Have an account? Log in
How Datawords Group built an AI engineering team without adding permanent headcount
A 950-person content group working for L'Oréal, LVMH and Michelin needed engineers it could switch on when a client project landed and switch off when it ended.
12Interviews their teams ran
4Hires across two group companies
1 in 3Interviewed candidates hiredTL;DR
- Who. Datawords Group, a French multicultural content group founded in 2000 and headquartered just outside Paris. Around 950 people across 25 countries, adapting content for more than 160 cultures for brands including L'Oréal, LVMH, Michelin, Henkel and Carrefour. Its creative and video agency, 87seconds, is one of the companies inside the group.
- What they needed. Senior engineers on four different briefs over four years: a frontend engineer to lead a video adaptation platform, a Python engineer who grew into an AI engineer, an AI project manager, and an audiovisual AI engineer for 87seconds.
- Why it was hard. Datawords staffs against client projects. A permanent hire becomes a liability the moment a client cancels, so it needed senior people on terms it could end. Its internal engineering team is small, its technical bar is high, and it had no appetite for a long interview process.
- What we did. A separate role intake for every brief, signed off with the VP of Engineering. Sourcing in Portugal and across Europe instead of France, starting with our own community. Our own technical screening on every candidate before anyone reached them. A technical challenge built together with Datawords when the AI brief was still forming. Contractor terms throughout.
- What happened. Four hires from twelve interviews across two companies in the group. When the first project was not renewed, we moved our engineer to another client instead of ending his contract. Datawords has come back four times in four years without adding a single permanent engineer.
About Datawords Group
Datawords was founded in 2000 to solve a problem global brands still have: content that works in Paris rarely works untouched in Seoul, São Paulo or Riyadh. The group now runs around 950 people across 25 countries, adapting content for more than 160 cultures for clients including L'Oréal, LVMH, Michelin, Henkel and Carrefour. It does that work on its own platforms, Wezen and Whatsquare, which is why engineering sits at the centre of the business rather than off to one side of it.
Across the group, Datawords has moved further into AI than most of its competitors, using it to produce and adapt content at a speed manual localization cannot match. That runs from text and translation to AI-generated video, synthetic voice and lip-sync, which is where 87seconds and tools like ElevenLabs and HeyGen come in. Those tools only pay off if you have engineers who can actually build with them, and in 2021 those engineers were not easy to find in France.
The brief
Datawords did not give us one role to fill. It gave us four, each one arriving as the business changed underneath it, and each one needing to be staffed on terms the group could unwind if a client walked away. We ran every search from role intake to signed contract, sourced across Portugal and Europe instead of France alone, screened every candidate ourselves, and kept the shortlists small enough that a stretched internal team only met people worth its time. Twelve interviews later they had four engineers and no permanent headcount.
The 5 problems we had to solve
Problem #1: "We cannot put permanent headcount against project work."
Datawords is a services group. Its engineering capacity is funded by client projects, and client projects end, get cut or get postponed. Hiring a permanent engineer against revenue that might disappear in six months is how services businesses end up carrying salaries they cannot justify. But the alternative, going without the engineer, means turning down the work.
It is the classic agency squeeze, and it is the reason Datawords was not simply posting a job ad.
What we did
We placed every engineer on contractor terms Datawords could end when the underlying project ended. No notice periods to negotiate, no redundancy conversations, no permanent contracts on their books across four years of working together.
That structure was tested almost immediately. Our first placement, a frontend engineer, was on a project that was not renewed inside the first year. Datawords ended the engagement exactly as the arrangement allowed. We then moved him to another of our clients, where he is still working today. The group got the flexibility a services business needs, and the engineer did not pay for it with his job.
Problem #2: "The engineers we want are not in France at a price we would pay."
The first brief was a senior frontend engineer to lead the front end of a video adaptation platform, the product clients use to upload video and order subtitling, voice-over and format changes. React and Next on the front, Python and Django behind it, talking over a REST API, inside an innovation team of two or three developers running Scrum on Jira. Datawords wanted someone senior enough to own that front end and shape the architecture around it.
Hiring that profile in Paris meant competing with every other French company for the same short list of people, at Paris rates. Guilherme, who is Portuguese, thought the group could do better elsewhere.
What we did
We sourced in Portugal and across Europe rather than France, which is the argument Guilherme had already won internally with the CTO. Datawords works remotely on European time, so location mattered far less than the group had assumed.
We started with our own community rather than cold outreach. Rui Fonseca, the engineer who took the role, answered a post we put up in the Remote Crew Slack. Every engineer we have placed at Datawords since has been in Portugal or Spain, working remotely, and the group has not once needed to open a French search to fill one of these roles.
Problem #3: "Our teams are too small to interview a lot of people."
Datawords' internal engineering team is small, and 87seconds is smaller still. When a squad is two or three developers, every hour spent in an interview is an hour not spent shipping the client work that pays for the squad. The standard agency answer, forwarding every CV that half-matches the brief and letting the client sort it out, would have cost Datawords more time than it saved them.
Their technical bar is also high, which makes volume actively counterproductive. Sending ten mediocre profiles to a demanding team does not increase the odds of a hire. It just burns the team's patience.
What we did
We interviewed heavily and shared sparingly. Every candidate went through our own technical screening before Datawords saw a name, and we only forwarded people who had already cleared it.
Across all four roles their teams sat twelve interviews and made four hires. On the first search that meant five profiles, four of which Datawords approved, two of which they interviewed, and one they hired. On the two most recent roles, at Datawords and at 87seconds, they interviewed two people each and hired from both. Everybody else was filtered out before reaching their calendar.
Problem #4: "We do not fully know what this AI role is yet."
By the time Datawords needed Python and AI depth, the market was moving faster than anybody's job description. The group knew it needed to build with LLMs. It had not settled on whether that meant a backend engineer who could work with models, a machine learning specialist, or something in between, and the internal answer kept shifting as the group learned what it actually wanted to ship.
Screening against a brief that vague is guesswork. Good candidates get rejected for the wrong reasons and weak ones look fine, which is exactly what started happening: this is the one search where we put a lot of profiles in front of them before it clicked.
What we did
We changed the process mid-search rather than pushing more candidates through a broken one. Together with Datawords we built a technical challenge and made it a screening step.
It did two jobs at once. Candidates were tested on the work instead of on a CV, and Datawords was forced to make the requirement concrete enough to write a challenge for it, which is what finally pinned the role down. We briefed candidates, set deadlines, chased submissions and delivered them to the team, who owned the evaluation. That role became João Martinho, a senior Python developer who now works almost entirely on AI and has been with the group since 2024. The challenge stayed in their process afterwards.
Problem #5: "Our video agency needs a different engineer, sometimes for a week at a time."
87seconds does not need what Datawords needs. It produces video, and its AI work is audiovisual: synthetic voice, lip-sync, generated footage, using tools like ElevenLabs, HeyGen, ArcadsAI and N8N, alongside Node and PHP. That is a genuinely rare skill set, and it sits closer to the marketing team than to a backend squad.
It also does not need that person full time. The work arrives in bursts, a week here, two weeks on a project there, tied to whatever the agency is producing. No permanent role could be justified, and most agencies will not staff at that granularity.
What we did
When 87seconds came to Datawords needing an AI engineer, Guilherme recommended us to the group company directly. We ran the brief the same way we run the others, and placed an audiovisual AI specialist who works with the agency on demand rather than on a fixed contract, coming in for a week or a fortnight when a project calls for it.
87seconds set the location it wanted and we found the person near its Barcelona office. Two candidates, two interviews, one hire.
How we actually ran it
1. One intake per brief, signed off by the manager who owned it.
We never recycled a job description across roles. Each brief got its own intake with Guilherme Gaspar: the product and the daily work, the exact stack, essential versus preferred skills, the seniority band and day rate, the timezone, and the interview loop with a named owner on every step. The frontend role, for example, was scoped as a senior engineer inside the innovation team at 300 to 350 euros per day, with a technical and behavioural interview owned by Guilherme and a final owned by the CTO.
2. Our own technical screening on every candidate.
Not a CV sift and a friendly phone call. A full interview where we described the role honestly, went through the person's experience, asked technical questions and wrote down what came back: years on the stack, availability, notice period, day rate expectation, location and our own read on them. Only people who cleared that reached Datawords.
3. Sourcing in Portugal and Europe, starting with our own community.
For the engineering roles we worked Portugal and the wider European market instead of France, on the basis that Datawords already runs remotely on European time. Rui Fonseca came directly from the Remote Crew Slack. For 87seconds we searched near its Barcelona office because that is where the agency wanted the person. Everyone we placed works remotely from Portugal or Spain.
4. A short shortlist, approved before anyone entered a loop.
The hiring manager gave a yes or a no on each profile before it went further, so nobody entered a technical stage the team had not already agreed was worth its time. Twelve interviews across four searches is the result of that filter, not of a thin pipeline.
5. A technical challenge, built with them and run end to end by us.
For the AI role we built the challenge with Datawords and then managed it completely: briefing candidates, setting deadlines, chasing submissions and delivering them to the team, who owned the evaluation. It screened candidates and sharpened a brief that was still forming, and it stayed in their process after the hire.
6. Contractor terms, and somewhere for our people to go.
Every placement was structured so Datawords could end it with the client project behind it. When that happened on the first search, we moved our engineer to another client rather than ending his work. The flexibility we sold Datawords is only sustainable because we carry the other side of it.
What Datawords ended up with
Four engineers across two companies in the group, covering frontend, Python and AI, AI project management, and audiovisual AI, from a single partner.
Four years of hiring without a single permanent engineering contract on their books.
Twelve interviews for four hires, because their teams only met candidates we had already screened.
Access to the Portuguese and Spanish engineering markets for roles they were struggling to fill in France, at rates that made the argument for themselves.
A first engineer sourced from our own community rather than a cold search, hired after two interviews.
A technical challenge, built with their team, that both screened candidates and forced a vague AI brief to become a real one. They kept using it.
An engineer at 87seconds who works by the week when a project needs him, which no permanent hire could have covered.
A partner they have come back to four times as the business changed, including once by internal recommendation to another company in the group.
