LatestAll news

Home / Use cases / PCT-008

Underwrite · PCT-008 · Sourced card

EBITDA add-back autopsy

Takes a sponsor’s adjusted EBITDA apart add-back by add-back, tests each against the QoE and the agreement’s caps, and shows what is left.

Direct lendingSpecial sits / distressedPublic + sample docsVendor + in-house
INPUTEBITDA definition and QoE scheduleAI STEPSExtract every add-back · Classify andcap · Show headroom at riskOUTPUTEach add-back classified, capped andsummedHUMAN CHECKWhether the add-backs are credible forthis business

Today vs automated

TODAY
  1. Find the EBITDA definitionAssociate · manual
  2. List add-backs in ExcelAssociate · manual, re-keying
  3. Apply caps by handVP / counsel · manual
  4. Email / meeting hand-offAssociate → VP / counsel
AUTOMATED
  1. The QoE or sponsor model lands in the data roomTrigger: starts on its own
  2. Extract every add-backAutomated · Document parser
  3. Classify and capAutomated · LLM
  4. Show headroom at riskAI draft, human review · Model template
  5. Whether the add-backs are credible for this businessApprove / override · VP / counsel

AutomatedAI draft, human reviewHuman onlyTools: Data room, Document parser, LLM, Model template

Estimated valueestimate

60–820 analyst hours a year
$9k–$206k of time, vs an assumed $20k–$100k tool cost

Net: −$91k to $186k a year net of tool cost; at the low end of the assumptions it does not cover the tool cost.

Manual3–6 h per dealWith AI0.5–1.5 h per deal, review included
Assumptions
  • 3–6 analyst hours per deal today; 0.5–1.5 hours with an agent, including review
  • 40–150 deals a year for one team
  • $150–250 per hour loaded cost
  • Tool or build cost $20,000–100,000 a year (both)

Estimate, based on the assumptions shown; your numbers will differ. How we estimate

What else it can do for the team

  • Show basket capacity created by uncapped add-backs
  • Compare CIM, QoE and agreement EBITDA side by side
  • Draft term-sheet asks on caps and windows
  • Track realised vs projected savings after close
  • Flag revenue-synergy language
How you get thereFour steps from prompting to an agentic workflow; most teams start at step 1.
1 · PromptingAn analyst asks a chat tool about one deal.
Needs: an approved AI tool, a confidentiality policy.
Risk: unchecked answers, data pasted outside policy.
Typical starting point
2 · WorkflowA fixed pipeline extracts the same fields into your template every time.
Needs: field definitions, a template, document parsing.
Risk: breaks silently when formats change.
3 · AgentIt watches the inbox or data room, runs the checks and drafts the output.
Needs: inbox/doc-store access, system write-back, logging.
Risk: acting on a misread; needs review gates.
The automated lane above
4 · Agentic workflowSeveral agents work across the deal or book; people approve exceptions.
Needs: shared data model, permissions, audit trail, exception queue.
Risk: errors compound across steps; needs monitoring.
Solution optionsBuy (9 vendors listed), build or hybrid, with the trade-offs of each.
Buy
Hebbia, 9fin, Octus, Indus Valley Partners (IVP for Credit), Blueflame AI, Alphastream, CovenantIQ, CredCore, CreditBase
Fastest start; you adapt to the product's data model and pricing.
Build
An LLM API, a document parser, a vector or document store, your data warehouse, and a review queue.
Fits your definitions exactly; you own maintenance and controls.
Hybrid
Vendor extraction feeding your own checks, templates and systems.
Common middle path; two things to integrate and govern.

Listing is not endorsement.

What changes for the team3–6 h → 0.5–1.5 h per deal (estimate); coverage goes from the sections someone had time to read to every clause and schedule checked against the term sheet.
ROLESAssociates move from reading and retyping to checking cited exceptions.
CYCLE TIME3–6 h → 0.5–1.5 h per deal (estimate)
COVERAGEFrom the sections someone had time to read to every clause and schedule checked against the term sheet.
CONTROLSEvery output cites its source; approvals and overrides are logged.
How it works

Full title: EBITDA add-back autopsy: how much of this borrower's EBITDA exists only in the credit agreement?

“30% of this EBITDA is allowed by the contract, not earned by the business.”

Scenario

SAMPLE: A sponsor's model shows $40m Adjusted EBITDA. Reported EBITDA is $28m. The credit agreement allows run-rate synergies with a 25% cap.

What the human still decides

Whether the add-backs are credible for this business.

Why it matters

Add-back caps (25–35% on larger deals) are a 2026 documentation battleground.

What the record shows

Checked 2026-10-15.

In Viavi's October 2025 term loan, run-rate savings tied to the Transactions have no cap (24-month window, 'reasonably quantifiable and factually supportable'); savings from later deals are capped at 30% of Consolidated EBITDA; and add-backs in the lenders' February 2025 model and QoE are permitted with no stated cap. Grower baskets key off that EBITDA: general debt is the greater of $122.5m and 50% of LTM EBITDA. Every dollar of uncapped add-back is also basket capacity.

Try it: prompt and data
Quote the Consolidated EBITDA definition verbatim. List every add-back with its cap, lookback, and section. Then build a bridge from reported EBITDA {r} to maximum covenant EBITDA and state the add-back share.

Data

Public filings plus clearly labeled SAMPLE documents. Never paste confidential deal data into a consumer AI tool.

Who uses it, vendors and build
Segments
Direct lending, Special sits / distressed
Who uses it
Associate, VP, Counsel
Build
Vendor + in-house
In-house
LLM + clause extraction prompts + a 20-agreement test set from EDGAR. Estimate: 2–3 weeks.
Last verified
2026-10-15

Vendors that claim this capability

Tags reflect each vendor's own public claims at the linked page (checked 2026-10-15). Not tested, ranked or endorsed.

Sources and related

Part of the deal lifecycle: stage 4 of 7, Documents.

Next step

Get new walk-throughs

Subscribe to the newsletter

Not investment advice. Names marked SAMPLE are fictional. Found an error? See the corrections log.