How Flaire Built an Agentic Layer on Top of Spott

Company: Flaire

Who: ~15 people in Paris, hiring across Europe and the US

What: Data science, machine learning, and software engineering search for frontier AI companies. 100% outbound, no job ads, no inbound.

Before Spott: One CRM for sales, one ATS for operations, Notion for documentation, a separate notetaker

Nothing was broken at Flaire. The firm was barely a year old, the recruiters were hitting their numbers, and the stack was doing its job. What nagged at co-founder Adrien Moulias was what came after that: he could see what he wanted to build on top of his data, and he could not build it on what they had.

Adrien had spent two years building and selling an ATS before he started Flaire. So when he went shopping for one, he was not watching demos. He was checking foundations.

"I thought, these guys are really building software for external recruiters, and you can feel it in the details."
Adrien Moulias, Co-founder, Flaire

The problem: the truth was spread across four tools

Flaire is a young firm growing fast, recruiting data scientists, ML engineers, and software engineers for companies building at the frontier of AI. Its recruiters do 100% direct approach: no job postings, no inbound applications, pure headhunting at volume. Between 2,000 and 2,500 candidates contacted every week, 100 to 150 screening interviews, 30 to 40 candidates sent to clients.

That machine worked. The stack underneath it did not scale.

• A CRM for sales, an ATS for operations. Two systems, two records of the same relationship.
• Notion for documentation. Client briefs and market knowledge lived outside both.
• Transcripts on a separate notetaker. The richest data the firm produced, what candidates actually said, sat in a fourth place.
• Sharing candidates depended on memory. A recruiter had to think "this person is brilliant, maybe someone else on the team has a role for them" and say it out loud. Great candidates matched one job, when they could have matched four.

Adrien is precise about the friction: it was not that the team was drowning. It was that the foundations would not support what came next.

"We felt that with our stack, we were not equipped to build the foundations of our future growth."

Because Flaire works for companies building the agentic layer of the web, the team saw where this was going earlier than most agencies. Recruiting is repetitive work over data that needs cross-referencing: CVs and career histories on one side, soft signals from interviews on the other. That is exactly what agents are good at. But agents need clean roots.

"You need good tools at the root, so the data is collected and stored clearly, and then you can extract it easily with agents, manipulate it, and push it back in. That was not possible with our previous CRM. It was not possible with our previous ATS."

Why Spott: the API and the MCP

Flaire talked to several vendors. Functional coverage was not the deciding factor, because no tool ticks every box on the wish list. What separated Spott was what sat underneath the interface.

• A granular API and an MCP server. Not every ATS had an MCP at the time. Flaire already used Claude heavily, so this was decisive. Today most of the firm's recruiter workflows happen in Claude, working on Spott data through the MCP: pull data out, work on it, push it back in. "That fineness of possibility, we did not have it before."
• A database structured rigorously enough to build on. Flaire calls itself a data-driven studio. A well-structured database is not a nice-to-have for them, it is the product substrate.
• AI-native by design, not by retrofit. "What interested us was precisely that the company is young, because it was built after the arrival of LLMs, and by design I think that changes a lot of things in your technical trade-offs." (More on what that architecture looks like: building a modern ATS from scratch.)
• Features that only make sense for agencies. Dashboards where you set targets. Speculative applications, so you can push a CV to a client without a live vacancy. "Those are features you do not find in ATS built for internal hiring. I was very pleasantly surprised, because I thought, okay, we are in the right place."
• References that checked out. Flaire's main concern was reliability. They called Spott customers and investors before signing. It was, in Adrien's words, swept away quickly.

What changed: no more blind spots

Four months in, Flaire's workflows look similar on the surface. The recruiters still hunt. What changed is what happens after hours.

Flaire built an agentic layer on top of Spott. Every night, an agent reads every conversation the team had that day, across every candidate: transcripts, emails, LinkedIn messages, sourcing notes. It cross-references them with every open role in the portfolio and detects matches nobody flagged manually.

Every morning, recruiters get a Slack notification. They open Flaire's internal platform and see the matches waiting for validation, with intro emails already drafted from the firm's own documentation, complete with the context a candidate needs to decide on the company and the role. The recruiter validates or rejects by hand, because a micro-niche demands human judgement. Then the emails go out.

"What has driven our productivity gains, though it is hard to quantify, is that today we have no more blind spots. We cover 100% of the conversations we have with 100% of our candidates and clients, automatically."

The mechanics of that are simple, and they compound. For 100 candidates in conversation, Flaire now surfaces roughly 250 options instead of 100, because no single consultant can hold the whole firm's portfolio in their head.

"With the same number of conversations, I think we can do twice as many placements. Before, we had blind spots. Now we can be sure that for every candidate in active search, we present the maximum number of jobs that fit what they are looking for."

Adrien is careful to call that an intuition rather than a proven number: the layer has been running for six to eight weeks, and the team is still learning to use it. But the direction is clear enough that he expects a 2x to 3x effect on revenue at constant headcount.

Two other things moved:

Advisory work with clients got sharper. Fine-grained data means complete reporting, and the ability to crunch a stalled search and understand faster where the firm can have impact. Flaire builds acquisition strategies with clients and iterates on them, rather than reporting after the fact.

Client calibration happens on day one. On a search for Brighter, a medtech company building AI that detects cardiac conditions in prenatal ultrasound scans, Flaire used the client portal to calibrate sourcing immediately after the brief. Twenty profiles pushed to the client, comments back on who fits and who does not, all before the first real sourcing wave. The search closed with a PhD-level machine learning researcher moving from academic medical research into the team.

The migration: what it actually took

Adrien is straightforward about what moving a core system involves. The CRM sits at the centre of the business, and the team lives in it eight hours a day.

Two concerns going in. First, the migration delta: you back up a database, it gets reinjected into the new tool, and the week or two of work your team did in between needs manual reconciliation. Second, change management: people need to relearn the tool they use all day.

"You can make it as smooth as possible, but it will never be as easy as staying on the old tool. It is a necessary friction to step up."

What made it work: a small team that can be trained individually, a founder who knew exactly how to configure the database and the pipeline stages, and onboarding sessions from the Spott side. "Thanks to Dimitri, who really took the time to train the team." (If you are planning the same move, our recruitment CRM migration guide covers the delta problem in detail.)

There is also a configuration decision hidden inside every migration, and Adrien treats it as an opportunity rather than a chore: how do you classify jobs, how do you classify candidates, what are the funnel stages. Changing tools is the right moment to reinterrogate all of it.

"We are a small team, around fifteen, so it goes fairly fast. Honestly it went well. We had no big scares."

What he brings up unprompted

Ask Adrien what is good about Spott and he does not start with features.

Support that answers. "When we ask a question on support, we get an answer immediately." During the sales process, he asked for a copy button on transcripts so he could drop them into Claude. It shipped the next day. Feature suggestions from Flaire around email enrichment showed up in the product later. There is a public feature portal with voting, and Adrien encourages his team to use it.

Speed of development. "I am very impressed by the speed at which the tool is developing. The foundations are excellent, and it opened up a horizon that is genuinely exciting, by letting us build internal tools on top of this database."

"Honestly, it is a pleasure. We are very, very happy with this migration, and we know it is a partner we will be able to rely on for the future, and for Flaire's next phases of growth."
Adrien Moulias, Co-founder, Flaire

Flaire is a specific kind of customer: technical, opinionated, building their own layer on top. But the lesson generalises. The ATS you choose is not just where your recruiters click. It is the root of every automation, every agent, and every report you will want to build in the next three years. Choose one that can be read from and written to properly, and see what your team builds. See what other firms automate first in our Spott automations guide, or how AI matching reads a database in how AI finds the best candidates.

Want a database your agents can actually read? Book a demo.