AI will eliminate most reading in due diligence. It will not eliminate judgment.
That’s the asymmetry every deal team needs to think about.
I’ve spent over twenty years on the buy side â residential, commercial, affordable housing, stressed assets, slum rehabilitation, across Indian cities. The process hasn’t fundamentally changed: read the documents, verify the claims, stress-test the numbers, visit the site, form a judgment.
AI is about to change some of it. But not the parts most people think.
What Due Diligence Actually Involves
For those outside the PE and real estate world, due diligence is the process of verifying everything about a potential investment before committing capital. It typically covers:
Document review: Land title chains, mutation records, encumbrance certificates, development agreements, regulatory approvals (RERA, environmental clearances, municipal sanctions), existing loan documents.
Financial analysis: Historical financials, cash flow projections, sensitivity analysis, valuation under multiple scenarios, tax structuring.
Legal review: Litigation search, regulatory compliance, agreement terms, representations and warranties.
Site diligence: Physical verification, construction progress, infrastructure, market survey.
Management assessment: Track record, capability, integrity, alignment of interests.
A single deal might involve 500â1,000 pages of documents, 3â5 site visits, dozens of verification calls, and weeks of analysis. Across a portfolio of 15â20 active deals, you see why PE firms have entire teams dedicated to this.
What AI Accelerates
Extraction and comparison at scale. “Compare the indemnity provisions across all 12 JDAs (Joint Development Agreements) in our portfolio.” Currently, a team member reads all 12 agreements, extracts the clauses, writes a comparison memo. Days of work. A well-built extraction system does it in minutes â with source citations to every clause. The same logic powers portfolio-level queries: “In our last 50 deals, how many had cost escalation tied to a price index?” Weeks of analyst time manually; trivial with structured extraction.
Deviation detection. A counterparty sends a draft agreement. What changed from our standard template? What was quietly removed? What was inserted? Red-lining is tedious human work that machines do faster, more reliably, and more consistently than a tired associate at 11 PM.
Unstructured data extraction. Land records in India are notoriously messy â handwritten mutation entries, scanned certificates spanning decades, formats that vary by sub-district. AI-assisted extraction doesn’t eliminate legal interpretation, but it dramatically reduces the mechanical reading.
A 100-page commercial JDA or DSA (Debenture Subscription Agreement) used to take a senior associate 6 hours to read end-to-end. Mechanical extraction now does it in 10 minutes. That’s 97% of the reading time, gone.
What AI Doesn’t Touch
Judgment on management quality. Is this developer capable of executing a 150-acre township? Do they have the team, the capital access, the regulatory relationships? This requires sitting across a table, asking probing questions, reading between the lines. No model does this.
Market intuition. The data says commercial office absorption in this micro-market was 2 million sq ft last year. But you drove through the area and half the buildings have “Available” signs. The absorption number includes pre-leases from a single IT company that might not take possession. This kind of ground-level intelligence doesn’t exist in any dataset.
The encumbrance that wasn’t on file. A residential parcel shows clean title â no encumbrances, no pending suits. AI tags it ready for transfer. A site visit shows two old houses on the land. Three months in, a local broker mentions that the houses have been there since 1985 â over forty years of open possession. The title is clean. The land is not vacant. No document in the deal file said this. No model would have caught it. The risk surfaced before signing because someone had picked up the phone.
Creative structuring. The deal doesn’t work at standard terms. Tax implications require a different entity structure. Stamp duty makes direct transfer prohibitive. The landowner wants revenue share but the developerâs / fund mandate requires equity / outright purchase. Designing a structure that satisfies all constraints is a creative act spanning finance, law, tax, and regulation. AI can flag the constraints. It can’t design the structure.
Relationship-dependent information. The most valuable diligence insights often come from informal conversations â with local brokers, government officials, competing developers, former employees. These networks are built over years and can’t be replicated by technology.
The Hybrid Model
The best deal teams of the next decade won’t be all-human or all-AI. They’ll be hybrid:
AI handles: Reading, extracting, comparing, flagging, cross-referencing. The mechanical work that requires attention and endurance but not judgment.
Humans handle: Evaluating, structuring, negotiating, judging, deciding. The intellectual work that requires creativity, experience, and risk appetite.
The handoff: AI flags the deviations; humans decide which ones matter. AI extracts the financial data; humans decide what story it tells. AI identifies the risk factors; humans set the risk appetite.
A deal team that spends 60% of its time reading and 40% thinking inverts to 20% reading and 80% thinking. That’s a better team. That’s better decisions. That’s better returns.
AI reduces reading. It does not reduce responsibility.
Who Builds This?
The person who designs the AI-assisted due diligence workflow can’t be a technology vendor â they don’t understand what matters in a deal. It can’t be a consultant â they’ll give you a framework but not an implementation. It has to be someone inside the organization who has done enough deals to know what’s signal and what’s noise, and enough technology to know what’s possible.
The difficult part is not building the AI layer. It’s knowing what deserves human attention.
The future of deal-making belongs to teams that combine deep domain expertise with technology literacy. Not either-or. Both.
Hashtags: #DueDiligence #AIStrategy #PrivateEquity #RealEstate #EnterpriseAI #AILeadership
Rohit Nanda | CA, CFA, MBA, LLB