About
Adam DeJans Jr.
Applied operations research scientist. Author of The Decision Product.
- 13+ years designing and deploying large-scale optimization and simulation systems
- Work spanning supply chain intelligence, optimization engineering, and decision science at Amazon, Gurobi, Toyota, and Ford
- Four published books, and one US patent
- M.Sc. Industrial Applied Mathematics and B.Sc. Mathematics, Oakland University
- Co-founder, Bit Bros LLC
- Personal site
- adamdejans.com
- adam@bitbrosdata.com
- in/addejans
- All books
- Amazon author page
- Research
- Google Scholar
- Code
- GitHub
I am an applied operations research scientist. I have spent more than a decade designing and deploying large-scale optimization and simulation systems for supply chain, logistics, production planning, and allocation, which is to say the unglamorous work of getting a decision into production.
I have worked across automotive, consulting, solver engineering, and retail supply-chain intelligence, and I am a co-founder of Bit Bros LLC.
I am not interested in models that live in notebooks. I am interested in policies that still work after the meeting ends.
What I actually do
Three things, in descending order of time spent. I coach individuals, mostly technical people whose analysis is correct and whose recommendations are not getting run. I run audits and workshops for organizations where a decision-velocity problem turns out to be a brief problem. And I write, both here and in the book, which is what makes the other two possible.
Why this subject and not another
Because the gap is unusually large and unusually tractable at the same time, which is a rare combination.
Most performance problems in technical organizations are hard. Forecasting is hard. Allocation under real constraints is hard. Against that, the distance between a correct analysis and an installed rule is a problem with a known method, a short learning curve, and effects visible inside a single planning cycle. It stays unaddressed largely because it gets miscategorized as a communication problem and therefore assumed to be a matter of personality.
It is not. It is a method, and methods can be taught to people who like methods, which is exactly who this is for.
Other books
Four published so far, with a fifth in progress. The full set is on my Amazon author page.
The Decision Factory
Policies, uncertainty, and why a static optimal plan so often perishes on contact with ordinary operations. That book is about what a decision system is. This one is about what you do when the email arrives. They are companions, not substitutes.
The MILP Optimization Handbook
Formulation and practice for mixed-integer programming, for readers who want the mathematics this book deliberately does not teach.
The Linearization Handbook for MILP Optimization
The modelling tricks that turn an intractable formulation into one a solver can actually handle.
You Got the Data Job… Now What?
For people at the start of the path this book assumes you are already some way down.
Where the method comes from
The pyramid is Barbara Minto’s, taken and pointed at a decision rather than at a deck, which is the only liberty taken with it. The trees are a narrowed version of Conn and McLean’s, aimed only at pruning. Diagnosis before prescription is Rumelt. Interests rather than positions is Fisher and Ury, used once and on purpose. Systems and local scoreboards is Meadows. What a policy can know at the moment it fires is Warren Powell. The further reading lists all of it, with the chapter each one sits under.
What is not in that literature is the application to the artifacts technical people actually produce, and the specific problem that a correct model can be complete and still change nothing. That translation is what the book is for.
How to get in touch
Directly, at adam@bitbrosdata.com, or through the contact page. I reply to every serious inquiry within two business days.
This site is about one book and one method. For everything else, the optimization work, the other books, and whatever I am currently building, adamdejans.com is the place.
Positions
Things I believe that are not universally held
Stated plainly so you can decide whether to engage me before rather than after.
Accurate decoration is still decoration
An analysis that could not have changed which alternative was chosen did not become valuable by being correct. This is the sentence technical people find hardest, because accuracy is the thing they were trained to defend, and it is the one the whole method rests on.
Write the answer down before you look at the data
This sounds like confirmation bias and is the opposite. An unstated prior cannot be falsified, because it deforms quietly to fit whatever the data shows. Writing the candidate ranking down, with its kill conditions, before any query runs is what converts a bias you cannot see into a claim you can attack.
Heuristics are first-class, and legibility is a feature
A rule a planner understands and follows frequently beats a better rule they route around. Effective value is the value when followed multiplied by the probability of being followed, and the second term is almost never estimated because compliance is assumed.
Most proposals to gather more data are not worth their delay
"Let us get more data first" has a price, and the price is the freeze that passes and the quarter that runs on the current rule. Against that, the value of the information is bounded by the probability it flips the ranking times the cost of being wrong. That calculation takes five minutes and is almost never done.
Next
Start with the free material and decide from there
The method is complete and ungated. If it changes what gets run after your next meeting, that is the only evidence worth acting on.