Why has the AI section of a DDQ grown so much?

Monel:Grant, not long ago the AI section of a DDQ was two or three questions. What are you seeing now?

Grant:Pages of them. AI moved from experiments to real workflows, and the questions followed. Allocators used to ask "do you use AI?" and "which model?" Now they ask how it is governed and what it can access. Managers feel that shift in every response cycle.

Is "which model?" still worth asking?

Jose:Monel, is "which model?" still worth asking?

Monel:Yes, with follow-up. Most firms use several models, because different models suit different tasks. Model choice affects data handling, cost, resilience, and concentration risk. It is not a proxy for maturity. The better questions are: why that model, where is it used, and can you switch if pricing, performance, or availability changes?

Are asset managers building AI or buying it?

Monel:Grant, who is building AI internally, and who is buying?

Grant:I see two camps. In about 2 in 10 conversations, the manager has hired AI engineers and data scientists, and RFP and DDQ tools are often their first project. The other 8 in 10 also have a mandate from the top to use AI. But they buy for work like RFPs and DDQs. They point their own build effort at what drives returns, which is their core business.

What does a build versus buy answer reveal?

Grant:Jose, allocators are now asking managers how they make build versus buy decisions. What does the answer tell you?

Jose:It shows management discipline and how the firm thinks about ROI. A strong answer is: "We build where proprietary data, models, or processes create differentiated investment advantage, and partner where specialized platforms already provide domain expertise, data connectivity, and infrastructure." A weak answer is: "We build everything because AI is strategic." AI has made building faster, so allocators should expect more of it. The question is whether it is aimed at the right work.

Do managers know which AI tools are in use?

Jose:Whether a firm builds or buys, someone has to know which tools are in use. Grant, is there a formal approval process and an inventory?

Grant:For vendor tools, yes. Most managers who adopted AI through us took it to an AI committee first, and I have sent them our AI FAQs to help with that review. Governance is real once a tool touches confidential data, supports a material process, or communicates externally. What I don't see tracked the same way is the AI that employees build themselves. That may be where the least visible risk sits.

Is citizen coding a real risk?

Grant:Monel, citizen coding is growing fast. Is that a real risk?

Monel:Potentially. The tools are getting easier faster than governance is adapting. An employee can now build an agent that touches internal data or connects to external APIs or MCPs without any development process. That is a new form of shadow IT. The risk is not citizen development itself, which can be very valuable. The risk appears when a prototype quietly becomes part of a production process. Who tested it? Who owns it if the employee leaves? What can it access? What happens when the underlying model changes?

Can one AI solution serve IR and Legal & Compliance?

Jose:Monel, how are firms managing AI across departments? Could a tool deployed in Investor Relations share data or workflows with Legal and Compliance? Or are those separate use cases needing separate platforms?

Monel:Both patterns exist. Some firms would prefer specialists especially when they are larger. While smaller firms with less mature functions will likely opt for general purpose solutions.

Should allocators care if managers license their data for AI training?

Grant:Monel, you recently shared a post about firms approaching DV to license data for AI training and evaluation. Should allocators care if their managers take part?

Monel:Yes. It raises a question of information ownership. A firm can monetize its data and expertise. But it should know exactly what it is giving away, what rights the recipient gets, and whether client, portfolio, investment IP, or employee information could be involved. Institutional knowledge is becoming more valuable because models need high-quality, domain-specific judgment. Firms need a clear policy on what can be commercialized and what stays proprietary. It is an additional governance question.

Is "AI saves time" enough to show ROI?

Grant:Jose, every manager says AI saves time. Is that enough?

Jose:No. Hours saved are useful, and every manager will quote it. But it is not the same as value. If someone saves five hours a week and nothing changes in capacity, cost, quality, or output, the benefit is hard to show. The question is what happened to the saved time. Did the team review more opportunities, cut external spend, move faster, make fewer errors? Stronger firms tie AI to results, not just hours on paper.

Monel:That is why we built an ROI calculator for clients. It gives both sides a shared way to measure it.

Does AI shift the burden from creating content to reviewing it?

Monel:Jose, do you see AI shifting the burden on teams from generating content to reviewing it?

Jose:Yes, and it is easy to underestimate. Drafting gets cheap. Judgment does not. The work moves to review: checking accuracy, sourcing, and tone before anything goes out. Firms should plan for that. Reviewers need clear standards, and the review step needs an owner. Otherwise, the time saved in drafting comes back as time spent fixing.

What happens if the AI model provider goes down?

Grant:Monel, say a firm has embedded AI in important workflows and its model provider goes down. What should happen?

Monel:This is a real continuity question now. The firm should know which workflows depend on AI and what the fallback is. That could be another model, another provider, a degraded mode, or a human process. And the fallback must be tested. Can you switch under real production volumes? Are the integrations ready? Have you checked output quality after the switch? A fallback that exists only on paper is not much of a fallback. Recent outages make the point. When many firms rely on a few providers, one failure can reach all of them. If your plan assumes only one narrow component fails at a time, test that assumption.

How should firms handle employee fears about AI?

Monel:There is a human side too. Jose, what are you hearing from teams?

Jose:AI still worries many employees, and the fear of job loss causes real anxiety. Firms are struggling with that cultural shift. It is a blind spot in most DDQs. Firms focus on the tech stack and forget that change management decides whether adoption works. If staff see AI as a threat, they resist it or work around it. Allocators should ask how management supports the team, addresses those fears, and retrains people toward higher-value work.

What does it mean when a manager declines to answer an AI DDQ?

Monel:Grant, some managers are declining to answer detailed AI DDQs because the space moves so fast. What does that signal?

Grant:It is telling. If a firm has not deployed AI beyond basic tools, saying so is better than manufactured answers. But if a firm uses AI across critical workflows and will not answer because governance has not caught up, that requires follow-ups. The concern is when they cannot explain governance, ownership, approved use cases, and controls at an appropriate level.

What is the one thing allocators should understand about a manager's AI program?

Grant:Jose, if you had to boil all this down to one thing an allocator should understand about a manager's AI program, what would that be?

Jose:Whether adoption is deliberate. Experimentation is normal. But once AI touches important data, decisions, or workflows, it has to give way to governance.

Monel:For me, it also comes down to dependencies. Which data, models, vendors, employees, and workflows does AI now rely on?

How should managers prepare for AI due diligence?

Monel:And for managers preparing for these conversations, Grant?

Grant:Two things. First, do not treat it as a compliance checkbox. The strongest answers come from firms that have worked through these questions in practice, not firms reciting a policy. Second, do not oversell. A firm that says "We are early. Here are our three approved use cases, here is how we govern them, and here is what we are testing next" is more convincing than one that claims AI is everywhere but cannot explain how it is controlled.

Three Questions Worth Asking From Either Side of the Table

Three questions worth asking from either side of the table. They come from the 25-question AI Due Diligence Questionnaire DiligenceVault built for clients.

1. Is there a live inventory of every AI tool touching confidential data, including those employees built themselves?
2. If your primary model provider went down for three hours tomorrow, what happens? Has the fallback been tested?
3. What did the time AI saved actually get used for?