Your first Data-to-AI hire: Run a lovable process

Hiring well is hard. You’re trying to find a match for an under-specified role from a pool of unique humans. That said, running a reasonable hiring process shouldn’t be that hard. This is especially true for startups, which are unconstrained by the legacy requirements of a large bureaucracy.

Recently, I gave a webinar on hiring a startup’s first Data-to-AI specialist. As practice, I recorded myself going through the slides.

The first part of the webinar covered what startups can expect from Data / AI / ML roles (posted last week). The second part gets into advice on hiring decisions and running a reasonable process:

The full slides are here (use the left/right and up/down arrows to navigate). The webinar page contains a recording that also includes audience questions.

The rest of this post discusses the key points from the video. Much of this applies to any startup role – especially to hiring the first specialist outside your expertise.

Before hiring

First, you should clarify the role and consider not hiring. Specifically:

• Ensure that the role is needed for near-term growth.
• Start with responsibilities rather than titles, especially given that data titles are often uninformative.
• Make a 90-day plan for the hire, including successful outcomes.
• Consider alternatives to hiring a full-timer: doing urgent work in-house, getting a contractor or fractional (i.e., part-time) person, and deferring non-urgent work.

Deciding to hire

Assuming you’ve decided to hire, you should further clarify the role. This includes:

• Clarify the activities and successful outcomes for the first 30 & 90 days, so the person can hit the ground running.
• Define aspirational goals for the 6-12 month horizon. These are likely to shift in startups, but should help with aligning expectations.
• Decide on the person’s level, which is likely to be mid-senior: a hands-on person who can lead a team and drive tech decisions. If they’re too junior, they will struggle to ship quality work independently – the opposite of what you want from your first Data-to-AI specialist. If they’re too senior, they’re less likely to be hands-on and more likely to be expensive. That said, keep in mind that seniority is correlated with years of experience, but not perfectly – curiosity and a growth mindset are key.
• Set a compensation range. This includes salary and equity, but you can trade monetary compensation for perks like 80% or 90% time. Further, some people may be willing to accept lower compensation for learning and growth opportunities, and for contributing to a mission that aligns with their values. This should all be within reason, though – if you try to offer half the prevailing rates for similar startup roles, you will miss out on quality candidates.
• Give the role a title. As discussed in the previous post, this is likely to be a Data Tech Lead (Analytics Engineer) if the person will be setting up data pipelines and analytics, or an AI/ML Lead (AI/ML Engineer) if they’ll be focused on contributing to the core product of an AI/ML-centric startup.

After getting clearer on the role, you should consider getting help. This includes:

• External recruiters for sourcing candidates and understanding the market. Many recruiters are terrible, but the better ones can expedite the process and help you get better outcomes. Even if you don’t retain an external recruiter for their full set of services, some will offer advice on your job ad, compensation range, and process.
• Internal recruiters for initial screening and running the process. While your startup may not be hiring enough people to justify a full-time internal recruiter, you should consider retaining a temporary recruiter if you’re trying to fill multiple roles (e.g., after a raise). This would help with aligning incentives in comparison to retaining an external agency.
• Data / AI / ML specialists for technical screening. Keep in mind the different areas of Dataland, though – an analyst won’t be of much help in screening AI/ML engineers.

The hiring process

My main message here is to run a process you’d love. Some startup founders were never employees, but you should try to put yourself in the shoes of candidates. Career arcs are long, so the person you’ve put through a poor process may interview you one day – or they may end up working for a potential client. Further, people share experiences on platforms like Glassdoor and Blind, and it’s hard to undo reputational damage there.

With that in mind, my main suggestions for running a process you’d love are:

• Be clear and honest. This includes honesty about the challenges at the company, which would help with retention in comparison to selling a false narrative.
• Be responsive and efficient. As a startup, being quick is a competitive advantage. You also want to get back to building your product as soon as possible. According to a recruitment leader I spoke to, ghosting is the number one complaint they get from candidates. You can stand out by responding quickly. A quick rejection is better than no response.
• Ask relevant, informative questions. Assuming the role and 90-day plan are well-defined, they should be the source of your questions. Also check industry best practices, but…
• Remember you are not Google. Startups should run faster, more efficient processes than big companies. There’s a lot of silliness in big tech hiring, including hackable assessments that rely on memorising a handful of techniques. Personally, I’ve had identical questions asked by Google, Microsoft, Qualcomm, and Intel. Such questions mostly test candidate motivation and their ability to prepare for interviews, which are only partly correlated to on-the-job performance.

Potential process flow

1. Job ad:
   • Make it clear and honest by including key challenges that a successful candidate is likely to face.
   • Include the hiring process, compensation range, perks, 30-90 day plans, and 6-12 month aspirations.
   • Conclude with a simple call to action and custom questions (e.g., how would you deal with the key challenges). Relevant custom questions are better than asking for an open-ended cover letter, especially as many cover letters are written by ChatGPT these days (i.e., they’re useless).
2. Quick rejections: For each applicant who’s neither a spammer nor a bot, reject quickly with a template note if they’re not a good fit.
3. Initial screen: Invite each candidate who is a potential fit to a 30-minute intro call. This is a low-pressure two-way street to understand candidate expectations and see if there’s a high-level alignment. Tips:
   • Try to avoid wasting time on reciting details from the job ad, or on getting candidates to talk about details that can be found in their resume. Instead, you can dig into how their past work and current motivations make them a good fit based on the role’s definition. This may include clarifying and elaborating on the questions you asked as part of the application.
   • Consider asking simple technical questions to filter out people who look good on paper but may be clueless, e.g., any AI/ML person should be able to explain the difference between classification and regression or between supervised and unsupervised learning. Further, explaining technical concepts to non-technical people is a required skill for your first Data-to-AI hire.
4. Technical screen:
   • Personally, I’m a fan of take-home tasks that can be completed in reasonable time. They are superior to tightly-timed assessments, as take-home tasks are closer to real-life work. However, completion in reasonable time is key. For example, you could design a task that can be completed in 1-2 hours, then give candidates 3-4 hours to complete it. If it takes too long, some candidates will drop out of the process.
   • To ensure that candidates understand what they’ve done, schedule a call to dig into their task answers. We live in the age of AI assistants, so I believe candidates should be allowed to use AI as part of a realistic screening process. However, blindly copying chatbot answers is foolish – the live call will help you filter out candidates who did that.
   • As alternatives to a long technical screen, consider paid trials and contract-to-hire arrangements. The best indication of work performance is doing actual work, but whether this is feasible depends on your candidate pool.
5. Final screen and offer: This may include a chat with the founders and reference checks. Keep in mind that reference checks are typically time-consuming and provide little signal – one recruiter I’ve spoken to said they’ve done thousands and only a handful changed the decision. If you’re ready to extend an offer, you may do it orally at the final screen call.

Common pitfalls

Common pitfalls in hiring the first Data-to-AI specialist include:

• Ignoring experts. The main experts I have in mind are recruiters. The best ones have gained a deep expertise through handling thousands of candidates – it’s worth consulting them.
• Trusting the wrong experts. For example, while I can help with navigating Data / AI / ML roles and technical screening, I’m not a recruitment expert. Take anything I say with a grain of salt.
• Google worship. One aspect of this is blindly copying Google (or other BigName) processes. Another is getting dazzled by pedigree – big companies hire many people who end up under-performing. Further, those who have only worked at a big company may not transition well to a startup role.
• Hunting unicorns. As noted in the first part of the talk, unicorns who are good at everything do exist, but they’re hard to hire and retain. Also, there’s a limit to how effective a single person can be when they’re wearing many hats.
• Expecting magic. Dataland has experienced many hype waves – from Big Data through Data Science to Generative AI. Therefore, data experts are often expected to wave their AI wand and deliver magical results. However, getting long-term business value from Data / AI / ML is a non-trivial matter, and recruiting perfect humans is impossible. Aim to align your expectations with what’s feasible for your startup.
• Hiring narrow specialists. People who are too narrow in their skills are rarely a good fit for a startup, especially since the first specialist will need to wear many hats. Exceptions include specialists who have a narrow expertise in the core value proposition of the startup.

Recap: Key takeaways

Overall, the key takeaways from both parts of the webinar are:

• Data & AI are all about plumbing, decisions, and automation.
• Before hiring, get clear on your needs.
• Outcomes first: Consider alternatives to hiring.
• If hiring, run a process you’d love.
• Keep learning, get help, and try to avoid pitfalls.

Feedback is always welcome!