Brendan Sterne

Here’s a memo I sent out to the engineering team this week:

Time for some real talk people.  We have a problem that we’re either not aware of or we’re tolerating.   I had to stop a meeting last month to address it.

A variety of stakeholders were getting a status report from a team, and it went something like this* (* names & details changed to protect the innocent):  “The Montesquieu project is blocked but the Montserrat project is complete.”  And we moved on.

I stopped the meeting and did a quick little quiz to see if people knew what the Montesquieu and Montserrat projects were. Let’s say that there was about 50% correct internal assignment of project name to project understanding for the non-team stakeholders.

As George Bernard Shaw once said, “The single biggest problem in communication is the illusion that it has taken place.”

I’ve seen this before.  People mix up the two project names.  Or they have no idea and don’t ask. Or they are actually thinking of that other project: Montagne!

It’s super fun to name things super fun, but it impedes our mission to help people get jobs.

Collaboration is how we accomplish great things together.  Rapid and clear communication is essential to collaboration.

Here’s why you should name projects as plainly as possible:

• Recognition – when people hear the name, they need to quickly associate it with the project.
• Recall – people need to be able to recall the project name when thinking about the project.

You should apply the ‘unprimed association’ test when you name something XYZ.  The trick is to ask a few people what they think of when they hear of project/product XYZ.  Code names fail this pretty quick. It doesn’t have to be perfect, it just has to be helpful and close.

In the Incubator we established a rule to name projects as plain and clear as possible, because we needed our busy executives (SLT) to understand them and keep them clear in their heads.  That’s why we named them things like: Referrals, Hiring Events, Chat, Guaranteed Response, Text2Apply, Interview Now, etc.

Imagine if they had accidentally mixed them up when conversing about funding (“I think we should fund Montesquieu and not fund Montserrat”).  Yikes.

Here I’m going to quote from the Harvard Business Review: “Use words that will resonate with those whose support and influence you must earn. If they can’t follow your ideas, they won’t adopt them” (hbr.og)

I’m going to give some examples – not to pick on them, but because they merely representative – and suggest alternatives:

• Itasca =>  Resume Bank
• Iris =>  Native App Vision
• Captain Planet => Jobs-on-HP
• Hodor => MatchFilterService
• Beaker => AggFeedStore

The plain alternatives are longer, and in practise people might adopt acronyms or shortened versions.  That is ok. I will assert (without evidence) that acronyms are better at recognition and recall than code-names – when people use acronymed names (like Amazon SES), internally the acronym and the expanded-form (Simple Email Service) are connected in the brain.  But my brain simply doesn’t have any neurons connecting the name Itasca with the concept of a resume bank.

Harder to recall/disambiguate (all real project/product/tool names at Indeed): Wendy, Wenda, Waldo, Waldorf, Basestar, Deathstar, Stargate, Goldengate, Gate, Iris, Aurora, Orion, Janus, Philae, Curious George, Yellow Hat, Yamanote, Peregrine, Naboo, Nigma, Hodor, Horton, etc.

Easier to recall/disambiguate: Session Viewer, Query Manager, ITA Manager, Job Metadata Viewer, AdCentral, AggCentral, User Personalization Service, Notification Preferences Service, TitleNorm, Labeler, JobSearchService, TestStats, Salary Transparency, etc.

I feel like a grouchy old person as I write this.  I’m fun. Really, I am. But please stop having so much fun with your names.  This is not a mandate. This is just Brendan sharing his belief that we should use boring, simple, clear, recallable and recognizable names for our projects, services and tools.

Operating Mechanism is a fancy term for processes (and the corresponding tools) used to run an organization.  Examples of operating mechanisms are:

• The quarterly talent review process
• The developer interview process
• Weekly product review
• The growth-board / project-funding process

These are necessary to help the business do its work effectively.  Each operating mechanism has its own purpose (which should be clear to those involved – but sometimes isn’t).

After participating in running, changing and building a variety of operating mechanisms, I wanted to share my observations.  These are some inconvenient truths according to me:

1. Non-compliance is real

People ignore emails and instructions.  Things that are too complicated don’t get done.  People go on vacation. Higher priority items crowd out lower priority items.

I’ve seen people design operating mechanisms without understanding the need to be simple / fast / valuable.  People don’t comply well with mandates – especially if they don’t understand the why of it.

Also, people are forgetful.  If you put an operating mechanism in place that requires people to remember something (e.g. “please remember to tag these 3 random people in every ticket that affects some subgroup”) it will not happen.

And so you need to think about how the operating mechanism handles non-compliance.  Does the whole thing fall apart if someone doesn’t do their part?  Or is it more resilient – does it simply degrade? Interview accuracy degrades if you have to cancel an interview.  Approval chains grind to a halt if there isn’t a way to delegate or escalate.

2. Operating mechanisms operate within constraints

Almost every step in an operating mechanism operates under a time or people constraint.  We are always doing the best we can within the limits. It could always be better if we gave it more.  E.g. more peer-feedback during reviews, more time during calibrations, another phone screen before an onsite, more time to discuss in loupe, etc.  

Building an operating mechanism means using judgement about the diminishing returns from doing more.

3. You can’t A/B test an operating mechanism

When discussing an operating mechanism, someone at Indeed will inevitably propose that we A/B test it.  Usually accompanied by a wink 😉

You can’t A/B test an operating mechanism, because you can’t do random assignment and/or get enough power and/or align on metrics.  The closest you can do is have different groups pilot or try different approaches, then compare. Most often you evaluate and improve operating mechanisms by cycles of feedback, assessment/judgement, modification.  There are business studies that you can read about, but it can be hard to know if they are really valid, and if they apply to your particular circumstance. But studies can often inform various adjustments to try.

4. There will always be problems and complaints

“It has been said that democracy is the worst form of Government except for all those other forms that have been tried from time to time” — Churchill

Operating mechanisms don’t get much love.  The best they can hope for is some respect.  But there is no such thing as perfection – where nobody complains.

To keep things efficient and fruitful, if you are discussing operating mechanisms, keep complaining to a minimum and stick to discussing concrete proposals for change.  Keep it real. Abstract discussions are rarely helpful.

5. Organizations outgrow their current operating mechanisms regularly

What got you here won’t get you there.  Things change. 

There will always be a sense that things are broken.  This is no excuse for not fixing things. It’s an explanation for those who are surprised that at fast growing companies like Indeed the operating mechanisms aren’t running perfectly smooth.  That’s because circumstances have and continue to change rapidly. It can be hard to get ahead of exponential growth.

So what should you do about these?

Here are tips for working with operating mechanisms:

• Start with why.  Clearly and repeatedly articulate the purpose of the operating mechanism, whenever you first create it, and when people are proposing changes.

• Remind everyone of the context.  Operating mechanisms are temporary, imperfect structures designed to accomplish goals in a changing environment.  Feedback, discussion and iteration/change are necessary part of the process. Remind everyone regularly.

• Organize around written proposals.  Discussions around operating mechanisms can get abstract, unclear and wasteful.  Try as best as possible to stick to gathering feedback, then discussing written proposals.  To respect everyone’s time, focus on written proposals for change – there’s no point discussing a criticism if it isn’t accompanied with some suggested change.

The odds of developing a second successful product – that is commensurate with a first successful product – are low.

As someone who works on the product exploration side (a.k.a. launching new products) vs the exploitation side (a.k.a. optimizing and improving existing products) I am continually on the lookout for companies that are able to pull off this challenging feat.

It was heartening to hear other people acknowledge how rare this is.  On the Acquired.fm podcast covering PowerPoint they touched upon this:

Ben: I pitched you on two podcast ideas. One was, we should cover acquisitions that actually went well, because the trope is that none of them ever go well and they implode. The other one was, why don’t we do a podcast series where every episode is about the rare company that manages to have a billion-dollar-plus innovation twice.

David: That’s right. I remember that. And we decided not to do that because there just weren’t that many. It would be a very, very short podcast series.

Ben: Yeah. And I think our thesis was basically like, companies have the magic moment just once, because they’re so rare, that they find that product-market fit, the monetization, like just everything works. And then after that, companies keep finding ways to grow and expand that thing, but they generally fail to find a second complete one of those and particularly one that works under the same roof and doesn’t interfere with the other main business. Microsoft did it. They’re one of the few in the world.

I always thought gmail was an example of an employee coming up with an idea, and having automatic-permission to explore it due to Google’s 20% time.  It is constantly referenced in articles about the policy.

But there is some nuance here.  It turns out Larry Page asked Paul Buchheit to explore email:

Larry Page assigned the project to me. He said, “Build some kind of email something” and chose me because I had an interest in email. I tried to create a web-based email service in 1996, shortly before Hotmail, but didn’t complete it.

Interesting.  This takes away one powerful piece of evidence often used to justify policies that enable employees to pursue their own innovative ideas on company time.  In this case there was an executive sponsor right from the beginning.

[update 10/19/2018]  Paul Buccheit addressed this again:

As Google grew, did you keep the freedom to hack on things you found interesting? Was creating Gmail part of that?

Around 2001, Larry got frustrated that each group was setting their own priorities and not working on what he thought were the most strategically important things for the company. His fix was to eliminate management and organize engineering around specific projects. He and Wayne Rosing, who was the VP of Engineering at the time, would sit down with engineers and give them projects. When they sat down with me they said, “we want you to build an email something.” That was all the specification I got! So I went off to build something with email, which became Gmail.

At Indeed we’ve been working on continually improving our processes for funding, staffing, executing and assessing new experimental products.

One of our challenges is assessing early products and figuring out whether to kill or continue them.  One approach used in the past has been ‘make-or-break’ goals.  This has typically been represented as a top line goal to reach at a point in time, such as: X hires, Y sales, etc.

Over time we have learned that these make-or-break goals do not provide enough flexibility for the product team and also do not provide enough flexibility for Indeed.  In certain circumstances the team might make the goal – but Indeed leadership is simply convinced that Indeed’s limited resources to test out new products should be spent elsewhere.  Conversely the team might not make the goal – but leadership is convinced that this was the wrong goal, and that further investment is warranted.  So we are no longer doing make-or-break goals.

What we are trying instead, is to use a metered-funding approach, where an investment committee provides funding in stages – and the project teams have autonomy to use their funding and pursue their vision, but when the money is running out they need to return to the investment committee for more funding.

So how does the investment committee decide if projects get further funding?  Here’s what I’ve been coaching Incubator teams with:

“For projects to continue they need to provide compelling evidence that is surprisingly positive that shows their product can produce sustainable value in a meaningful market.”

Let’s pick this apart:

• Compelling Evidence – Bring verifiable data and evidence.  No evidence-based-learning means no more funding.
• Surprisingly Positive – Sets the bar high, if there aren’t aspects of the experimental product that are working surprisingly well we should probably put our investments elsewhere.  Most projects will not find something surprisingly positive.
• Sustainable Value – There is confidence that the product can achieve sustainability.  We don’t want to launch new products that are simply re-packaging our sponsored job advertising business into a different model and selling it for a loss.  Teams need to understand the cost of acquiring users, the cost of servicing them, and how these are trending.
• Meaningful Market – We believe this market could be sizeable enough or strategic enough to be important to Indeed.

There is room for interpretation with these and over time we will provide more clarity and guidance, with examples.

By continually experimenting with a portfolio of new products, Indeed should be able to compare projects side-by-side, and invest more in the ones that better meet this criteria (and kill projects that fail to meet this criteria).

Indeed is investing more and more in discovering and building new products.  As the product team discussed and evaluated our successes and failures – I found it difficult to communicate without some common terminology around the different types of innovation projects underway.

So here are a couple of terms I proposed that were intended to help the discussion.  In all cases I’m talking about trying to find new successful products – new sources of revenue that are pre product-market fit.  Initiatives to grow core business are outside of this discussion.

Seeds: Small projects; less than one quarter long; pursuing a solution to an opportunity; developing and piloting an MVP solution; gather some initial benchmark performance data.

Pods: Medium size projects; 3-4 quarters long; pursuing solutions to an opportunity; enough time to gather benchmark KPIs, to run some tests to see how the KPIs improve, and to try a few different approaches to the product.

Big Bets: Large projects; 6+ quarters long; strong commitment to a new business opportunity; try to launch and scale it quickly.

Projects of any size that have found product-market fit are no longer seeds, pods or big bets.  They’re new products and they need to be scaled as large and as efficiently as possible.  Also, these are not strictly stages.  These are different approaches to product discovery and development.

What are they good for?

Seeds

Seeds are useful because they bring an idea to life.  It’s fine to research an idea – like creating a site dedicated to jobseeker events – but the requirement to actually pilot something adds a level of fidelity to it.  It’s way easier to engage with something tangible – to both poke holes in the idea, and to see the full potential.  Show, don’t tell.  The nice thing about seeds is that they don’t need Senior Leadership support.  You can – and should – support seed projects that few people would bet on.

Seeds also support the creation of pods and/or big bets.  Both pods and big bets are a substantial investment of limited resources (product, engineering, sales).  Starting with a seed is a way to get some clarity and hopefully some real data before committing to a pod or big bet.

You should expect to stop after the seed phase and then make a decision about what the next step is.  That might mean putting the seed idea on the shelf.  It might mean gathering the resources to create a pod.  But it’s a mistake to continue at the seed level too long.  They need more resources (engineering, marketing) to have much of a chance of real success.

Pods

Pods are like a small funded startup with a year of runway.  They have a problem identified, they are building and launching a product.  Their goal is to find product-market fit, or to show a clear trajectory on the KPIs towards product market fit – before the $ runs out.  Lean startup approaches to product discovery and validation apply here.

For pods to be successful they require a certain amount of investment.  They require 2-3 engineers.  They require product marketing support – with a budget.  They might require some integration with other products.  And they require some Senior Leadership support to help secure resources.

Big Bets

If you think you’re onto something, if you’re committed to it – then go big.  Pros: bigger product faster.  Cons: what if you don’t have product-market fit?  Then you’re burning lots of resources as you try to find it.

Big bets require an engaged executive sponsor because of the significant investment involved.

Ok, I understand the proposed terminology, so what?

Companies should be pursuing a mix of these concurrently.  They all have a role to play.   You should make it easy for entrepreneurs in the org to pursue a seed project they’re excited about.  This could be an innovation rotation, a labs/skunkworks project, or the permission to launch a nights-and-weekends project.

I would propose that pods are one of the most efficient ways to find product-market fit; enough resources to actually try something out and learn, but with limited cost to the company.

Finally – You shouldn’t be afraid to place a limited number big bets on the most promising opportunities in your space.  Bigger risks but bigger rewards.

Haters gonna hate.  Engineers gonna build.  Unsurprisingly, engineers are drawn to the challenges that interest them: reliability, scalability, and reduced technical risk.  In a product space where Market Risk dominates (see Part I), these are the wrong priorities.

A lot of startup advice makes sense when you realize it’s designed to address market risk not technical risk.  YCombinator basically exists to beat entrepreneurs over the head with this lesson – same thing with the Lean Startup methodology.

• Get out of the building and spend time with potential customers.  In other words, stop coding (addressing technical risk) and start validating your market hypothesis.
• Do things that don’t scale.  Because who cares how scalable it is if there’s no market?  And by the way, there’s probably no market for your Airbnb-for-garden-tools startup.
• Build, Measure, Learn.  If you do have to build something, build just enough to start testing whether there’s a market for your product.

Even if you have a large amount of technical risk that needs to be addressed, don’t forget the market risk.

Consider the agile software development movement.  It began more than 20 years ago in the contract software development community, and was designed to address the high percentage of failed projects.  Since a lot of contract software development is for internal systems, there was an assumption of product-market fit embodied in the requirements.  However, all too often the resulting product failed to properly address the needs of the contracting organization.  It wasn’t that there was some technical invention missing – but rather that existing technology was shaped into a solution that failed to solve the problem.  The first principle behind the Agile Manifesto is:

Our highest priority is to satisfy the customer through early and continuous delivery of valuable software.

This addresses technical risk – ensuring subsystems work together as early as possible.  It also addresses market risk by providing regular checkpoints to validate product-market fit (even when there is someone writing a check who says they know what their organization needs).

But be careful!  Agile approaches don’t necessarily address product-market fit.  In a product company, building something in sprints doesn’t protect you from building the wrong thing in the right way.  If you aren’t validating the market as you go along (or better yet before), you’re just addressing technical risk (which is good, but not sufficient).

I was delighted the other day by an engineering manager who really demonstrated an understanding of technical risk vs market risk.  When discussing whether to put together an engineering team to start building a product – or use a more lightweight rapid-prototyping and piloting approach (that may need to be replaced if the product succeeds), the engineer manager said:

“I know we can build something that scales, what I don’t know is if the product will succeed”

That pretty much sums up most startups these days.  So spend more time finding out if the product will succeed and trust in your ability to scale.

If someone could build a teleportation device that worked, and that cost less than $2,000 to transport a person across the planet, they would be rich.

If someone were to build a bluetooth-controlled LED display that could be affixed to the back of a car and allow the driver to display messages from their phone, they might be rich.

What’s the difference between these two ideas?  In the first case, we know there’s a market for rapid transportation of people.  There’s evidence all around (cars, trains, airplanes).  What’s lacking is the technology – it’s not clear if the technology can be made to work.  In the second case we don’t know if there’s a market for communicating with the people behind you using your phone.  The gap is the market (or as some say, product-market fit) – it’s not clear if customers will adopt the product.

If you were thinking of investing in companies pursuing these two ideas, and you were talking about the risks, you would say the first has technical risk, and the second has market risk.

To figure out which kind of risk your product is facing, ask yourself: If the technology was available to buy off-the-shelf, cheap, and any startup could build it, would it be obviously successful?

• High resolution mobile screen, with full web browsing capability (aka iPhone)?  Yes: tech risk.
• Social network to share filtered photos (aka Instagram)?  No: market risk
• Container that keeps vegetables fresh without losing any nutrients for a month? Yes: tech risk.
• Macaroni & Cheese Flavored Ice Cream?  No: market risk

Interestingly, the rapid pace of technology has meant that more technology is available off-the-shelf (hardware and software), cheap and any startup can build it.  A lot of the lower hanging fruit of products with technical risk have been created; and new opportunities are now in reach. These new opportunities are for things that aren’t just better, faster, cheaper.  These are new opportunities to address unrealized needs; things you didn’t realize you would want or love.  I would put things like Twitter, Instagram, AirBnB, Pinterest and Snapchat in this category.  It wasn’t obvious that there was a market for these products.  And while they definitely had technical risk too, the bigger risk was market risk.

In the real world new products face a combination of technical risk and market risk.  But usually one dominates.  And I believe that we have moved into a world with more market risk than technical risk.  Take a look at the Unicorn List and ask yourself how many of these represent new technology that was risky, vs the application of technology into a new area with an uncertain market.

In part II I will show how this shift from technical risk to market risk explains the rise of Agile development and the Lean Startup movement.

I’m inclined to agree that the recent New York Times article on Amazon “Inside Amazon: Wrestling Big Ideas in a Bruising Workplace” appears pretty slanted and seems written to match a pre-conceived narrative.

I’ve never worked at Amazon, but I have former colleagues that I respect who work there.  A lot of the practices described in the article are much more benign in practice (e.g. systems for peer-feedback, meetings to calibrate the evaluation of employees, etc) and are pretty common among tech companies.

For those interested in these types of things, there’s a good counterpoint to the article written by Nick Ciubotariu, a current engineering manager at Amazon, here: An Amazonian’s response to “Inside Amazon: Wrestling Big Ideas in a Bruising Workplace”

However, Nick is writing from the perspective of a recent (18 mo tenure) engineering manager who is presumably very good at his job and well-suited to thrive.  Average performers, or folks in other departments such as warehousing, sales, marketing might have a very different experience.  Also, on occasion Nick has chosen his words very carefully, for example:

During my 18 months at Amazon, I’ve never worked a single weekend when I didn’t want to. No one tells me to work nights. No one makes me answer emails at night. No one texts me to ask me why emails aren’t answered. I don’t have these expectations of the managers that work for me, and if they were to do this to their Engineers, I would rectify that myself, immediately.

Notice that he doesn’t address whether he actually works weekends and nights.  Maybe he’s just a happy Amabot.  🙂

My final thought it that even if Amazon were as the NYTimes article described, it would be a legit approach to business:

“Amazon is O.K. with moving through a lot of people to identify and retain superstars,” said Vijay Ravindran, who worked at the retailer for seven years, the last two as the manager overseeing the checkout technology. “They keep the stars by offering a combination of incredible opportunities and incredible compensation. It’s like panning for gold.”

For more perspective, read the comments (on LinkedIn) on Nick’s rebuttal, or the NYTimes Article comments.