Year Retrospective @ SkuVault

Last year guys from SkuVault offered me an amazing opportunity to help the company manage a growing development team, create organized schedule, establish workflow that reflects the company goals .

For those who don’t know – SkuVault is a Warehouse Management System (WMS). Like a swiss army knife, SkuVault manages and syncs your inventory across e-Commerce platforms, POS, Logistics and Warehouses, providing accurate quantities in order to prevent out of stocks. Headquartered in Louisville, KY – SkuVault helps to manage the inventory for hundreds of clients all across the globe.

2540ridgemar

It’s time to list some of the achievements we accomplished during this year.

Observational Research and Optimization Scenarios

For the first 2 weeks, I was examining the flow within the project, and getting to know the team. Each couple of days I published blog posts on my findings with ideas on how to improve and optimize the workflow. Some things in my new team were completely different from my previous experience:

  • No teamleads. That meant that developers split up into Code Review teams, and reviewed each other.
  • Technology stack (.NET at SkuVault vs Scala / Riak / react.js at Storia). With all the pros and cons, .NET development teams don’t have that clear BE / FE differentiation: backend developers can work on frontend tasks via ASP.NET MVC, so our devs are more like (..universal soldiers).
  • No UX / UI design step in the workflow. This particular part makes every decision much faster. The product itself (SkuVault Warehouse Management App) uses Bootstrap, and is very utilitarian from design perspective. Key factor here is ease of use (as much as it can relate to industrial application).
  • Distributed team on both sides of the Atlantic, covering almost 24 hour period.
  • SkuVault is used by hundreds of customers around the world, bugfixing happens daily, and there are different bug priorities. This particular moment doesn’t work well with typical “sprint->release” cycles (because the priorities may change quite fast, or something needs to be urgently released).

With all those differences in mind, I started to streamline the workflow in JIRA.

Statuses, Transitions, Workflow

I managed to decrease the number of statuses.

  • Used to be: 23 (with any transitions allowed between any statuses)

shawshank_redemption

  • Became: 9 (with clear status sequence that reflects state of a bug / new feature).

SkuVault v1.7 - JIRA 2016-06-08 15-12-47

Most statuses were redundant, I’ve changed some of them with combination of “labels + status”, some were eliminated and substituted by generalized statuses (for example statuses “Design Holds”, “Client Clarification” changed to status: Hold + label “IncompleteDescription”).

The Workflow is constantly being refactored and improved per developers’ suggestions and whole team feedback. Last week I’ve released the 7th version of the workflow in a year.

Flow in general, and for every team member (BA / PM / Dev / QA) is described in our wiki, as well as terminology, list of labels to apply (there is a special glossary for labels). I explain the workflow as a sequence of steps, so that there is an instruction in case of emergency, or a new person onboarding.

On Duty Teams

During the first week we started to ask developers to fill a small questionnaire to find out how often they are distracted from new feature development by urgent client requests or bugfixes requiring immediate attention. It turned out that significant time (up to 80%) had been taken by Urgent Tasks, which distracted devs and made their work less efficient.

So the management team (well, actually it’s more like PMs, CEO, CTO and Support Lead) decided to establish teams of on call devs. These teams (2 devs: Frontend and Backend) would work only on urgent tickets, which allowed the rest of the team to work on regular tasks, ideally, without distraction.

On duty team concept has been rethought a couple of times, and currently we’re aiming at having 3 devs on call each shift, as client base grew significantly, and so have the requests, tasks and points of attention. Some of developers still get pulled to urgent tasks, because SkuVault heavily relies on integrations with other SaaS / eCommerce / Shipping systems, that change their APIs, improve their products, and may occasionally alter the way they interact with our system. And on duty devs may have questions for the developer who built the integration originally.

However, the concept itself proved to be extremely helpful, and overall the issue is resolved.

Mentorship

It’s a common thing to establish, when you have senior and junior devs 🙂 In order to clarify overall system architecture questions, seniors mentor other developers and code review.

Ticket Description Standards

Creating a clear guidance on filling out the fields and required info on a bug / new feature / or any other issue type is essential to streamline development.

Rules: Filling out ticket fields in JIRA - Project Management (SV) - Agile Harbor IKB 2016-06-17 12-01-08

Changing Kanban approach to Hybrid Scrumban

In a nutshell, a year ago development boards (one for planning, one for development) included lots of statuses, was extremely heavy (as you gotta display ~1k tickets), and hard to manage. Kudos to Tim Jannace and the team , who managed to bravely (and successfully) operate and maintain this board!

However, an agile board should focus on one goal: to show a piece of flow relevant for particular scenario / area. So those two boards were split up, so that each board reflects a single scenario:

  • Development Board, where the tickets transition from ToDo to Ready to Release: Scrum Board;
  • Urgent Board, which is used by on duty teams, and includes only Critical and Blocker tickets: Kanban Board;
  • Quality Control Board, where developers and test managers can to see the scope of tickets they need to review: Kanban Board;
  • Release Board, for the release manager to overview and manage the tickets that should be merged to master: Kanban Board.

There are boards for DevOps tasks, of course, as well as for other projects, but developers mostly have to check 2 boards maximum. And both of the boards are easy to use and lightweight.

On the other hand, pure sprint -> release cycles do not reflect how SkuVault operates, because of the Urgent bits that need to be released almost daily. So sprints are more like folders here, which allow us to forecast approximate or particular start / release dates for the tickets, and limit feature scope in a given time period. That’s why it’s called ScrumBan 🙂

Notifications, Due Dates, etc

I’ve also established automated email notifications on Pull Requests or Tasks are not Reviewed / Tested for more than 2 days.

We started to use labels trigger notifications for tickets that will soon miss due date, or for ones that shouldn’t be rescheduled.

There are a lot of other specifics, changes, undergoing improvements – over the year team grew significantly, as well as number of clients – and we adjust the company flows accordingly. Developers look motivated, and I couldn’t be happier to work in such an environment.

Key findings this year:

  • Don’t make a release the goal itself. Quality product is the goal. So you can skip a release or two, but deliver something good. Even if there are lots of clients,they would understand the importance of stability, not the feature they want firsthand;
  • Write up retrospectives on problematic moments, so that you solidify foundation of your experience for yourself and others. Try to gather additional data and opinions inside the team, in order to provide a broader angle to the problem;
  • Make everything possible to have a good human relationship with developers and other team members. You are colleagues, and a good person will always try to do her best, if she’s motivated (see motivation reference article);
  • Horizontal hierarchy and a little bit of dev anarchy is always good. Every team member should have his voice at least heard;
  • Always update team feedback on how things are, this is essential to keep the flow up to date and address concerns that devs may have. Cause you know, in IT, team is what defines success, and good manager’s work is to facilitate work and motivate the people;
  • Maintain comfortable release pace for the team and the clients;
  • Read professional literature, but don’t forget to check how this works in reality 🙂
  • There is always room for optimization. You just don’t have enough time! You can spend days micromanaging things, to extrapolate optimization on global flow later. Neverending exciting job.
  • Maintain work/ life balance. Don’t let team overwork.

Aside of your professionalism, key things to stay motivated are team spirit and ability to apply and improve your skills. For the past year we became mature, overcame challenges, and continue to create awesome WMS for our clients. Looking forward for the next adventurous year at SkuVault 🙂

Thanks to Ksenia, Slav and Kim for the review, and SkuVault team for the support.

Tuning up Scrum Approach

IMG_2016-06-17 13:21:33

Recently my colleague, Tim, decided to try out Planning Poker, to have better estimations. Planning is essential, and scrum already offers a framework of how to deal with planning. But over the course of my work and experience with scrum techniques, team usually shapes

Previous experience showed that daily scrum meetings are merely pointless. Direct communication / skype / IM is much more efficient. Especially in distributes teams.

And following each and every ritual from scrum routine is time and efficiency consuming during the first iterations, since agile methodologies need a good deal of instructions. Usually, after some time teams shape up scrum as they want, and it just works, so from my experience it’s not essential to follow scrum by the book (Agile Estimating and Planning book by Mike Cohn, written 10 years ago). Here are additional thoughts on why our transatlantic distributed team doesn’t fully fit into planning classic scrum and it’s rituals:

  • Due to out product having two versions, we sometimes have developers jumping in and out of projects;
  • Since we’re distributed and flexible – it’s hard to incorporate planning poker with all it’s “team that takes the hint” practice. Distributed team of two-three people can handle classic scrum, but not bigger one.
  • In order for classic scrum to work, team must be onsite (together in one place), and everyone should be in one timezone. That’s not our key point, we’re strong in our flexibility, adapting to challenges and different projects. Scrum meetings are not that effective, when one part of the team has finished work day, and the second one only starts with the fresh brains (smile)

Story Points vs Ideal Days and time estimates

  1. Story Points are valuable when it comes to relative complexity (e.g. that task is twice more complex as this one), and when the team has sort of fog of war before them. However, when elaborating on stats from burndown chart and calculating Focus Factor, we go to the point of calculating how many points / ideal days of uninterrupted development do we need. All because we need to know how to squeeze features into sprint timebox.
  2. If you work with JIRA
    1. Story Points are not as comfortable to work with, as Original Estimate field. Subtask story points do not sum up in parent ticket, unlike original estimates.
    2. Story Points lack ‘remaining story points’ bit, which is uncomfortable once user story has spilled over to the next sprint. Original Estimate, in this case, can be complimented with Remaining Estimate field.
  3. It’s not convenient to estimate buffers for unplanned work with Story Points.
  4. At the end of the day, Story Points are calculated to that very same hours developers spends effectively inside a timebox. Do we need an additional layer of calculations, if it will eventually come to measuring time?

Planning Poker

Planning poker is a ritual before the sprint, where the team (devs, qa) estimates upcoming user stories by a consensus-estimate (average estimate of all team members), assigns user stories to developers, and discusses possible roadblocks collectively.

Don’t have anything against that, but it often comes out time consuming (not that critical as it sounds, actually), and shows lack of detail from other estimators. Moreover, planning poker usually means that devs themselves think about which ticket is to take, which is quite hard to do when we have such a vast scope (~1300 tickets in backlog) + ~2 sprints planned ahead.

But let’s omit devs and tickets self-assignment and time consumption. That’s all tunable.

There are online tools for planning poker:

Estimations and Forecasting

Man Day !=  Calendar Day, because developer gets distracted during man day. So none of these terms reflect what we need.

Story Point is too abstract. Let’s use Ideal Day term, meaning 6 hours of undistracted work.

Key questions to answer when making a good plan (cynical comment: plan is worthless, planning is essential, as Napoleon said) are the following:

  1. How many ideal days on average are in sprint.
  2. How many ideal days can certain developer actually works per sprint.

Once again, Ideal days != calendar days.

Buffers and Planned Days

While starting to estimate and plan back in August, I started by making buffer of 0.5 day, thus making development occupy other 4.5 days in the sprint.

Currently it’s 1.5 days buffer, and 3.5 days of development. This may not be enough, as I’m continuing to tune and gather stats on that. I think that somewhere closer to 3 days is tolerable.

MONTH
TIME BUFFER / SPRINT
PLANNED DAYS
August 0.5 4.5
September 1 4
October 1 4
November 1.5 3.5
December 1.5-2 3-3.5
  • I’m aiming at 3 days of development for devs, and 2 days for code review / scope creeps / finishing up tickets that are reopened;
  • Time buffer includes time for Code Review, fixing Reopened Tickets, other distractions;
  • Planned Days = Ticket Estimated in Ideal Days multiplied by Complexity Multiplicator.

Such empiric way is basically the same focus factor scrum is offering, but without a layer of story points that you later need to convert. And btw, it falls into same ratio I had during previous two projects, which is 2/3 development, 1/3 buffer for fixes and everything else. Seems like more or less ratio across projects then.

Complexity Multiplicator

Plus we have complexity factor, which helps to form buffers. Complexity multiplicator is a combination complexity and unknown unknown. 3 ticket complexity levels:

  • easy <x1.2>
  • moderate <x1.5>
  • complex <x2>

The common equation for one person will look like:

Sprint = SUM(User Story x Complexity Factor) + Time Buffer
5 days = SUM(User Story 1 x Complexity Factor; User Story 2 x Complexity Factor) + 1.5

Individual numbers differ among developers.

All in all, these are estimation basics. Questions asked will add up to this post. Meanwhile, some literature to read:

resized_high-expectations-asian-father-meme-generator-you-are-scrum-master-why-not-scrum-phd-f56adf

I’m not telling pages behind those links are true / correct, but they are certainly allow to overview issues from different angles.