2 of 47

Hello

David Colls, semi post post technical

Project & program management, organisational change
Visualisation, and occasional development

Today brings together a few of my favourite things:

Finding new ways of looking at problems
Exploring with maths, metaphors & visuals
Sharing what I’ve found

3 of 47

Today

Scaling production - factory
Sustaining software - home loan
Scaling change - nightclub
Conclusion

4 of 47

Factories

CC BY 3.0 Andreas Praefcke

5 of 47

So you run a factory

You produce stuff
Once upon a time the game was to scale production
Lean teaches us that overproduction is waste and that we should allow customers to pull production
But even the most customer-driven organisations need to scale production

6 of 47

http://media.treehugger.com/assets/images/2015/11/tesla-gigafactory-solar-roof-01.jpg

7 of 47

Factory maths

So let's assume that what we're producing is valuable
Most managers will think in terms of fixed, variable, marginal and total cost
Accounting and economic (opportunity cost) lenses. Lean teaches us that missing customer expectations is cost

C = f + v•N

M = dC/dN = v

8 of 47

Visualising factory maths - marginal

9 of 47

Visualising factory maths - total

10 of 47

Teh big-O of factories

Less MBA, more CS pls
Factory production is O(N)
At large enough scale, variable cost dominates fixed (no matter f, v)
kthx

11 of 47

Visualising factory maths - big-O

12 of 47

Managing with factory maths

Shift variable costs to fixed to scale
Depending on the anticipated volume of production, we can determine sensible fixed cost investments

13 of 47

So can we manage software like a factory?

What, no, of course not - it's creative, not just churning out widgets at volume
There’s variation - if we strive for efficiency we lose responsiveness
But Lean applies to factories and software
Ok, so maybe some concepts are applicable and others are not

14 of 47

What does software produce?

Experiences for users
But as a software delivery team, we don't produce individual experiences, the software does
Our product is a factory

15 of 47

To scale production of experiences

We can manage operations with factory metaphors and maths
We can automate, simplify, improve quality to shift variable cost to fixed
To delivery a variety of experiences, we can even use Lean concepts like SMED
And therefore production cost may become more predictable and reduce over time

16 of 47

To scale the factory

However, changes to our factory - software delivery - often become more expensive and less predictable over time - why?

17 of 47

What stops us sustaining software change?

Many would say technical debt, being the retarding complexity in a code base that builds up over time

Retarding - slows you down
Complexity - is the root cause
Over time - half right

19 of 47

The cost of debt

Adopting a debt metaphor implies an equation to cost this retarding complexity in software delivery, the same one we would use to calculate the cost of a home loan

C = P • (1 + r)^t

M = dC/dt = Q • (1 + r)^t

(Ignoring repayments)

20 of 47

Teh big-O of technical debt

O((1+r)^t)
Over a long time, interest rate dominates
And marginal cost is the same
But how do we manage the passage of time?

21 of 47

Management with the debt metaphor

We can’t manage the passage of time
Time is a proxy for change
We borrow responsibly
We make repayments in a timely manner
The bigger the outstanding balance, the bigger or more frequent the repayments
We get in trouble when we borrow too much, or don’t make repayments in a timely manner

22 of 47

Is there another way to look at it?

Technical debt is a great metaphor, but...

Can we talk about the cost of change directly?
Can we use a model that applies equally to business as software change?

24 of 47

What is the dominant cost of change?

It's the previously unanticipated or knowingly deferred cost of interaction between changes.

Although we have fixed costs, and each change has its individual cost, the dominant factor is that each new change interacts to some extent with every previously introduced change.

25 of 47

Visualising interaction costs - marginal

26 of 47

Visualising interaction costs - total

27 of 47

Visualising interaction costs - big-O

28 of 47

So why nightclub?

Because in a nightclub, a problem could occur between any pair of patrons

But nightclubs want to scale patrons through the door

How do they manage scale?

By screening at the entry, by creating separate spaces, and by ejecting patrons who have outstayed their welcome

So let’s talk about socialisation costs

30 of 47

Interaction cost maths

Each change i has variable cost a & interaction cost b

Add up all N for total cost

Re-write 2 with 1

Expand sums with average values for a and b

31 of 47

Interaction cost maths & big-O

Re-write like our factory maths, use s for interactions

Figure out marginal cost

Marginal cost for a new change is O(N), dominated by number of changes to date

32 of 47

Change heat map

Visualise the cost of pairs of interactions as a heat map
Note that cost per pair varies significantly
For N elements, we have approximately N²/2 interactions

33 of 47

Change interaction heat map

Visualise the cost of pairs of interactions
At scale
We get in trouble when we fail to manage these interaction costs
Explore 4 management techniques: socialise, screen, seclude, surrender

34 of 47

Socialise

Make sure every change plays nice(ly) with every other
Equivalent to making technical debt payments
Shift to variable cost
Same N²
And we can only guess how to socialise with future changes

35 of 47

Screen

Be selective in the changes you adopt
Prioritisation and deliberate capacity constraint
Shift to variable cost
Screening ½ cuts cost to ¼

36 of 47

Seclude

Eliminate interactions between groups of elements
Architecture, in the broad sense
Shift to variable cost
2 buckets → ½ N² cost
3 buckets → ⅓ N² cost
...
N buckets → variable cost only

37 of 47

Surrender

Screening at the other end of the lifecycle
Actively manage out changes
Shift to variable cost
Surrendering ½ cuts cost to ¼

38 of 47

Where else do we see socialisation costs?

Start-ups - screened or surrendered - more agile than incumbents because they have less history of change
Product management - secluded - Facebook/Messenger/Instagram/Hyperlapse/Layout/Boomerang/etc separate apps
Hiring - screened & socialised - TW interview process and TWU shifts interaction cost to variable cost
Brand architectures - secluded changes - Unilever’s Dove can campaign for real beauty while Axe/Lynx offends

40 of 47

Design == change?

From the change perspective, deployment is a variable cost & ops is fixed
Designing the change is what attracts the interaction costs
Design in the broad interpretation covers UX, IA, coding, etc
So can we say design is O(n^2) too?

41 of 47

Bonus Metaphor

42 of 47

Pivoting

I.e., rotating
Rotational inertia is O(r²) - mass far from the centre has a bigger effect
Though we may be accused of mathematical mysticism, consider the impact of changes far from your purpose on your ability to pivot

Axis of rotation

Increasing inertia

43 of 47