## **Key Takeaways**

1. Sourced candidates make up only 4.8% of applications on average but account for around 16% of hires. Companies that have really leaned into their sourcing motion see 30–50% of hires coming from outbound.
2. LLMs have fundamentally changed AI sourcing — platforms can now read the full candidate profile, research additional context, and make a more in-depth assessment of fit in a largely autonomous way, rather than just ranking based on ML pattern matching.
3. JuiceBox's search splits your job description into two categories: hard filters (job title, location, skills) that every candidate must meet, and criteria (for ranking and assessment) that identify who is the best possible fit among matches.
4. The Ashby integration surfaces which candidates in your search results are already in your ATS — showing their last stage and last interaction date — so you can choose to focus exclusively on net-new candidates and avoid reaching out to people you've already engaged.
5. JuiceBox agents operate autonomously — finding a new batch of profiles every day for a given role. Running both manual search and agents in parallel produces a combined, non-overlapping list that is stronger than either approach alone.
6. Auto export ensures that all profiles found in JuiceBox — whether shortlisted, emailed, or responded to — automatically sync back into your Ashby instance with full profile details, contact information, and outreach history, keeping the ATS as the single source of truth.
7. If you could reduce your sourcing time by 50% (a conservative estimate), the teams seeing the biggest impact are reinvesting that time in deeper talent mapping and more meaningful candidate conversations.

## **Transcript**

*Note: transcript is auto-generated and may contain minor inaccuracies.*

 **David Paffenholz**

Super excited for the session today. We're going to be talking sourcing and how sourcing works amazingly with Juicebox plus Ashby. To start off, I'll do a quick intro. I'm David, co-founder of Juicebox, we're an AI recruiting platform. Really what we're focused on is your top of funnel — can we help you find better candidates and plug them into your pipeline? Today we'll be talking about why outbound matters, we'll hop into a live product demo exploring how Juicebox and Ashby work together, including some of our newest features on personalization. And then we'll have a few questions that you can take away with you and think about how you want to construct your outbound funnel in the future.

**David Paffenholz**

Ashby publishes some really amazing data. Sourced candidates make up a disproportionate share of hires. We see about 4.8% of candidate applications are sourced — far majority being inbound — but they make up around 16% of hires. This is across a pretty large sample set. What we see with some of our customers who are really leaning into their sourcing motion is they can see 30, 40, 50% of hires made through sourced, especially when the company is in a rapid growth phase and really focused on maintaining that quality bar for talent while rapidly scaling. So if it's clearly a solution we should be considering, why isn't everyone doing outbound today? First, manual search is pretty tricky — we've all used LinkedIn Recruiter, we've all manually gone through thousands of profiles, and we then still spend time reviewing each one, assessing in our head whether this person is a good fit. A lot of the actual engagement we do is also manual. There's been a category of AI sourcing for a while, but it was actually really hard to build before LLMs. What that really meant was an ML-based solution — platforms were ranking lists of candidates, essentially a better form of a potential ranking algorithm. What we can do now with LLMs is a lot more akin to what we do when actually looking at a profile: we can read the full profile, research more context on that candidate, and make a more in-depth assessment of whether that profile is a good fit in a largely autonomous way.

**David Paffenholz**

Let me do a quick overview of what Juicebox offers today. Juicebox is split into three main capabilities. First, search — can we help you find the right person, using a combination of hard filters with a set of criteria which we use to assess and rank profiles? Second, talent insights — what is the market map of the role that you're looking for? And then finally, outreach to engage those profiles. Now that workflow can be done in two ways: either by the recruiter or sourcer, or by an AI agent that is operating autonomously for that role, finding a new batch of profiles every day. We recommend a hybrid approach where you deploy both agents and manually work on those roles, because those two different strategies will net you a combined different list than if you're just investing in one of them. And to complete the picture and tie that into Ashby: this results in quite a neat funnel, from sourcing to outreach to then tracking interviews and getting to offer stage within Ashby.

**David Paffenholz**

So we're starting off in Juicebox in our search interface. We can either start off with a prompt, or if we have our ATS connected, we'll just open up this job description field, select it directly from Ashby. I have a backend engineer role — I'll just select that in here and automatically pull in that job description. Juicebox will start by reading that job description and splitting it into two different categories: hard filters — in this case, they should have a software engineering job title or similar, they should be located in San Francisco, and some skills inferred from the job description like Node.js or OpenSearch. The second thing Juicebox does is set a list of criteria — how can we assess these profiles and figure out who is the best possible fit? The first two criteria are generated directly from the job description, like having built something from scratch and having experience with AWS. But Juicebox also remembers criteria that I have manually added in previous searches and will tailor my criteria based on that — shown with the small brain icon. I like these criteria, so I'll click run search. Within just a few seconds, we'll see our total pool assessed and ranked. We'll also notice the Ashby icon at the top — of 16,000 potential matches, 300 already exist in my ATS. I can either focus on just those ATS matches, seeing the last stage they went through and when they last applied, or only focus on net new candidates — people that are not yet in my ATS.

**David Paffenholz**

Let's say I like these first three profiles. I'll select those three and add them directly into an email sequence. Traditionally, sequences are pre-built, but with Juicebox I'll just click create a sequence and have Juicebox generate this using AI. This ties into the Ashby integration — Juicebox knows what role I'm applying for, can actually insert that application link into the email sequence autonomously, and remembers some details about Juicebox as a company and why someone might want to work there. I can also input my Ashby scheduling URL or any other scheduling software I'm using. Just like the criteria, sequence generation remembers your edits and previous sequences, so depending on your style or the way you write emails, sequences will adjust based on that as well. We'll get a preview — backend engineer at Juicebox, a brief pitch on why they might want to work there, and a link to my Ashby application URL right here. In this case, I only did it with three profiles, but we can do the same workflow with dozens, if not hundreds of profiles at any given point in time.

**David Paffenholz**

There's one final feature I'd like to highlight — our auto export for those candidates back into Ashby. One of the biggest struggles we see with implementation of recruiting software on the sourcing side is having uniformity across the team. Are people actually using the same sourcing tools, or are people using a combination? With auto export, we can ensure that all the profiles found in Juicebox are automatically exported back into the ATS, ensuring data consistency and that we're still using the ATS as our source of truth. Any profile that you have saved, emailed, or gotten an interested response on in Juicebox is going to automatically sync back into your Ashby instance, including the full profile details, contact information, and outreach history, which Ashby pulls in from there. As a reminder, Juicebox sources over 850,000,000 candidate profiles, and so all of that profile data is at your fingertips, and ready to be synced back into the ATS. 

**David Paffenholz**

So far we've talked through what the manual version of this workflow looks like. We go from search, to evaluating profiles, to reaching out to profiles, and exporting them. That same workflow is also possible with agents. You start off with a brief calibration where you describe who you're looking for, and the agent will ask you questions back — a bit like interacting with a recruiter or sourcer who's really trying to understand the role. From there, agents go out, find an initial batch of profiles, ready for review by the recruiter or sourcer working on the role, and can ultimately reach out to those profiles autonomously as well. Because it's all happening in the same platform, your manual search and your agents are not conflicting with one another — if you're finding a profile manually, you know if the agent has already reached out, and vice versa. A lot of teams are already using this workflow today. I’ve selected a few teams that are outbound-heavy, and focused on sourcing. In some cases, like with Starbridge, over 90% of roles are being filled through outbound. We see this being particularly strong for companies that are still developing their talent brand, where inbound applications may not be as strong. There’s a greater dependency on outbound, but still maintain a high bar for the talent that they are finding. Starbridge has leaned into the concept of Talent Engineering. We think of this as establishing a systems process to find talent, rather than pure search. One way of doing this is by setting up a number of Agents, think 20 or 30 agents, running in parallel to fill those roles. We've seen two different approaches. One is the sourcer or recruiter sets up the agents, or they have a dedicated talent engineer whose job is to manage all of these agents across all roles, so sourcers and recruiters are focused on the individual roles instead. 

**David Paffenholz**

There's a couple questions I want to leave the group with. First, what percentage of your hires are coming from outbound sourcing today? We saw averages around 16 or 17% — we see companies that have over 50%, or in Starbridge's case, 90%. Second, what is the speed at which those candidates move through your pipeline? Teams that are more heavy on outbound tend to have a faster recruiting process in general. And finally, if you were able to reduce your sourcing time by 50% — and I think 50% is somewhat conservative — what can you do with that extra time and capacity? We're seeing teams investing into deeper talent mapping, and others investing more into candidate conversations with the amount of time that’s able to be spent with these candidates. We think AI-first outbound recruiting is possible today. LLMs have unlocked that for us. Our mission at Juicebox is to make that product and workflow accessible to everyone. Thank you so much.

See how Juicebox's AI sourcing engine works alongside Ashby to help talent teams discover and engage net-new candidates outside their existing pipeline. Presented in partnership with Juicebox.

AI Outbound Sourcing with Ashby & Juicebox

Speaker

Recommended Sessions

What TA Can Steal from Growth Hackers

Getting Back to Basics to Fight Candidate Fraud

Product

Resources

Company

Developers