Where AI actually helps a small fund team today, where it is getting useful, and where it is mostly hype. A practical guide from someone building it.
Every fund administration pitch deck now has an "AI" slide. Most of them are aspirational at best. As someone who builds fund management software, I want to give you a more honest picture of where AI delivers real value today, where it is getting useful, and where the industry is overselling.
Where AI helps right now
Document processing and data extraction. This is the most mature and immediately useful application. Fund operations involve enormous volumes of unstructured documents (subscription agreements, K-1s, account statements, portfolio company reports, compliance certificates). AI models can extract structured data from these documents with high accuracy, eliminating hours of manual data entry. We use this in our own operations and the time savings are measurable.
DDQ and RFP response drafting. Due diligence questionnaires are a fact of life for fund managers. LPs send them during allocation decisions, and many of the questions are repetitive across questionnaires. AI can draft initial responses based on your previous answers, your fund documents, and your operational data. A human still reviews and approves every response, but the drafting time drops from hours to minutes.
Investor communication summaries. When you have dozens of LP interactions across email, calls, and meetings, AI can synthesise these into structured summaries (flagging action items, tracking commitments, maintaining a record of what was discussed). This is useful for investor relations teams managing multiple fund vehicles.
Where AI is getting useful
Portfolio company analysis. AI is getting better at processing financial data from portfolio companies (normalising different reporting formats, identifying trends, flagging anomalies). This is not yet reliable enough to replace human judgment, but it is getting good enough to be a useful first pass that highlights what deserves attention.
Deal sourcing signals. Parsing news, social media, patent filings, and job postings to identify companies at inflection points. The signal-to-noise ratio is still low, but the tools are improving. Most useful as a supplement to your existing sourcing channels, not a replacement.
Where AI is mostly hype (for small funds)
Fully autonomous fund operations. No AI system today can run a fund's back office without human oversight (processing capital calls, calculating NAVs, filing regulatory reports). The stakes are too high and the edge cases too complex. Claims of "autonomous fund administration" should be treated with extreme scepticism.
AI-driven investment decisions. For quantitative strategies with massive datasets, machine learning has a legitimate role. For a venture capital fund making fifteen deals a year based on founder quality, market timing, and competitive dynamics, AI is not making your investment decisions. It can inform them, but the judgment remains human.
How to evaluate AI tools
Start small. Pick one manual process that is high-volume and relatively standardised (document data extraction is often the best starting point). Implement an AI solution for that specific task. Measure the time saved and error rate. If it works, expand to the next process. If it does not, you have lost very little.
Be sceptical of vendors who cannot give you specific, measurable results from existing clients. "AI-powered" is a marketing term. "Reduced document processing time by 70% for Fund X" is a verifiable claim.
I get pitched AI fund admin tools every week, and most of them are a wrapper around a chatbot. The honest version of what we actually build is less glamorous. We use models to read documents, extract numbers, and check our work. That alone removes thousands of hours of manual entry per year. The chat is incidental.
What we are doing at Infra One
We are integrating AI into Backbone in areas where it delivers clear, measurable value: document processing, data extraction, DDQ assistance, and investor communication tools. We are not bolting on an AI chatbot and calling it innovation. Every AI feature we ship has a specific use case and a measurable improvement metric.
