What we’ve learned so far from the 100,000+ AI requests made through Juniper Square
When we launched Headless GPX in June, we knew we were giving GPs a powerful tool to securely extend their firm’s data into AI and agents. What we didn’t yet know were all the ways our customers would use AI to speed up and improve core workflows.
Four months later, GPs at more than 200 firms have made over 100,000 tool calls through Headless GPX. Each call represents an AI client, like Claude, Copilot, ChatGPT, or an agent the firm built itself. These clients are reading data, including things like an investor’s commitment history from Juniper Square, to complete tasks.
Many firms start their AI journey with investor tearsheets and meeting prep, as these are the kinds of questions and answers most AI vendors demo. However, those use cases make up a small share of the total volume of GPX calls. What we’re seeing instead is that most calls come from jobs where one request leads to hundreds of lookups, like checking an entire outreach list against the CRM or confirming K-1 delivery for each investor in a fund.
Here’s what those high-volume jobs look like in practice.
How firms are using Headless GPX
Outreach screening. This is the largest use case. Before an IR or business development team sends a LinkedIn invite or an email, an agent checks each name against the CRM to see whether the firm already has a relationship, who covers it, and whether the contact has been flagged. Some firms re-screen the next day's outreach list every morning. Others require a prospect to clear a series of checks before anyone reaches out.
Scheduled jobs. A large share of activity runs nightly, daily, or weekly without anyone typing a prompt. Agents refresh IR dashboards and master trackers; some save snapshots of investor data for later review. One fund manager runs a nightly reconciliation that compares its positions in Juniper Square against its LP master tracker and its shared drive, then flags anything that doesn't match.
Keeping other systems current. Firms use agents to move Juniper Square data into other systems such as accounting software and market data platforms. Before Headless GPX, this usually meant a custom integration project or someone exporting and re-uploading files on a schedule.
Tax and document audits. These are common too. A multifamily real estate manager checks every fund entity to confirm each investor received a K-1, and flags positions missing payment instructions. Another firm ties out 1042-S withholding forms across several funds. A venture firm pulls positions and capital account balances across its entities for year-end audit confirmations.
Fundraise monitoring. This is the fastest-growing category. Firms build daily briefings on where each prospect stands in the subscription process, and check data room activity several times a day to see which investors are engaged. A private credit manager runs a regular digest of subscriptions still waiting on a signature or countersignature.
The list of one-off uses are varied. A farmland operator pulls position data for each ranch entity to build cap tables for a lender package. Another firm filters its CRM by custom fields to find debt funds that lend on hotel construction. Several firms use it to build their holiday card lists.
What the most AI-forward firms have in common
Most GPs are experimenting with AI, but fewer have built it into how they operate. The ones who have share a few practices in how they use Headless GPX and other AI tools.
They've grown past a single champion. In the early weeks most firms had only one enthusiastic person, but today more than half have multiple users actively building. Once colleagues start running the same workflows, the team has usually started to see real value.
They run one job at scale. Nearly every one of the most active firms has a single workflow that makes up most of its volume, such as an outreach screen, a K-1 audit, or a weekly position count. We expect this to quickly grow to multiple jobs on average.
They run that job on a schedule. About half of the most active firms run at least one job daily, nightly, or weekly. Many of these jobs surface what changed since the last run, like new prospects in a raise or new positions in a fund, so the team gets the update without asking for it. Once a team runs autonomous workflows, usage tends to climb quickly because the work no longer depends on someone remembering to start it.
They write the process down. Several of the most active firms give their agents written instructions, with numbered steps and checks that have to pass before the agent moves on. Firms already document their own operating procedures this way, and the most active ones have extended the habit to their agents.
See where your firm stands
If your firm is still at the stage of one person experimenting, that's normal, and it's also the point where a clear next step helps most. Juniper Square’s new AI Assessment is built to help you understand what steps will help you take AI firm-wide. It places your firm on a six-level scale, from no AI activity to a firm-wide AI strategy in execution, and explains what the next level up requires. For example, one of the steps on that scale is the same shift described above: moving from one person building tools for their own use to a team that shares and maintains them.
The Assessment is just 13 questions and takes about 5 minutes to complete. You’ll get a detailed report that rates your firm on the eight areas it needs in place to be AI-ready, including things like governance, systems, and talent, along with three recommended changes to make in the next 30 days.