge, struggle to drive adoption, and often capture less value than they create.
- Acquiring established service businesses gives an AI company access to trusted brands, experienced teams, live operational data, and direct ownership of outcomes.
- Long Lake looks for mission-critical service companies where human judgment, accountability, and regulatory constraints remain important, then uses AI to improve their operating leverage.

## Summary
Prathik Naidu argues that enterprise AI deployment is hard because the important parts of a business are often hidden in human judgment, informal relationships, and decades of operating experience. An outside vendor may integrate with a CRM or document repository, but still struggle to understand exceptions, gain adoption, and capture the value created. Building a new AI-native service company solves some of these problems, yet it starts without customers, brand trust, or specialist expertise. Long Lake takes a different route: it acquires established service businesses and transforms them from inside. Ownership gives its teams access to employees and live workflows, aligns incentives around business results, and provides an existing customer base and cash flow. Naidu describes a focus on mission-critical services where customers trust the company with end-to-end outcomes. The goal is to use AI to reduce the labor cost of services, improve quality, and turn capacity savings into revenue and profit.

## Key ideas
### Outside-in AI deployment gets blocked by the business around the model
[01:31](https://www.youtube.com/watch?v=J7n3WJnFTto&t=91s)
Naidu describes several walls between an AI product and production use. Vendors must pass security reviews, coordinate internally, and integrate with systems that may be 20 years old. After deployment, someone still has to train users, collect feedback, and drive adoption. The vendor and customer may disagree about who owns that work. Pricing and value capture create another problem because the vendor has to decide between tokens, seats, or a mix, then prove return on investment. The model may work well for a narrow use case, but getting it embedded in the full operation takes much more than model quality.

### The important business knowledge lives in people and exceptions
[02:10](https://www.youtube.com/watch?v=J7n3WJnFTto&t=130s)
A CRM, ticketing system, document repository, or collection of SOPs gives an AI vendor useful information, but it does not contain the full operating reality. Naidu says complex businesses run on human judgment, including edge cases, exceptions, tacit knowledge, and quick questions answered by experienced staff. A useful agent therefore requires close work with the people doing the job. The team needs to learn what good work looks like and understand the veterans who influence decisions. This is difficult for a third-party vendor that sits outside the organization.

### Ownership connects AI work to the value it creates
[04:28](https://www.youtube.com/watch?v=J7n3WJnFTto&t=268s)
Naidu compares an AI engagement with an established business to the value created by improving an existing P&L. In his example, a client pays $1 million a year and receives a $5 million annual EBITDA impact. Using a roughly conservative 10x multiple, that impact could create $50 million in enterprise value for the client. The vendor may have done the deployment work, but the largest effect appears in the customer's P&L. Owning the business changes that relationship because the AI operator owns the outcomes and the resulting business value.

### Acquiring an established service company supplies trust, access, and operating data
[06:46](https://www.youtube.com/watch?v=J7n3WJnFTto&t=406s)
Long Lake's ownership model gives its technology team direct access to the employees who perform the work, including people with decades of experience. The business also provides a live environment with real tasks, edge cases, and answers about whether an agent got something right. Ownership aligns incentives because the team wins when the business wins, without separate revenue goals for unrelated clients. The acquired company already has a brand, customer relationships, expert staff, and cash flow. Naidu says these advantages combine into a compounding flywheel: agents and people operate together, produce real-world data, and improve future deployments.

### Service businesses have a scaling problem that AI can change
[09:44](https://www.youtube.com/watch?v=J7n3WJnFTto&t=584s)
Software companies can add revenue with relatively little additional delivery cost, while service businesses usually need more people as they win more work. Naidu gives examples such as an HOA community manager handling only so many associations and an architect designing only so many hospitals or schools. He says a software company may keep about $80 from each additional $100 of revenue, while a service business may spend about $80 on delivery and overhead. That economics creates an 80% marginal tax on growth. AI can reduce service costs while improving quality, making service economics look more like software economics.

### Long Lake buys mission-critical services where accountability remains expensive
[13:38](https://www.youtube.com/watch?v=J7n3WJnFTto&t=818s)
Naidu says Long Lake evaluates businesses by asking whether customers trust them with complete outcomes, whether the service requires human judgment, and whether regulation creates accountability. The target is not a narrow task such as drafting a homeowner newsletter or extracting data from documents. It is the full outcome, such as handling an urgent leak or making a complicated tax decision. Long Lake wants companies with strong business fundamentals and a clear opportunity for close AI partnership. Naidu says the firm sees about 5,000 deals a year and applies a tight filter.

### The transformation starts with a J-curve and only pays off when capacity becomes business impact
[15:33](https://www.youtube.com/watch?v=J7n3WJnFTto&t=933s)
After an acquisition, Long Lake first invests in the platform, technology, workflow capture, and time with the operating team. The early phase requires iteration, validation, model training, and detailed study of the company's workflows and data. Once employees use the product, the business can produce the same output in fewer hours and unlock capacity. Naidu says that is only an intermediate result. The team then has to put the capacity to work through growth, better customer and employee NPS, and close coordination with management. The desired result is higher revenue per employee and better service quality.

## Notable quotes
- "The reality is businesses, especially in complex domains, run on human judgment." (02:51)
- "And so, as a vendor, you did all the work. You've deployed your agents, you went through all the walls of integrating with the systems, working with the team. But, the largest impact actually shows up in the P&L of the business that you actually worked with." (04:47)
- "Ownership solves for this. You only win when your team and your business wins." (07:47)
- "The important piece here is you have to own the outcome." (14:18)
- "But this isn't the end of the journey. Importantly, we want to actually capture value and we want to generate real business impact with our partner companies." (16:27)

## Tools & references mentioned
- Long Lake
- Elad Gil
- General Catalyst
- Alpha Wave
- D1
- Palantir
- American Express Global Business Travel
- Amex GBT

## Who should watch
- You are building an AI product for enterprises and keep running into security reviews, integrations, adoption problems, or unclear ownership of results.
- You are evaluating whether to build an AI-native service company or buy an established operator with customers, specialist staff, and a trusted brand.
- You work on AI deployment inside a services business and need a framework for choosing workflows where human judgment and accountability still matter.

## Related talks

- [Leadership in AI Assisted Engineering](https://aietalks.com/talks/leadership-in-ai-assisted-engineering) (Justin Reock, DX, 18:11)
- [How do you diffuse AI into the real world?](https://aietalks.com/talks/how-do-you-diffuse-ai-into-the-real-world) (Varun Shenoy, Long Lake, 17:46)
- [Reverse-Engineering the AI Buyer](https://aietalks.com/talks/reverse-engineering-the-ai-buyer) (Aliisa Rosenthal, Acrew Capital, 19:10)
- [What We Learned Deploying AI within Bloomberg's Engineering Organization](https://aietalks.com/talks/what-we-learned-deploying-ai-within-bloombergs-engineering-organization) (Lei Zhang, Bloomberg, 18:21)
- [Build Dynamic Products, and Stop the AI Sideshow](https://aietalks.com/talks/build-dynamic-products-and-stop-the-ai-sideshow) (Eliza Cabrera, Workday & Jeremy Silva, Freeplay, 18:10)
