WALL 01PRODUCTION
Get inside the business.
A successful build still has to survive the organization.
- Security review
- Internal coordination
- Integrations with 20-year-old systems
The hard part of AI
is the business.
Getting an agent to work is only the beginning. Getting it into production, into daily use, and into the P&L is the real challenge. Ownership changes all three.
Explore the thesis Long Lake case study ↗The ownership equationFIG. 01
Deploying from the outside
AI vendor
→Client business
Contract revenueP&L upside → client
One owner · one outcome
AI + people
⇄Operations
↳ Real work → data → learning → better work ↲
3 wallsBETWEEN A BUILD
AND P&L IMPACT
AND P&L IMPACT
5 advantagesCOMPOUND THROUGH
OWNERSHIP
OWNERSHIP
$70T+GLOBAL SERVICES
MARKET IN THE NOTES
MARKET IN THE NOTES
01 / The deployment problem
Three walls between
a demo and a business.
Every external AI vendor has to cross the same organizational boundaries. Building the technology is one step in a much longer chain.
Build →
Pilot →
Adoption →
P&L impact ↗
WALL 02ADOPTION
Make the work change.
Ownership is unclear between vendor and client. Who is accountable for getting people to use the system and changing the workflow?
WALL 03VALUE CAPTURE
Turn impact into revenue.
The vendor sells tokens or seats while the client owns the economic outcome.
- Token / seat pricing
- Price sensitivity
- ROI measurement
The org chart is not
the operating system.
Organizations are more complex than their formal structures. Decisions are entangled in decades of tacit knowledge and human judgment.
Edge cases, exceptions, and “quick questions to the expert” govern how work actually gets done.
Systems
of record
of record
CRM and ticketing systems: the structured layer vendors can integrate with.
Documents
SOPs and contracts: the written rules vendors can ingest.
Operating
reality
reality
Conversations, hallway chats, and exceptions: the context that lives outside software.
02 / The strategic fork
Build the business.
Or buy the starting point.
Building an AI-native firm from scratch offers control over the whole workflow. It also means creating the customer base, reputation, and expertise from zero.
The appeal of greenfield
Own the full workflow.
- 01Improve the whole workflow, not just a narrow slice.
- 02Sell the work itself, rather than a tool the customer has to operate.
- 03Own the outcomes, so the P&L is yours.
The cost of starting from zero
Trust takes time.
- 01Zero customers at the starting line.
- 02Minimal brand reputation in a market that depends on trust.
- 03Hard to hire domain experts into an unproven organization.
03 / Why acquisition changes the equation
Five advantages.
One compounding system.
Acquisition brings AI into an operating business with customers, experts, and feedback already in place. Access, data, incentives, trust, and scale reinforce one another.
01 / ACCESS
Sit with the experts.
Work alongside the people doing the job. Understand veterans with 20 years of experience and the judgment they use every day.
02 / DATA
Learn in production.
Every acquired business is a live environment with real tasks, edge cases, and ground-truth feedback.
03 / INCENTIVES
Win together.
No separate revenue goals or competing clients. You only win when the business wins.
04 / TRUST
Inherit credibility.
A brand built over decades and existing customer relationships come baked in.
05 / SCALE
Start with a business.
Paying customers, expert teams, and cash flow from day one.
The learning flywheel
Each turn improves
the next one.
Agents + people act in live operations → work gets done and creates real-world data → outcomes are measured across the platform → learning compounds and improves the system → repeat.
01 — Act in live operations
Agents and people do real work inside the business, with access to the context and expertise that sit outside formal software systems.
04 / The services opportunity
A software-sized margin.
A services-sized market.
The thesis: AI reduces cost to serve while improving quality, allowing labor costs to decouple from revenue. Service economics begin to look more like software.
$70T+
GLOBAL SERVICES
~$1T
SOFTWARE
Each teal tile ≈ $1T of services. One white tile ≈ $1T of software.
Scale illustration using figures from the supplied notes.
The “80% marginal tax” on growth.
In a traditional services business, more clients require more hires. Labor scales with revenue, so much of the incremental revenue is consumed by the cost of delivering the work.
The notes frame this as an 80% marginal tax on growth: per $100 of additional revenue, a service business retains $20 while software retains $80.
Where the next $100 goes.
Incremental economics illustrated in the notes.
What AI changesBreak the link between
Break the link between
revenue and headcount.
Lower cost to serve and better quality can unlock capacity without proportionate hiring. The opportunity is to grow revenue without recreating the same labor bill.
More output per employee
→ capacity to serve more clients
→ service economics move toward software
→ capacity to serve more clients
→ service economics move toward software
The margin comparison is a framing from the notes, not a measured margin for Long Lake or a forecast of its results.
Interactive / Value capture
Who keeps the upside?
A vendor deploys the agents; the client captures the P&L benefit and the associated enterprise value.
Vendor captures$1MAnnual contract revenue
Not enterprise value
Not enterprise value
Client captures$50MIllustrative enterprise value uplift
From the EBITDA impact$5M EBITDA × 10 = $50M EV
From the EBITDA impact$5M EBITDA × 10 = $50M EV
Original example: $1M/year contract → $5M annual client EBITDA impact → ~$50M in client enterprise value at ~10× EBITDA. The controls illustrate that example; they are not a valuation of either firm. Annual vendor revenue and client enterprise value are different measures. EBITDA is used as entered, with no separate fee deduction; the notes do not specify whether the impact is before or after the vendor fee.
05 / Thesis in practice
Long Lake.
Acquire. Deploy. Compound.
An applied AI holding company targeting mission-critical services businesses where customers trust the firm with end-to-end outcomes.
Applied AI holding company
Long Lake
Founded ~2023 · Approximately three years old at the date of the notes
$7.5B+CAPITAL RAISED
40+PARTNER BUSINESSES
55PEOPLE ON THE TEAM
~5,000DEALS SEEN PER YEAR
CAPITAL PARTNERS
Elad Gil, General Catalyst, AlphaWave, D1, and others.
TEAM BACKGROUNDS
Former founders, Palantir, and military backgrounds.
PARTNER BUSINESSES
Residential services, architecture, HR, and more.
INVESTMENT FILTER
A very tight filter on business fundamentals and the AI opportunity, despite seeing ~5,000 deals per year.
Recent deal cited in the notes
Amex GBT take-private
Entering corporate travel, where accountability extends through travel disruptions.
$6.3B
Buy responsibility for the outcome.
The target is a mission-critical business customers already trust to solve the whole problem. The notes illustrate the distinction with three examples.
Resolve the urgent leak.
End-to-end responsibility
Beyond sending the HOA newsletter.
Stand behind the tax decision.
Accountability for the decision
Beyond extracting the data.
Handle the last-minute disruption.
Own the travel outcome
Beyond making the booking.
The path through the P&L
Transformation follows a J-curve.
Invest in AI deployment → unlock capacity → put that capacity to work → drive revenue per employee and better client NPS.
Schematic only. No numeric P&L scale, deployment duration, or realized Long Lake performance is implied.
Invest in deployment
AI deployment requires investment before the operating benefits appear. That initial investment creates the dip in the transformation curve.
The central thesis
Own the context.
Improve the work. Capture the value.
Source & scope. This page presents the supplied “AI deployment challenges and strategies” notes dated Tuesday, October 13, 2026, including their slide observations. All company, deal, market-size, and operating figures are attributed to those notes; they have not been independently verified here. The calculator, market tiles, and J-curve are explanatory illustrations. This is a research note, not an official Long Lake publication.