Wednesday, November 24, 2010

Buy vs. Build Decisions & User Stories

Where does the typical engineering Buy vs. Build decision making process happen within Agile software development?  How does Agile's User Stories help us with this decision making process?



Case Study
In the late 1990s I worked with a talented group of people creating a product to deliver high speed Internet service via satellite download links.  The version 1.0 product was done and functioning well, venture capital was secure for version 2.0.  There was a window of opportunity to release a 2.0 product into the market place and we were racing to that market place with a competitor.

Although we were not using any formal Agile process (the term had yet to be coined in Snowbird, UT), we were like many start-up companies using such a lightweight process that it had no name.  It is best to describe the development process as "just make good decision - and do it fast."

One of the features for the 2.0 version was greatly enhanced product licensing.  The new licensing feature wish list from Marketing had these desires:
  • license codes easily generated & transmitted to customers
  • demo license & timed trial licenses expire
  • annual licenses managed & easily renewed
  • perpetual licenses
  • add-on product features individually licensed
The existing 1.0 licensing algorithms were hand made by the company and known to have  some defects.  It was a much simplified and last minute design that allowed simple on/off behavior based on some hash-key techniques.

We were not using Agile User Stories and we were not estimating in relative story points and deriving duration.  However, we had experienced developers working in a known domain.  We estimated the License Management feature to be about 2 man-months of work for the team.  We had the "What" of the story - the "How" was up to the team.

Doing some initial design investigation, there appeared to be significant risk in designing our own solution for this complex problem space.  I had previous experience with many product's license management tools, and recommended we investigate a buy vs. build decision.  Our 2 man month estimate gave us a starting point for the data going into the decision.  In rough terms that would be $200,000 of development time.  Along with the opportunity cost of not doing other core competency development on the satellite networking code.  What would our alternatives cost?

An alternative solution was FlexLM - a best of breed license management solution that ran on all of our target platforms, except one.  That one missing platform was in development and could be considered functional beta on Novell's Netware.  It provided all of the features desired by our marketing group and provided an easy to integrate API for the development team.  This solution was going to reduce our work load to a week, and cost and upfront investment with recurring annual fees.

Working on our BATNA (Best Alternative To a Negotiated Agreement) with the FlexLM company we expanded our offering by agreeing to use our core competency, Novell Netware development, to help them with their Novell beta of FlexLM in exchange for waving the initiation fees.  This required about a week of consultation time, sharing code bases and cross-compiling and debugging techniques on Netware, which was our forte.

Our solution then was to use the FlexLM product, integrate into our code base their simple API for license management and pay the annual fees, in exchange we consulted with their development team on their code port to Novell.

Application of User Story model
Given that this case study took place before the advent of Agile User Stories, one must make some assumptions to draw conclusion on the usefulness of User Stories to the Buy vs. Build decision.  The decision is an economic model based upon the scarcity of capital, and the trade-offs of opportunity cost.  The inputs for the decision are monetary amounts, however, Agile User Story units of Story Points for effort don't compute.

Can one derive the necessary dollar amounts from the Story Points on an Epic feature to use in a buy vs build decision.  Yes, using a team's known Velocity (Story Points completed per Sprint) and team cost per Sprint (typically about $100,000 for a 7 person team) the cost of a feature may be derived - do the math.  The assumption is that a known velocity is applicable to the domain.

In the case of the license management this might not be true.  There were known risk associated with developing outside the core competency of the team.  These risk would tend to increase (not decrease) the cost of in-house development.  In the case study these risk were mitigated by the purchase decision and the partnering agreement.



Futher reading:

Using Agile for Buy Vs. Build Decisions - IEEE Xplore Digital Library - Agile 2008 Conference


Post a Comment