How Do You Ensure Ownership of Agentic AI Workflows During an M&A Deal?

    How Do You Ensure Ownership of Agentic AI Workflows During an M&A Deal?
    Simone Vascotto··7 min read

    Key Takeaways

    • 1.Agentic AI is spreading through deal teams faster than anyone is assigning ownership for it. The risk isn't the tool. It's that nobody owns what it produces.
    • 2.Every agent output that reaches an IC memo needs a named human owner per workstream. Accountability doesn't transfer to software.
    • 3.The workflow is an asset too. Someone has to own its maintenance and versions, or it leaves with the associate who built it.
    • 4.Seller data comes with NDA and data room terms. Know where an agent sends it before the first document is uploaded.
    • 5.Ownership has to be handed over at each deal stage, especially from the deal team to the integration team.

    Short answer

    You ensure ownership of agentic workflows by assigning four owners before the agent runs: a named person who owns each output, an owner for the workflow itself, an owner for the data it touches, and the person who signs the decision. Then you make ownership travel with the deal, from pre-LOI screening through diligence, close and the handover to integration. The agent does the work. A person answers for it.

    I spoke with Matthew Tigwell, founder of Fulcra AI, an AI diligence platform. We covered a lot of ground in our earlier piece on the contract clauses that can break an AI synergy thesis. Ownership never came up by name. But almost everything he described is an ownership question in disguise, so this piece builds the framework from his observations.

    Why is ownership the missing piece in agentic deal workflows?

    Start with how AI actually enters a deal team today. In Matt's words, "a lot of firms are just giving Claude to their associates and their deal team." And then: "there's a couple people, typically younger, on the deal team who are starting to build out skills and starting to build solutions internally a lot of the time."

    That is a healthy start. But the firm's diligence method is now partly encoded in workflows nobody formally owns. Matt estimates AI handles roughly 20 to 25% of the diligence workload today, mostly in an elementary way. As that share grows, so does the unowned part.

    Agents raise the stakes. Anthropic describes agents as systems that "dynamically direct their own processes and tool usage" and warns that "the autonomous nature of agents means higher costs, and the potential for compounding errors" (Anthropic, Building effective agents). Matt sees the same thing on scope: "if you hand it a whole data room of 500 documents or 500 pages of documents, you know, you're gonna get a ton of hallucinations." The team's reaction, as he put it: "When I double check it, it's wrong half of the time."

    The opposite failure is just as common. In our conversation I described what I see on the people side: "They just trust everything." The EU AI Act names this risk directly. Its human oversight article, written for high-risk systems, asks that the people overseeing AI stay aware of "automation bias," the tendency to over-rely on its output (EU AI Act, Article 14). Most deal tools won't fall in that legal category, but the principle holds: an owner neither dismisses the output nor trusts it blindly.

    Who owns an AI agent's output and decision?

    One named person per workstream: financial, legal, commercial, IT, HR and people. That person owns every agent finding in their area. They check it against the source, decide whether it is material and put their name on what goes into the IC memo.

    The NIST AI Risk Management Framework makes this a core governance outcome: "Roles and responsibilities and lines of communication related to mapping, measuring, and managing AI risks are documented and are clear to individuals and teams throughout the organization" (NIST AI RMF 1.0, GOVERN 2.1).

    For lawyers it is already a professional duty. The American Bar Association's Formal Opinion 512 says AI tools "cannot replace the judgment and experience necessary for lawyers to competently advise clients," and that the right level of verification depends on the tool and the task (ABA Formal Opinion 512). Deal teams outside the law firm should hold themselves to the same standard. Humans sign off. AI doesn't.

    Who owns the workflow itself?

    This is the ownership question most firms skip. A diligence skill, prompt chain or agent configuration is a piece of institutional method. It needs an owner who:

    • Maintains it. Updates it after each deal post-mortem, when a check was missed or a clause type turned out to matter.
    • Versions it. Records which version ran on which deal, so a finding can be traced back to the method that produced it.
    • Documents it. Writes down what it does, what it doesn't do and how it was tested, so it survives staff turnover.

    The institutional memory problem is real. If the junior associate who built the firm's best data room workflow leaves the deal team, the workflow often goes with them, or stays behind with nobody who understands it. I've written about how to build Claude Skills so the work is done once and reused on every deal. The step that matters most is giving each skill an owner.

    There is also a legal layer: who owns the workflow's IP. If it was built in-house, the firm should own it explicitly. If it runs on a vendor platform, the contract should say who owns the configuration, the outputs and any improvements built on your deal data.

    Under the Hood of M&A

    Get articles like this one weekly.

    Subscribe on Substack

    Who owns the data an agent touches?

    The seller does, and the buyer holds it under the NDA and the data room terms. Before an agent reads a single document, someone has to own three answers. Do our NDA and VDR terms allow this tool to process the data? Where does the data go, and is it retained or used for training? Who can see the outputs?

    Data security is already the top concern among dealmakers using AI. In Deloitte's 2025 GenAI in M&A Survey of 1,000 US corporate and private equity leaders, 67% named data security as a leading barrier, followed by data quality and availability at 65% (Deloitte).

    Data quality is an ownership issue too. In smaller deals, Matt describes sellers drip-feeding documents over months, which means the analysis has to rerun as each batch arrives. Someone has to own the question "what has the agent seen so far?" Otherwise an early, incomplete finding hardens into a conclusion.

    How should ownership move across the deal stages?

    Ownership fails at handoffs. Matt notes that AI diligence work is moving earlier, into pre-LOI screening, and that bolt-on volume is now too high for deal teams to diligence every add-on manually. More stages and more deals mean more handoffs. The table below is a starting RACI. Adjust the roles to your firm.

    Deal stageAgent workResponsible (runs and checks)Accountable (signs)ConsultedInformed
    Pre-LOI screeningTarget screens, outside-in research, first-pass risk flagsAssociateDeal leadSector partnerInvestment committee
    Post-LOI diligenceData room review, clause extraction, financial checks, contracts read against the thesisWorkstream ownersDeal leadExternal advisors, legal counselOperating partner
    IC and signingIC memo sections, findings summary, open questionsDeal leadInvestment committeeWorkstream ownersIntegration lead
    CloseFindings handover pack, unresolved risks, data disposal or transferDeal leadOperating partnerLegal counselPortco management
    Integration and portco100-day plan drafts, milestone tracking, status consolidationIntegration leadOperating partnerPortco management, workstream ownersInvestment committee

    Two rows deserve attention. At close, hand over the workflows and their versions along with the findings, so the integration team knows what was checked and how. In integration, ownership shifts to people who weren't in the data room. I called this the knowledge chasm in The Handover Problem. Agentic workflows can narrow that gap, but only if they are handed over with an owner.

    Build or buy: who owns what?

    Matt's view is direct: "it makes a lot more sense to buy rather than build" for most firms. He also sees large firms building AI centers of excellence, mostly focused on portfolio companies, and he flags the security risks of vibe-coded tools.

    Each path moves ownership differently:

    • Build. You own the workflow, its IP and its maintenance. You also own every failure, the security review and the key-person risk of whoever built it.
    • Buy. The vendor owns the platform and its upkeep. You still own the outputs you rely on, the data you upload and the decision. The contract decides the rest.

    NIST treats this as a governance item in its own right: policies and procedures should address "AI risks associated with third-party entities" (NIST AI RMF 1.0, GOVERN 6.1). Buying a tool doesn't buy away accountability.

    How do you assess an AI vendor on ownership?

    Matt's criteria are a good start: real security certifications, a whole team with experience building platforms (not just the founder), and M&A practitioners informing the product. Add the ownership questions:

    • Who owns the outputs, configurations and any improvements built from our deal data?
    • Is our data retained, and is it used to train models?
    • Does every finding cite the document and page it came from?
    • Is there a log of what the agent did on each deal?
    • What happens to our data and workflows if we leave?

    Where do documents stop?

    Even a well-owned workflow is bounded by its inputs. Matt is candid about this: "we're completely limited by what we see in the documentation provided by the seller." Documents show what a company owns and has signed. They don't show how teams decide, coordinate and handle change, which is where many value creation plans succeed or stall. That question needs its own review, run at team level, with its own owner.

    An ownership checklist for agentic deal workflows

    1. Name an owner for each workstream's agent outputs before diligence starts.
    2. Name an owner for each workflow or skill, with a version log.
    3. Confirm NDA and data room terms allow the tool, and record where data goes.
    4. Require a source for every finding: document, page, clause or ledger line.
    5. Test each workflow on a closed deal before using it on a live one.
    6. Track which documents the agent has seen, and rerun when new ones arrive.
    7. Keep a log of what the agent did and what a human changed.
    8. Hand over findings, workflows and open risks at close, with named receivers.
    9. Agree data disposal or transfer when the deal closes or dies.
    10. Review vendor contracts for output ownership, retention and exit terms.

    About the Author: Simone Vascotto is the Founder of HumanAQ, the AI-powered platform for Cultural Due Diligence and leadership alignment in M&A.

    This piece is part of the Under the Hood of M&A newsletter.

    Frequently Asked Questions

    Who is accountable when an AI agent gets a diligence finding wrong?
    The person who signed off on it. The tool doesn't carry accountability. That is why each workstream needs a named owner who checks agent findings against sources before they reach the IC memo.
    What happens to an AI workflow when the person who built it leaves?
    Unless it has a named owner, documentation and version history, it usually stops being maintained or leaves with them. Treat workflows as firm assets with an owner, not personal tools.
    Can we upload seller documents to an AI tool?
    Only if your NDA and data room terms allow it, and you know where the data goes, whether it is retained and who can see the outputs. Confirm this before the first upload.
    Should PE firms build or buy AI diligence tools?
    It depends on scale and resources. Building gives you ownership of the workflow and all of its risk. Buying shifts platform upkeep to a vendor but leaves you accountable for outputs, data and decisions.