Don't go to market for AI leaders with champagne tastes on a beer budget
Using a decent position description, and your common sense, to hire the best data and AI leader you can actually afford
Apologies in advance for a minor rant but there are a lot of woeful position descriptions out there. We all believe that people are the most critical component of a good software / data / AI team right? They are certainly the most expensive component (unless you’re training your own foundation model of course).
And many believe that assembling a team of people with diverse backgrounds is important too. So simply relying on your personal network or your alma mater, while a popular strategy, is one rife with mid to long term dangers.
So how do you raise the bar on position descriptions and why should you care?
Create clarity
The first and most important reason to be disciplined, sit down, write some words, come back to it later, write some more and then share what you’ve written with a colleague and discuss, is that it clarifies your thoughts and lets you sharpen up your understanding of the true ask. In my experience, most people resist writing a position description, or pass it off to their HR business partner, because they’re overwhelmed by the idea of being specific about what they need.
Something else on the dopamine hit inducing ‘to do’ list always seems easier. I get that. Bringing on someone new is a massive investment of time and the future need is murky at best. Triple the strength of the aversion field if you’re hiring for a role you’re not terribly familiar with.
But you’ll improve your odds of a great hire at least from ‘dismal’ to ‘fair’ by writing down and debating what can otherwise quickly become a long list of carelessly assembled ‘must haves’. To be concrete, you can:
Clarify your future tech stack. While PD writing 101 tells you not to put specific languages in your ‘must haves’ (great folk, the kind you should be shooting for, can pick up another language), having a discussion about which languages belong in your ‘nice to have’ list can be illuminating. If there’s a need to future proof / migrate / disestablish in your near future, it might shake out through conversation about why Tensorflow is out but Spark makes the cut.
Clarify your org structure. Will this new team member need to lead people? How many? How many in six months?
Expose weaknesses in your current skill mix that need attention. How critical will written communication skills be in this new human? How about stakeholder engagement? Verbal comms? Sprint planning?
It can also help you spot the ‘laundry list’ PDs which generate those embarrassing job ads that ask for ‘five years experience in X’ where X has only been in commercial use for two years. Tanks your credibility with all the savvy future colleauges before you even get to know they exist.
And yes, I hear all the folk who are saying this problem doesn’t only exist when hiring for data / ML / AI roles. That’s true, but it’s significantly more pronounced in the ‘data people’ universe, particularly when deep data skills don’t exist in the founding team.
Leave a record
The second reason to invest time in a decent position description is that you’ll leave a written record of what you agreed was needed. Useful in so many ways:
Checking your own expectations a few months in by refreshing your memory of what was agreed
Giving your new hire a useful document to refer back to when planning their career growth
Laying the foundations for a career ladder with a clear progression journey from junior through senior roles
And in worst case scenarios, acting as a solid foundation for performance management conversations and redundancies
(Optional extra rant: A position description is not a job advert)
Having spent three happy years working in a job search platform company, I can’t resist this slight off tangent mini rant. Once you have a great position description, your work isn’t done. Your job ad needs to pop and speak directly to the top drawer candidates. Craft the ad from your PD of course but don’t just copy-paste! A good PD for a senior role might be 3-4 pages long. A job ad has to grab attention so it needs to be concise, hook the reader (your amazing new colleague!) and leave them wanting more.
Some of the important dimensions of the job of leading data & AI
I’ll leave the nuanced conversation about all the flavours of technical jobs (people who build warehouses, people who build pipelines, people who build ML models, people who build reporting, people who build product analytics, people who build AI products, etc, etc, etc) for a future post.
From here, we’ll focus on the dimensions you need to consider when shaping up a data & AI leadership role in a modern SaaS company where there is little or no physical product - which is every marketplace, every ‘platform company’ and every fintech just for starters. Think of the list that follows as a framework to work from when drafting, re drafting and debating that position description that I know you’re now super motivated to write 😝
What I want to open your eyes to is the most common mistake I see companies of all sizes make - going to market with a budget of $Y and a wishlist of experience and expertise that commands a 2x or even 3x $Y salary in the open market. Don’t be that company 🙏 🙏 🙏
If you do that, the ‘best’ case scenario is that you hire someone who just really needs a job and leaves you as soon as an opportunity more commensurate with their market value comes along. Much more often, you’ll hire someone with a random subset of the skillset you went to market for. Trust me, that is never going to be better than realistically constraining your expectations to meet your budget and thoughtfully selecting the most important subset that you can get, while ‘eyes open’ noting what you’re choosing to forgo.
OK, the dimensions I’d start with:
Commercial data strategy - e.g. when we sign contracts with other SaaS providers, who ‘owns’ the data and what restrictions are placed on usage of the exchanged data? All of these contracts have implications on your technical solution and your operating processes. With the enormous surge in AI regulation and the flux around the ownership of data underpinning foundation models (such as GPT-4) this isn’t a space to take lightly.
Responsible data use - e.g. creating and ‘policing’ the company wide approach to capturing, storing and processing data. In a regulated or semi-regulated industry this in a huge and messy area and extremely hard to do well at scale.
Architecture - in many SaaS companies, it’s pretty much impossible to cleanly separate ‘data’ architecture from product architecture. There will be interconnections and implications both ways, and that’s before you even begin to consider the (often significant) implications of your proposed data/AI usage on your security and network architecture.
AI product development - the blind leading the blind is both dangerous and costly here. A leadership team without a decent level of algorithmic literacy are likely to ask for the impossible, pick the wrong partners, run up enormous bills, create unmaintainable products and be unable to attract top notch practitioners. And no, this problem didn’t go away with the advent of an API you can call to unleash the magical LLM 🪄
Data warehousing - yes I know the cool kids think they don’t need a warehouse but trust me, you do. What was old is new again, the technology solutions have changed, the vendors have had a makeover, and a lot of ideas from 25-40 years ago have new names. But the basic problem of having a consistent vocabulary with which to quantitatively explore your business has not gone away.
Business intelligence - often conflated with warehousing but another nuanced challenge where it is super easy to over invest and under deliver. And you really want your revenue reporting to be robust and accurate 😬
Marketing analytics - attribution, robust CAC numbers, channel strategies driven by data, consent based customer and prospect contact. At scale, all of this rests on solid data foundations so invest wisely.
Instrumentation & product analytics - so many young companies (and some not so young) have no idea how users really use their product and hence are often making new product bets based on gut feel and focus groups for far too long. Knowing what to instrument, how to do so coherently across channel, which clicks to capture and what to pick up from the backend isn’t rocket science but it is complicated. Like so many things, easier when you’ve done it at least once before.
Experimentation - this one strikes me as the secret sauce that separates the sophisticated players from the rest. And much more subtle than turning on the A/B testing option in your feature flagging tool.
Clearly with this breadth of responsibility, you would also need someone comfortable with formulating and discussing strategy to board level, handling not small budgets, planning and executing significant programs of work, negotiating seven figure software contracts and attracting & retaining senior folk with many other enjoyable and well remunerated career options.
As you also can’t possibly do all of the above well and quickly, you need someone with the confidence to hold the line on scope and the experience to plot the best sequencing possible for your market and your situation.
So, does your hiring budget meet your ambition? If not, begin your painful ranking exercise!
And don’t unsubscribe from this newsletter in a huff! I promise you, that being realistic about obtainable breadth of experience from the outset, or raising your budget if you can’t strategically afford to under hire, is a much better option than setting yourself and your new colleague up for a slow and painful dissent into intense frustration, ‘under performance’ (relative of course to an unrealistic expectation) and foregone market share from the slow start and the data debt accumulation.
Wishing you all a peaceful weekend. In Melbourne of course we are gearing up for the race that stops the nation. I kid you not. We have a public holiday for a ball sport too so we’re a very equal opportunity sports mad state.