How Ardian embedded Claude across the Buyout investment cycle, with JAKALA as strategic AI partner

An evolving industry landscape

For a decade, private equity returns had three engines: multiple expansion, leverage, and operational improvement. Since 2022, the first two have flipped from tailwind to headwind, as multiples have plateaued and cheap financing has disappeared. This leaves only one reliable engine: operational improvement and greater precision on every investment decision. For Ardian Buyout, AI is not an efficiency gadget: it has become an operational layer with the potential to reshape how the team sources opportunities, makes investment decisions, and creates value across portfolio companies.

The opportunity runs along both axes of value creation. Internally, the best investment decisions come from investment teams thinking critically, rigorously, early, on every deal. Process improvement is indispensable, in order to allocate maximum time to analyses of precisely selected data around both the market and target, which guides to exercising judgment, rather than expending time on internal document creation. Externally, the same discipline applies to value creation of portfolio companies, where AI can unlock operational improvement across the levers that now drive returns, from commercial excellence to procurement and reporting. In both cases, the scarce resource is identical: the time and critical thinking of experienced investors.

As the team assessed the market, screening and testing over 100 PE-tailored tools, a clear architecture emerged: Claude as the intelligence layer at the core. This was selectively complemented by dedicated external tools, plugged into the Claude ecosystem via MCP whenever they met a very specific need. However, leading technology is not enough. To extract full value, Ardian leveraged its own knowledge corpus: a proprietary 30-year history of deals, exit analyses, sector intelligence and decision criteria, spanning 100+ platform investments, 350+ build-ups, and every IC memo and lesson learned along the way, none of which can be bought or reverse-engineered. The challenge was to activate it inside Ardian's own workflows, in a secure environment, with humans always in the loop.

The Taskforce's starting conviction was that the opportunity was not a chatbot with deal documents attached; it was to reproduce, in software, the way an investment team actually reasons and thereafter, by scaling architectures and compute, expand the intelligence frontier of even the most experienced investment professionals, while reducing execution time. JAKALA designed and engineered that system on Anthropic's technology, combining the Claude models, Claude Code as the agentic harness, and the skills and sub-agent architecture it enables. This made it possible to build in a short timeframe what would previously have required a dedicated platform team.

One use case is the Diligence Sparring Partner: a system that challenges the team's investment thesis and surfaces upside angles it has not yet explored. Several architectures were tested for the job before selecting the released one. The first one was a mixture of experts — a panel of specialist agents, each an authority on one dimension of the deal, such as market, competitive dynamics, business model — researching in parallel with every claim anchored to a cited source. Their findings then fed into an adversarial debate between two agents, one arguing the investment case and one built to attack it, judged blindly by a third agent that never saw the deal materials. The second architecture was swarm intelligence: waves of lightweight agents exploring a deal in successive rounds, each leaving signals that steer where the next wave digs deeper. The pattern is inspired by pheromone trails and applied here to investment analysis for what is, to our knowledge, the first time. No human team can hold hundreds of parallel lines of inquiry open at once and let the strongest ones emerge; a swarm can.

Another is the accelerated drafting of investment documents. Grounded in hundreds of previous Ardian documents, the system reproduces the structure, tone and analytical standards the team expects, with every figure and claim traceable to the underlying deal materials. Investment Professionals review, challenge and finalize the draft rather than building it from a blank page; the quality bar stays exactly where it was, while production time has been cut by more than 70%. The investment professional has more time to do what is most value-add: exercise judgment and develop relationships.

Underneath both capabilities run a structured knowledge layer, distilled from raw deal documentation into a cited, versioned foundation. It is the single source of truth on a deal, accessible to anyone on the team; the firm's knowledge about an opportunity, once scattered across documents and data rooms, is now centralized in one place.

Across the stack, each task runs on the right class of Claude model: frontier models where depth of judgment matters, faster ones where speed and cost do. And because Anthropic shipped new generations during the build, such as Opus and Fable 5, the system was designed to absorb them; each release was adopted within days and benchmarked on real deals, and each one moved the frontier of what the architectures could attempt. The entire experience lives inside Claude Code: one environment where Investment Professionals launch, review and refine everything, without ever leaving it. It is also the environment where the system was made: JAKALA's AI engineers built it entirely through Claude Code itself, a compounding speed that allowed them to take both use cases described here from strategy to production in a matter of months.

Several more use cases are already under way, spanning pipeline generation and screening, M&A screening, and opportunity assessment.

The system runs entirely on Ardian's own infrastructure, governed by its existing enterprise permissions: no new servers, no new authentication surface, a design validated with Ardian's IT and Cybersecurity teams. JAKALA's role ran the full arc, from use-case strategy with the Taskforce, through architecture and implementation, to production deployment, documentation and training. What was handed over was not a tool, but a capability the team operates itself.

The Taskforce began its journey in 2026 by mapping 12 priority use cases across the three phases of the investment cycle: deal sourcing and screening, due diligence and investment process, portfolio monitoring and reporting. For each one, it mapped the data available, the process in place and the tools in use, then rethought the workflow around what AI makes possible rather than layering AI onto the old way of working. By summer, the first Claude Skills, complemented by the selective integration of external market tools, were live on Ardian's Claude Marketplace and tested on real opportunities — addressing identified use cases spanning from investment document drafting, in-depth due diligence sparring, and legal work.

For each use case, Ardian's knowledge corpus is organized through a Wiki Builder: a modular but highly practical approach that turns raw deal documentation into a structured, reusable, and verifiable knowledge base — one that compounds over time, as every deal processed enriches the foundation. In every case, investors review, edit and finalize the output; the system augments judgment, it does not replace it.

Adoption is driven from within through the Buyout AI Center of Excellence, a structured bi-weekly forum whose mission runs across both axes of the strategy: collecting real-world AI use cases and needs from portfolio companies, sharing knowledge and driving adoption across the whole Ardian Buyout ecosystem, so that a use case proven in one company can be replicated to the next. Each session puts one or two concrete use cases in front of the team, from optimizing a salesforce's geographic coverage to white-space analysis built on public or proprietary data sources. It is also where the team's own capability compounds; all investment professionals can build Claude Skills with no coding required, while a group of heavy builders across teams uses Claude Code for more advanced use cases — and what is learned internally feeds the portfolio, and vice versa.