Beware the data science pin factory: The power of the full-stack data science generalist and the perils of division of labor through function

Schedules of functional specialists are difficult to align as each specialist is allocated to several initiatives.

A one-hour meeting to discuss changes may take weeks to line up.

And, once aligned on the changes, the actual work itself also needs to be scheduled in the context of multiple other projects vying for specialists’ time.

Work like code changes or research that requires just a few hours or days to complete still may sit undone much longer before the resources are available.

Until then, iteration and learning languish.


It narrows context.

Division of labor can artificially limit learning by rewarding people for staying in their lane.

For example, the research scientist who is relegated to stay within her function will focus her energy towards experimenting with different types algorithms: gradient boosting, neural nets, random forest, and so on.

To be sure, good algorithm choices could lead to incremental improvements.

But there is usually far more to gain from other activities like integrating new data sources.

Similarly, she may develop a model that exhausts every bit of explanatory power inherent to the data.

Yet, her biggest opportunity may lie in changing the objective function or relaxing certain constraints.

This is hard to see or do when her job function is limited.

Since the research scientist is specialized in optimizing algorithms, she’s far less likely to pursue anything else, even when it carries outsized benefits.

Telling symptoms can surface when data science teams are run like pin factories, for example in simple status updates: “waiting on ETL changes” and “waiting on ML Eng resources” are common blockers.

However, I believe the more insidious impact lies in what you don’t hear, because you can’t lament what you haven’t yet learned.

Perfect execution on requirements and complacency brought on by achieving process efficiencies can mask the difficult truth, that the organization is blissfully unaware on the valuable learnings they are missing out on.

The solution to this problem is, of course, to get rid of the pin factory.

In order to encourage learning and iteration, data science roles need to be made more general, with broad responsibilities agnostic to technical function.

That is, organize the data scientists such that they are optimized to learn.

This means hiring “full stack data scientists” — generalists — that can perform diverse functions: from conception to modeling to implementation to measurement.

With fewer[3] people to keep in the loop, coordination costs plummet.

The generalist moves fluidly between functions, extending the data pipeline to add more data, trying new features in the model, deploying new versions to production for causal measurement, and repeating the steps as quickly as new ideas come to her.

Of course, the generalist performs the different functions sequentially rather than in parallel — she is just one person after all.

However, doing the work typically takes just a fraction of the wait-time it would take for another specialist resource to come available.

So, iteration time goes down.

Our generalist may not be as adept as a specialist in any one function.

But we are not seeking functional excellence or small incremental improvements.

Rather, we seek to learn and discover all-new business capabilities with step-change impact.

With full context for the holistic solution she sees opportunities that a narrow specialist won’t.

She has more ideas and tries more things.

She fails more, too.

However, the cost of failure is low and the benefits of learning are high.

This asymmetry favors rapid iteration and rewards learning.

It is important to note that this amount of autonomy and diversity in skill granted to the full-stack data scientists depends greatly on the assumption of a solid data platform on which to work.

A well constructed data platform abstracts the data scientists from the complexities of containerization, distributed processing, automatic failover, and other advanced computer science concepts.

In addition to abstraction, a robust data platform can provide seamless hooks into an experimentation infrastructure, automate monitoring and alerting, provide auto-scaling, and enable visualization of debugging output and algorithmic results.

These components are designed and built by data platform engineers, but to be clear, there is not a hand-off from the data scientist to a data platform team.

It’s the data scientist that is responsible for all the code that is deployed to run on top of the platform.

And, for the love of everything sacred and holy in the profession, don’t hand-off ETL for engineers to write.

I too was once lured to a function-based division of labor by the attraction of process efficiencies.

But, through trial and error (is there no better way to learn?) I’ve found that more generalized roles better facilitate learning and innovating[4] and provide the right kinds of scaling: to discover and build many more business capabilities than a specialist approach.

And, while there are some important considerations[5] that may make this approach to organization more or less tenable in some companies (see footnote), I believe the full stack data scientist model provides a better starting place.

Start with them, and then consciously (grudgingly) move toward a function-based division of labor only when clearly necessary.

Final thought.

There is further downside to functional specialization.

It can lead to loss of accountability and passion from the workers.

Smith himself criticizes the division of labor, suggesting that it leads to the dulling of talent — that workers become ignorant and insular as their roles are confined to a few repetitive task[6].

While specialization may provide process efficiencies it is less likely to inspire workers.

By contrast, generalist roles provide all the things that drive job satisfaction: autonomy, mastery, and purpose[7].

Autonomy in that they are not dependent on someone else for success.

Mastery in that they know the business capability from end-to-end.

And, purpose in that they have a direct connection to the impact on the business they’re making.

If we succeed in getting people to be passionate about their work and making a big impact on the company, then the rest falls into place naturally.

[1] I took the liberty of modernizing Smith’s use of pronouns.

[2] As J.

Richard Hackman taught us, the number of relationships (r) grows as a function number of members (n) per this equation: r = (n²-n) / 2.

And, each relationship bares some amount of coordination costs.

See: Hackman, J.


Leading teams: setting the stage for great performances.

Boston, Mass.

: Harvard Business School Press, 2002.


[3] It’s important to note that I am not suggesting that hiring full-stack data scientists results in fewer people overall.

Rather, I am merely suggesting that when organized differently, their incentives are better aligned with learning vs.

efficiency gains.

Consider the following contrasting deptarment/team structures, each with 3 people.

Fractional estimates and summed team sizes are illustrative only.

Specialist Model: organized for functional efficiency.

Workers are not dedicated to any one business capability, rather their time is allocated to many.

Generalists Model: Full-stack Data Scientists optimized for learning.

Workers are fully dedicated to a business capability and perform all the functions.

[4] A more efficient way to learn about this approach to organization vs the trial and error I went through is to read the book by Amy C.

Edmondson called “Teaming: How Organizations Learn, Innovate, and Compete in the Knowledge Economy” (Jossey-Bass, 2014).

[5] This process of iteration assumes low cost of trial and error.

If the cost of error is high you may want to rethink (i.


, it is not advised for medical or manufacturing).

In addition, data volume and system availability requirements should also be considered.

If you are dealing with petabytes or exabytes of data, specialization in data engineering may be warranted.

Similarly, system availability (ie.

uptime) and innovation are tradeoffs.

If availability is paramount, functional excellence may trump learning.

Finally, the full-stack data science model relies on the assumption of great people.

They are not unicorns; they can be found as well as made.

But they are in high demand and it will require certain conditions in order to attract and retain them (competitive compensation, company values, interesting work, etc.


Be sure your company culture can support this.

[6] Smith, Adam.

An inquiry into the nature and causes of the wealth of nations.

Dublin: Printed for Messrs.

Whitestone, 1776.


Page 464.

[7] Pink, Daniel H.

Drive: the surprising truth about what motivates us.

New York, NY: Riverhead Books, 2009.

Originally published at multithreaded.


com on March 11, 2019.

.. More details

Leave a Reply