Why AI in Private Capital Requires a Partner, Not Just a Platform
Last updated 4 June 2026
Microsoft and EY just committed $1 billion to do in enterprise what Xlagent has done in private capital from the start: put engineers and domain experts in the room with clients until AI actually works.
Key takeaways:
- 80% of enterprise AI projects fail to deliver intended business value (RAND, 2025)
- Microsoft and EY are now co-locating engineers with clients to bridge the pilot-to-production gap
- OpenAI's Frontier Alliance with BCG, McKinsey, Accenture and Capgemini confirms that technology alone cannot drive organizational change
- In private capital, regulation, document complexity and workflow specificity make hands-on implementation non-optional
The industry has confirmed the model
In May 2026, Microsoft and EY announced a $1B+ joint initiative to move enterprise AI from experimentation into governed production. The mechanism: Microsoft Forward Deployed Engineers co-located with EY industry professionals inside client operations, across fifteen countries. The problem they named explicitly is the one most private capital firms already know: most enterprise AI spend is currently stranded between experimentation and production.
Three months earlier, OpenAI launched its Frontier Alliance with BCG, McKinsey, Accenture and Capgemini. The rationale OpenAI gave: "technology alone cannot drive organizational change." OpenAI's own Forward Deployed Engineering team now embeds with consulting partners to integrate AI into client workflows, not to sell subscriptions.
OpenAI went further still with its Deployment Company, bringing together nineteen partner firms including TPG, Bain Capital, Goldman Sachs and Warburg Pincus alongside consulting alliances with Bain, McKinsey and Capgemini. The stated 2026 focus: practical adoption, not experimentation.
The pattern across all three is the same. The AI vendor provides the technology. The implementation partner provides the depth. Xlagent combines both, built specifically for private capital.
Why AI projects fail
RAND Corporation's 2025 analysis of more than 2,400 enterprise AI initiatives found that 80% failed to deliver intended business value, roughly double the failure rate of non-AI IT projects. McKinsey's 2026 Global AI Survey found that 73% of enterprise AI deployments failed to achieve projected ROI. Writer's 2026 enterprise adoption survey puts it plainly: 79% of organisations face challenges adopting AI, and 48% describe the process as a massive disappointment.
The cause is not the model. An analysis of 140 enterprise AI implementations found that only 23% of failures came from model performance, data quality or integration complexity. The remaining 77% came down to strategy, governance and change management. You cannot fix a 77% problem with a better subscription.
Why private capital is harder
Most enterprise AI products are built for large US corporates with dedicated IT transformation teams and clean, structured data. Private capital firms are something else entirely.
The workflows are document-heavy and specific: lease abstractions, payment-flow validation, insurance certificate checks, IC memos, LP report preparation. These are not generic productivity tasks. They require an AI layer that understands the document type, the clause structure and the decision it supports.
The regulatory environment adds further weight. DORA has been enforceable since January 2025. GDPR applies to any personal data in the workflow. AIFMD II transposition deadlines passed in April 2026. The EU AI Act adds auditability and human oversight requirements to AI-assisted decisions. A COO cannot deploy a third-party AI tool without documented vendor risk assessment, contractual data handling provisions and a clear auditability trail.
Then there is the EU data sovereignty question. Data must remain within EU jurisdiction, not simply hosted on EU servers owned by a US parent company subject to the CLOUD Act.
Finally, there is the team size reality. A 50-person fund does not have an AI transformation office. The COO making the deployment decision is often also the person reviewing the quarterly LP report. Standard enterprise onboarding, built for large organisations with dedicated project teams, does not land in this context.
What the Xlagent partnership model looks like
Discovery comes first. Before configuring anything, we sit with the COO, the CFO and the analysts doing the work today. We listen and document. We do not arrive with a fixed use-case list.
Use case selection follows. We identify the 2-3 workflows that are high-volume, rule-based and document-intensive. Not every workflow qualifies, and selecting the wrong one is where most implementations fail first.
Then we map the process and the data. We document what actually happens before building anything. This step reliably surfaces blockers that would otherwise appear after go-live.
The pilot runs on real data. Not a demo environment: a live workflow, real documents, success criteria agreed upfront. At the end, the firm has measurable results, not another proof of concept.
Post-launch, we stay involved. Monitoring, edge case handling, quarterly reviews, proactive identification of new automation opportunities. The relationship does not end at go-live.
EU compliance is handled as part of delivery: all data in Azure within Xlagent's EU tenant, DORA contractual provisions, full auditability. This is not handed back to the firm's compliance team to figure out.
Why the human layer matters
The people who determine whether the platform works are the analysts and operations staff, not the executive team. If they do not trust the output, they will not use it.
Every Xlagent agent shows its work: the source document, the specific clause, the reasoning behind each output. The analyst can check, override and sign off. Under DORA and the EU AI Act, AI-assisted decisions in financial workflows must be traceable and subject to human review. An agent that produces an answer without a visible source is not compliant. Transparency here is not a design preference. It is a regulatory requirement.
There is also a commercial reality. Analysts who see the platform remove hours of manual work in the first weeks become the strongest internal advocates for expanding its scope.
| Standard SaaS AI vendor | Xlagent | |
|---|---|---|
| Implementation | Onboarding docs and help desk | Sit with your team, map your processes |
| Use case selection | You decide | Identified together in discovery |
| Data preparation | Your responsibility | Built as part of implementation |
| Pilot format | Demo or sandbox | Live workflows, real data, defined criteria |
| Post-launch | Ticket system | Ongoing monitoring and quarterly reviews |
| AI transparency | Output only | Source, clause and reasoning shown |
| EU compliance | You verify | Handled as part of delivery |
The AI industry has confirmed the principle: technology capability does not equal operational results. Microsoft needed EY. OpenAI needed four consulting firms. For a private capital firm in the EU, the right answer is a specialist partner who understands your documents, your workflows and your regulatory environment: one who defines success criteria before starting and stays responsible for the outcome after go-live.
For a detailed breakdown of each implementation phase, see how Xlagent works with clients.