UX STRAT Online 2021 Presentation by Mike Kuniavsky, Accenture

COMPUTATIONAL
DESIGN
Niche manufacturing, AI and
A research framework from Accenture Labs
Mike Kuniavsky
UX STRAT 2021

Copyright © 2021 Accenture. All rights reserved. 2
EXECUTIVE
SUMMARY
• We propose enabling niche manufacturing using computational
design, so manufacturers and retailers can reach small markets faster,
more accurately, and with less waste.
• Niche manufacturing takes audiences identified by digital marketing
and makes unique products for each market.
• Computational design is a software-first approach that maximizes
designer productivity using data and AI assistants throughout the
design process.
• Designers and AI assistants collaborate to interpret live data streams to
create product design spaces with infinite variations, instead of
unique product designs.
• This approach requires new design practices and AI design assistants,
which Digital Experiences in Accenture Labs is developing.

00About me +UX

…backintheJurassicWeb(1994-2003ish)

…throughIoT(2004-2012ish)
4 5
Living Cube · Living
products Living Cube · Living products
each of these ever-changing needs. Specifi-
cally, the Living Cube introduces a built-in
food cooling and heating system to the living
room without adding additional furniture
to an already busy environment.
form
The Living Cube evokes a traditional coffee
table, but with a distinct difference: when the
leather ‘saddle’ slides to the side, the Cube’s
design ceases to be furniture and becomes
instead a technology, complete with displays,
controls, functional surfaces, and storage
compartments.
The cooling compartment includes a bottle
carrier for four to eight bottles, reflecting
the times when the living room serves more
as a drinking area than an eating area.And
for the times when snack food is the focus,
the heating surface has two different com-
partments underneath: one for the storage
of small induction heating cookware and an-
other for silverware.
interaction
When closed, the Living Cube emits only a
slight glow of diffused white LEDs to indi-
cate the temperature of the cooling compart-
ment, communicate that the Cube is on,
and help identify it in the dark.
In recognition of the task-specific way in which
people will use the product, the sliding motion
of the saddle has been designed to create a
small transformation ritual that signals the
transition from one kind of activity to another,
while revealing only the controls for either the
cooling side or the heating side.
← →
dimensions
Overall 50 by 90 by 50cm
Cooling unit 32 by 46 by 30cm
Warming zone 36 by 17cm
Saddlebag 24 by 22 by 6cm
materials
Leather,
Satinized glass,
Stainless steel,
MDF
• •• • • •• •• •••• •• •• ••••••••••• • ••• ••••• •• •• ••• • •
•••••••••• •••• •••••••• ••••••••••• ••••••• • ••• •••• ••••••• •
••• ••• •••••••••• ••• •••••••••••• •• •••• •••••• ••••••••••••• •
••••••• ••• ••• •• ••• •• ••••••••• ••• •••• •• •••• •• •• ••• ••
•••• •• ••• •• •••••••••••••••••••••••• •••• •••••• • ••••••••••••
••• ••••• ••••••• •• •••••• •••••••• •••••••• • ••••• • ••• •• ••
•• ••• ••• •••• •• • ••• • ••••••••••••• ••• • ••••••• •••• ••••

…toDesignforR&D(2012-2019)

Accenture Labs
…since2019.

01Niche
marketing
to niche
manufacturing

Howproductsaremadeandsoldischanging
Markets are smaller and move more quickly
• By geography: maximize regional focus
• By interest: address needs of medium-sized sub mass-market interest groups.
• By time: capitalize on momentary markets
Retailers want to reduce the risk of entering these
smaller markets
• Reduce the cost to enter small markets, and enter multiple markets
simultaneously
• Reduces the complexity of competing in a new market
Manufacturers want to design and engineer more
efficiently
• Reconfigure manufacturing lines quickly
• Reinvent product introduction processes
• Reduce work required in design changes
• Respond to regulatory changes

Continuous,data-drivenproductinvention
NEW PRODUCTS
MARKETING
CUSTOMER DATA
AND MONEY

Today:nichemarketing
Many markets + Same product = Many messages

2025:nichemanufacturing
Many markets = Many Products

Aslotsizegoesdown,productvariationgoesup
(ifyouwanttocapturethesamesizedmarket)
• Today: processes are optimized almost
exclusively for mass manufacturing
• Tomorrow (2022-2025): moving
towards niche manufacturing with
smaller lot sizes and more product
variations serving more diverse users
• Future (2025+): personalized
manufacturing where products are
designed for individuals and produced
in ultra-small lot sizes
• Niche manufacturing is the bridge
between lot size 1M and lot size 1.
P
E
R
S
O
N
A
L
I
Z
E
D
M
A
N
U
F
A
C
T
U
R
I
N
G
NICHE MANUFACTURING
MASS MANUFACTURING
Today
Tomorrow
Future
Zara
Apple
Dell
1 10
6
10
4
102
LOT SIZE
PRODUCT
VARIATION
1
10
102
10
3
Invisalign

Butwedon’tyethavenichemanufacturing.
Why?(atheory)
Many of the pieces are already in place
• CRM gives us customer interest data
• IoT systems give us product usage data
• CNC manufacturing enables automatically-reconfigurable tools
• CAD and PLM systems allow high precision digital product
definition
• There are clear market advantages: capture more total market
share and less total risk by making small bets
Why don’t we have widespread
niche manufacturing?

Traditionaldesigneffortquicklyincreases
beyonddesignercapacitybecauseofthetools
• Design teams have limited resources of:
• Time
• Money
• Knowledge
• Team members
• Designing one product is hard enough.
Designing 10, 100, 1000 unique variations
becomes untenable because design tools
scale linearly to the number of designers.
10x the designs = 10x the designers
• Increasing team size does not enable
teams to make better use of knowledge or
manufacturing flexibility because the
tools are still the same. 1 10
Design
Team
Capacity
Number of product variants
100 1000
Zara*
(300 designers on staff)
Currently
unfeasible
https://www.businessinsider.com/zara-design-process-beats-trends-2018-11

02Computational
design
principles

Human+AI+DATA=IncreasedDesignCapacity
=ComputationalDesign
• Change design effort from scaling linearly
to scaling exponentially
• Augmenting human teams with AI-based
tools has a multiplicative factor on team
effort – this collaboration changes the
math on design capacity
• Designers are supported by AI design
assistants
• AI assistants are supported by data via a
digital thread across the design process
including manufacturers and customers
1 10
Number of product variations
100 1000
AI
DATA
DATA
AI
DATA
AI
DATA
Design
Team
Capacity
HUM
AN
+M
ACHINE+DATA

Computationaldesignvscomputerizeddesign
Computationbakedinateverylevelredefinesdesignand
engineeringpractice
Computerized design (CAD) tools replicate
pre-digital design practices with digital tools.
Super paper, meet super pencil!
Computational design creates design
practices that can only be done with a
computer.
• Data-driven
• Parametric
• Automatically evaluated
• Simulated
• Instead of one-off detailed
representations, computational design
tools create product possibility spaces.
Scriptable CAD is just the start of computationally defined metaproducts.

Computational design:
Many names for the same idea
Computational Design
Parametric Design
Generative Design
Algorithmic Design
Performance-based Design
Evolutionary Design

Computationaldesign
components
1. Software-defined
everything
2. Metaproducts
3. AI-enabled agile design
4. Digital threads

1.Software-definedeverything
Everyaspectisencodedassoftware,notasvaluesinsoftware
• Physical appearance - keep a product on brand
even when its design is completely generated by
software
• Functionality – configuration of options into an
included bill of materials creation
• Firmware – software that enables all the options
to work together
• Digital twin – provision the cloud service that
enables specific functionality, collects usage
information
• Every possible digital product from industrial
equipment-as-a-service to consumer products to
interactive environments

2.Metaproducts
Interlinkeddesignsthatdefinegenerativeproductspaces
• Software-defined parametric metamodels describe an infinite range
of potential product designs, rather than a single product design.
• Changes cascade through the whole design.
• Difficult for products made up of many materials and manufacturing
methods today, but we’ll get there and we’ll start with simple designs.

3.Agiledesignpractices
Nothingiseverfinished
Abandon waterfall design: design isn’t a step in the
product development process. It is the process.
Bring rapid iteration/high-data design pioneered in online
design to physical products.
Expand: Augment design capabilities with AI assistants
• AI design assistants (requirements gathering, constraint
definition, simulation, customer/user data analysis)
• Feedback-driven design, assisted design and
manufacturing (generative design)
Enable: continuous data-driven design + manufacturing
• Digital twins collect data about how products are actually
used. This data is fed back to designers as they design the
next generation of a product
• CRM systems allow for customer niche identification by
behavior/values/desires/associated purchases
Software is a tool for the mind.
- Reas & Fry, 2014

Process Iteration
Data Feedback
Design
Subprocess
Problem
Formation
Preliminary
Requirements
Conceptual
Design
HW & SW
Detailed Design
Testing
Sourcing
Manufacturing
Quality
Verification
Product Launch
Sales
After Sales
DIGITAL
THREAD Simulation
Validation
Simulation and
validation become
parts of every
design
subprocess. Each
of these is
powered by data
on the digital
thread.
4.DigitalthreadforProductDesign
Connectdatasourcesdirectlytoresponsivetools
• As design and manufacturing
processes become increasingly
agile and iterative, teams need
to move fluidly between design
phases.
• Iterations will be powered be
new sources of product and
customer, and manufacturing
data.
• Digital threads connect digital
twins of products, customers,
and processes so relevant
information is always available
to designers at every step.

03Computational
design
practices

Computationaldesignprinciples
1. Designer + AI work as a team
Designers and engineers collaborate with AI assistants on teams.
Each team works together on a representation of the design.
2. Shared representations
At each stage of the computational design journey the designer
and AI take on different roles as they work together on a shared
representation of the design challenge.
3. Everyone does what they’re best at
AI assistants take on tasks that only computers can do, such as
analyzing large amounts of data and evaluating thousands of
variants of an idea.
Designers take on tasks that only people can do, such as defining
goals, evaluating simulations for unquantifiable criteria, and
making tradeoffs.
Snap-to-grid is a primitive assistant
Snap-to-product will offer
autocomplete assistance for
finished designs

Computationaldesignprinciples
DESIGNER + AI CO-ANALYZE AND CO-CREATE
DESIGNER
• Creating constraints
and relationships.
• Define a general style
• Sets functional goals
AI ASSISTANT(S)
• Analyzing data
• Modifies parameters,
generates and
• Evaluates designs
SHARED REPRESENTATION

Thetraditionaldesigndoublediamond
Source: British Design Council, 2004

Doublediamondphasesforcomputationaldesign
aproposal
Discover
• The team captures business, user, and customer needs and wants.
• Merge purchase behavior, IoT sensor, CRM, qualitative research, and business discovery
data.
• Defines goals for a parameter space definition of product value and effectiveness.
Define
• The team parametrically describes the product space
• Creates a metamodel of all potential product designs
• Encodes product goals as fitness functions
Explore
• The team explores the product space by creating and evaluating hundreds to millions
of design variants.
• Fitness functions help prioritize among design variants. Fitness functions can be
functional (such as mechanical performance), cost (such as estimated manufacturing
costs), or
esthetic (whether a design fits brand style guidelines).
Evaluate
• The team evaluates design variants using simulation. Simulations can be algorithmic
(crowd flow simulations in an architectural space) or sensory (AR simulation).
• The team uses results of the simulations to adjust parameter space definition,
metamodel descriptions and fitness functions.
Designers and AI
assistants work as a
team and take on
different roles to jointly
produce different
representations of the
product idea as they
interpret data and
respond to it with new
design constraints and
settings. There are
many different design
and engineering roles
throughout product
development, and
there are many kinds of
AI assistants.

Computationaldesignjourney
Design Goal
AI Assistant Role
Activity
Simulation Engineer
Evaluate subjective design factors
Provide suggestions / make adjustments
Select solutions to move forward with
Designer Role Curator
Generative Designer
Compose Generative Design System
Select fitness functions
Design Director
Constraint Manager
Define problem
Prioritize requirements
Define style grammar
Systems analyst
Quantitative Research
Document business challenges
Ask research questions
Generate insights
Researcher Director
Activity
Provide objective comparisons
Prioritize designs based on multiple factors
Capture feedback from humans
Generate solutions
Present solutions
Refine solutions based on feedback
Document constraints and
requirements
Debate priorities
Define parameters
Collect quantitative field data
Find patterns in data
Present summaries
Raise questions
Shared
Reprensatation
Intelligent research interface
Requirements prioritization
Parametric metamodel
Data-driven need list
Customer genome
Outputs
Generative Design System
Design options on fitness curves
Hi-Fi simulation results
Refined design option
Intelligent Requirement System
Style Grammar System
Generative Design System
Design space fitness map
High-Fidelity Simulators
Design feedback environment
DISCOVER DEFINE DEFINE
EXPLORE EVALUATE
Capture business, user, and
customer needs and wants
Specify problem, functions,
requirements and initial
parameters
Generate possible solutions
Evaluate solutions, refine
requirements and parameters,
select designs to more forward

04Computational
design
explorations

Explore:GenerativeDesignDemo
• A digital twin of a
customer’s driving
style.
• Prototype uses the
digital twin to generate
a unique wheel design
that meets the driver’s
performance style and
esthetic preferences,
while still meeting all
functional requirements.

Explore:ThemeInspiredStyleTransfer
• A product design space
of many potential
designs.
• Prototype generates
many potential 3D
designs based on
product
manager/designer
interest/inspirations or
market performance
data.

Evaluate:IntelligentPlannerToolDemo
• A digital twin of a FEMA
emergency facility
location, based on
photogrammetry of the
actual site.
• Simulated, learning
agents explore
potential layout designs
to explore impact of
various facility layout
decisions. Agents have
individual different goals
(get supplies, give out
supplies, etc), and the
capability to learn from
their mistakes.

“Ideas flow from [a design
material] to us and though we
feel to be the creator we are
involved in a dialogue with [it].
The more subtly we are tuned to
our medium, the more inventive
our actions will become. Not
listening to it ends in failure.”
- Anni Albers, Bauhaus professor, 1982
Computational
Design is a new
design material with
which we are just
beginning a
dialogue.

UX STRAT Online 2021 Presentation by Mike Kuniavsky, Accenture

More Related Content

UX STRAT Online 2021 Presentation by Mike Kuniavsky, Accenture