The document discusses how AI can be used to enhance enterprise architecture. It describes analyzing documentation to automatically detect relationships between system elements and classify projects. Natural language processing is used to extract entities from project descriptions and classify them. A graph is generated from the relationships to help identify migration paths between current and future states. Weights can be added to the graph to analyze transition complexity and costs. While fully automated transition planning remains difficult, AI shows promise in augmenting enterprise architects' work by recognizing patterns and analyzing large amounts of documentation.

2. AI
ENHANCED
ENTERPRISE
ARCHITECTURE

4. DISASSEMBLING MY
GRANDFATHER’S RADIO

5. LOOKING INSIDE

6. MAKING SURE IT STILL WORKED

10. HOW THEY WORK

11. HOW THEY WORK
don’t

12. HOW EVERYTHING FITS

13. doesn’t
HOW EVERYTHING FITS

14. WHAT IS ARCHITECTURE?

15. IT ISN’T SITTING IN AN
IVORY TOWER

17. ARCHITECTURE: FUNDAMENTAL CONCEPTS AND
PROPERTIES OF A SYSTEM IN ITS ENVIRONMENT
EMBODIED IN ITS ELEMENTS, RELATIONSHIPS, AND
IN THE PRINCIPLES OF ITS DESIGN AND EVOLUTION
ISO/IEC/IEEE 42010:2011

18. A System
It’s elements
It’s environment
It’s relationships
principles of design
& evolution

19. A System
It’s elements
It’s environment
It’s relationships
principles of design
& evolution

20. A System
It’s elements
It’s environment
It’s relationships
principles of design
& evolution

21. A System
It’s elements
It’s environment
Relationships
principles of design
& evolution

22. A System
It’s elements
It’s environment
It’s relationships
principles of design
& evolution

23. WHAT KEEPS ME AWAKE AT NIGHT
AS AN ENTERPRISE ARCHITECT?

24. IT’S NOT CLOWNS

31. CAN WE APPLY AI?

32. CAN WE AUGMENT EA
WITH AI?

33. IS IT SMART ENOUGH?

34. HUMAN + ELEPHANT
= CHAIR. GOOD JOB.
https://www.theregister.co.uk/2018/08/28/ai_image_recognition_tricked/

35. AI ALGORITHMS ARE
IMPROVING FAST

36. SOME OF THE TASKS ARE
REPETITIVE

37. SOME OF THE TASKS REQUIRE
RECOGNIZING PATTERNS

43. TOOLS USED

44. Graphviz
Viz.js

45. CONCRETE EXAMPLES

46. PROJECT
CLASSIFICATION

47. HUNDREDS OF PROJECTS
▸ Hundreds of projects are started every year, how can you keep track?
▸ In a perfect world, projects are driven by business needs, that drive
architecture changes, and those architecture changes drive project definitions.
▸ In the real world (at least in mine) it is a mix of strategic projects (coming from
Architecture changes) and tactical projects (coming from someone that saw a
nice toy, or from someone with an urgent business need). And what we’re
doing here is making it possible to monitor the latter kind of projects.

48. LEARN SPACY NEW ENTITIES
Application
List
Data
List
NLP
MODEL
Actors
List

49. CLASSIFY PROJECTS
PROJECT ID TITLE DESCRIPTION
P-0001
Onboarding
clients
The clients of
P-0002 Adding e-ID The web site is
P-0003
Add PSD2
support
The clients of
P-0004
Onboarding
clients
Freeing the
PROJECT ID Impacted Entities
P-0001 actor:CUST[2], data:MT[1], data:MX[1]…
P-0002 application:PRIVWEB[1], actor:CUST[1]…
P-0003 actor:CUST[1], data:PORTFOLIO[1]…
P-0004 actor:CUST[2], data:MT[1], data:MX[1]…
Run
Project List
Through
Model

50. GET THE APPLICATION LIST
▸ Interviews
▸ CMDBs
▸ EA Tools

51. PROGRAM MANAGEMENT OFFICE (PMO)

52. GET THE LIST OF PROJECTS INTO SPYDER

53. RUN THE CLASSIFICATION SCRIPT

54. ANNOTATIONS COME OUT

55. ANNOTATIONS EXPLAINED
"P-0001","PROJECT","Onboarding clients Swift and conversion MT/MX","The clients of our small entities need to be
on-boarded so that they are no longer managed in a different system. We'll need to define the portfolios in Pimmy and
add accounts to the CBS."
"P-0001", "actor:CUST[2]", "data:MT[1]", "data:MX[1]", "data:PORTFOLIO[1]", "application:PMS[1]", "data:ACCOUNT[1]", "application:CBS[1]"
Marketing must have a look at this because we’re handling customers, MT/MX is used so there’ll
probably be a need for integration —> integration architect. Several data objects are touched so we
need our data architect to look at it and finally our core banking and our Portfolio management
systems’ architects need to be involved.
INPUTOUTPUT

56. FINDING LINKS

57. AGAIN…
▸ In a perfect world the architect would open Sparx’ Enterprise Architect (or any
similar tool) and start looking for links in the diagrams.
▸ We do that, but now we also complement this with links we extract from
documentation automatically

58. MANUALLY MAINTAINED LINKS

59. AUTOMATIC LINK DETECTION

60. WE TAKE A POWERPOINT

62. AND THEN GENERATE THE DEPENDENCY GRAPH
DOC-1 actor CUST 2
DOC-1 data MT 1
DOC-1 data MX 1
DOC-1 data PORTFOLIO 1
DOC-1 application PMS 1
DOC-1 application CBS 1
DOC-2 application PRIVWEB 1
DOC-2 actor CUST 1
DOC-3 actor CUST 1
DOC-3 data PORTFOLIO 1
DOC-3 application PMS 1
DOC-3 application CBS 1
DOC-4 actor CUST 2
DOC-4 data MT 1
DOC-4 data MX 1
DOC-4 data PORTFOLIO 1
DOC-4 application PMS 1
DOC-4 application CBS 1
DOC-5 application CBS 1
DOC-5 data MX 1
DOC-6 data MT 1
DOC-6 data MX 1
DOC-7 actor CUST 2
DOC-8 application CBS 2
DOC-8 application PMS 2
DOC-8 actor PBO 1
DOC-8 actor FM 1
DOC-9 data PORTFOLIO 1
DOC-9 application PMS 1
DOC-9 actor CUST 1
DOC-9 application DWH 1
GraphViz
Generator
Script
strict graph G {
ratio=fill;
CUST [ label="CUST" fillcolor="#FCFAB5" shape="oval" style="filled" ];
MT [ label="MT" fillcolor="#C2FCB5" shape="parallelogram" style="filled" ];
MX [ label="MX" fillcolor="#C2FCB5" shape="parallelogram" style="filled" ];
PORTFOLIO [ label="PORTFOLIO" fillcolor="#C2FCB5" shape="parallelogram" style="filled" ];
PMS [ label="PMS" fillcolor="#C8FBF6" shape="box" style="filled" ];
CBS [ label="CBS" fillcolor="#C8FBF6" shape="box" style="filled" ];
PRIVWEB [ label="PRIVWEB" fillcolor="#C8FBF6" shape="box" style="filled" ];
PBO [ label="PBO" fillcolor="#FCFAB5" shape="oval" style="filled" ];
FM [ label="FM" fillcolor="#FCFAB5" shape="oval" style="filled" ];
DWH [ label="DWH" fillcolor="#C8FBF6" shape="box" style="filled" ];
CBS -- PMS [ label="P-0008" ];
CBS -- FM [ label="P-0008" ];
CBS -- PBO [ label="P-0008" ];
PMS -- FM [ label="P-0008" ];
PMS -- PBO [ label="P-0008" ];
DWH -- PMS [ label="P-0009" ];
DWH -- CUST [ label="P-0009" ];
DWH -- PORTFOLIO [ label="P-0009" ];
PMS -- CUST [ label="P-0009" ];
PMS -- PORTFOLIO [ label="P-0009" ];
CUST -- PORTFOLIO [ label="P-0009" ];
CBS -- CUST [ label="P-0003" ];
CBS -- PORTFOLIO [ label="P-0003" ];
CBS -- MT [ label="P-0004" ];
CBS -- MX [ label="P-0004" ];
MT -- PMS [ label="P-0004" ];
MT -- CUST [ label="P-0004" ];
PMS -- MX [ label="P-0004" ];
MX -- CUST [ label="P-0004" ];
PRIVWEB -- CUST [ label="P-0002" ];
}

63. DEPENDENCY GRAPH

64. FINDING MIGRATION PATHS
Bus Inf
App Tech
Bus Inf
App Tech
CURRENT STATE
FUTURE STATE
TRANSFORMED ENTERPRISE
Transition
Transition

66. ADDING WEIGHTS
▸ Adding weight to the edges:
▸ how much does each edge cost us? Example: we need MQ Series
vs. it’s just an API call
▸ how stable is each edge?
Example: it runs every day for months on end vs. every day we need
to go clean some messages because of format errors.
▸ how “online” is the edge?
Example: it is “online” vs “batch running every hour” vs “batch
running every day”
▸ how “easy” (i.e. how much time or budget) is it to replace by new
tech?
Example: it is something written in COBOL using some dark 1990
broker software vs JMS messaging vs SOAP API that can be easily
re-written in JSON
w1
w2
w3
w4
w5
w6
w7
w8
w9

67. SET KPIS
▸ The transition must by fast and can
cost whatever it may
▸ The transition can take all the time
you need but must cost as little as
possible
▸ A mix of the above

68. VERY SIMPLE EXAMPLE
www
DWH
CBS PMS
CRM
NW
www
DWH
CBS PMS
CRM
NW
x
x
x
www
DWH
WMS
CRM
NW
x

69. TEXT

70. TEXT
UNABLE TO CREATE
DETAILED ACTION PLAN
STUCK
Dynamic
Data
SAVINGS
SCHEDULE
SEQUENCE
COMPLEXITY
PRIORITIES
INVENTORY
DEPENDENCIES
AUTOMATION
SIZE
ANALYSIS
MIGRATION
ASSUMPTIONS
INTEGRATION

72. 169 NODES
1368 EDGES

73. 140 NODES
1068 EDGES

74. 110 NODES
622 EDGES

75. 77 NODES
409 EDGES

77. An AI Architect in a his Ivory Tower

80. Mulțumesc
@patrizz