AI product studio · private equity

The spend is up.
The EBITDA isn’t.

LightCI is an AI product studio for private equity. Inside portfolio companies, we enable the people and engineer the AI systems that show up in EBITDA.

100+
Portfolio companies
served
Claude Partner Network · Select Services PartnerPartnerAgent infrastructureLangChain
Private equity · 2023 → 2026From the 2026 report →
2023202420252026AI spendEBITDA impactTHE GAP

We close the gap inside each portfolio company, two ways.

EnablementEngineering

Some of the sponsors we work with

Vista Equity Partners
Thoma Bravo
Hg
Battery Ventures
EQT
General Atlantic
Warburg Pincus
Insight Partners
Berkshire Partners
Great Hill Partners

And dozens more across their portfolios.

Why it fails

The same four mistakes, in almost every portfolio.

All four come from one root cause: treating AI as a feature layer instead of a reason to rebuild.

SAME PRODUCT UNDERNEATH
01

Bolting AI onto legacy products

A chatbot on a ten-year-old codebase is still a ten-year-old product.

MARGINAI COST, UNTRACKED
02

No cost attribution

Token spend lands in COGS with nobody watching. Margin erodes silently.

✓ Data rights✓ Model logs✕ Evals
03

No governance

No data rights, no logs, no evals. Buyers find it in diligence.

ASSUMED3 YEARS · $MMACTUALMONTHS
04

Assuming rebuilds take years

They take months. The assumption is what keeps portfolios stuck.

And it shows up at exit

0%

of buyers walked from a deal because of AI risk.

Diligence now covers AI maturity, data provenance, and governance.

−0%

EV haircut for weak AI maturity. Already applied by 40% of investors.

A feature veneer does not survive it. Evidence does.

So we do it differently

Installed, not advised.

Two motions · one operating model

Enablement activates the people.
Engineering builds the systems.

Most portfolio companies need both, in that order.

01 · Enablement4 weeks per function

The workforce, working with AI by week four.

Governed AI in week one. Function by function, real processes become named skills.

  • Claude rollout with governance and audit
  • 5–8 shipped skills per function
  • Adoption and ROI the CFO can defend
Weekly active usageWeek 1 → 12
— — typical license rollout84% sustained
Skills shipped8 / 8
Variance narrativeFinanceNDA triageLegalTicket routingITVendor onboardingOpsBoard-pack roll-forwardFinanceRenewal briefSalesPolicy Q&AHRIncident summaryIT
02 · Engineering~3 months to production

Agents in production. Systems the company owns.

Senior engineers inside the company. Agents wired to the systems where the work lives.

  • Finance, customer, ops, board-pack agents
  • A semantic layer the company owns
  • Evals, telemetry, audit from day one
Collections agent · run #4,812running
Pull open invoices
4 ERPs · 14 business units
Score payment risk
Per invoice · history + terms
Classify disputes
Pricing · PO mismatch · delivery
Rank collector worklists
9 collectors · 1,240 accounts
Post to queues
Human approves top 5% by value
62→49
DSO, days
~$26M
working capital
3×
accounts / collector

Enablement finds the fifty processes. Engineering automates the five that move EBITDA.

How it fits together

Systems of record. Then a layer your experts built. Then agents.

Not a chatbot on the ERP. A layer built with the people who know the business, then agents on top.

Systems of record

What the company already runs

ERP · per portco
CRM & support
Billing & payroll
Data warehouse
Contracts & PDFs
Spreadsheets

Semantic layer

Built with your subject-matter experts

Definitions your SMEs signed off
Permissions per role and entity
Evals: 10–30 golden cases per skill
Every action logged and attributable

One layer. Reused by every agent and every skill that follows.

Agents

They act, they do not suggest

Finance agentslive
Customer agentslive
Ops agentslive
Board-pack agentslive

Confidence-gated: auto-route, ask, or escalate.

Outcomes

What the sponsor sees

EBITDA
cost out, revenue in
Working capital
DSO, DPO, close
Exit narrative
AI-native, evidenced

Across the hold period

From the first hundred days to the exit narrative.

One operating model. The same team from diligence to the data room.

DiligencePre-close

Know what AI is worth before you price it.

What AI is worth in this target, as a number the IC can price.

Both
First 100 daysDay 0 → 100

Governed AI in every hand. Shadow AI closed.

Governed AI in week one. The first function ships skills before the first board meeting.

Enablement
Value creationYear 1 → 3

Agents on the processes that move EBITDA.

Agents in production on the processes that move the P&L.

Engineering
Exit readinessYear 3 → 5

An AI-native story a buyer can verify.

Telemetry and audit trails that survive diligence. Evidence, not a slide.

Telemetry

PRISM · our enablement methodology

One program in.
A spectrum out.

A license is not a program. PRISM is how enablement becomes measurable.

01

Foundation

Governed AI behind SSO. Shadow AI closes in week one.

02

Function workshops

One function, four weeks, five to eight shipped skills.

03

Telemetry

Adoption, impact, cost, safety. One dashboard.

04

Company OS

Workforce, process inventory, and skills become one layer over every system.

Explore PRISM90-day program · scoped in writing before work starts
ONE PROGRAMPRISMFinanceIT & SecurityLegalOperationsEnablementGovernanceTelemetryA SPECTRUM OUT
84%
weekly active use, sustained
22
skills · 5 functions · 90 days
$4.2M
annualized at one retailer
100%
of actions logged, attributable

The gap is closing without you

Your portfolio will not wait.

Thirty minutes. Bring one portfolio company and one number you want to move.