1. LEMNOS
ALISSA ORLANDO
hustler
MBA
DANE RENNER
picker
MBA
FERDINAND LEGROS
designer
CS + MS&E
MAXIME VOISIN
hacker
CS + MS&E
IAN TAYLOR
hustler
MSX
CRAIG SEIDEL
mentor
110
INTERVIEWS
TODAYDAY ONE
Reduce production
interruptions
in Oil & Gas plants
Increase production
throughput in plants
EXISTING MARKET – BETTER PERFORMANCE
1
2. Setting the Scene
What is a plant, and how big is it?
200k
equipment
items
150
people
$170M
operational
budget
EXAMPLE: PLUTO PLANT, AUSTRALIA
2
3. Setting the Scene
Who are the people?
They have four objectives:
- Increasing throughput
- Reducing production interruptions
- Reducing costs
- Safety
3
Executive
Sign-off authority >$5mm
Plant Manager
Sign-off authority >$0.5mm
Maintenance Manager
Sign-off authority <$0.5mm
Engineer
Sign-off authority <$0.5mm
4. We were a group with AI expertise
looking for a problem to solve...
Week
1
4
Executive
Sign-off authority >$5mm
Plant Manager
Sign-off authority >$0.5mm
Maintenance Manager
Sign-off authority <$0.5mm
Engineer
Sign-off authority <$0.5mm
They have four objectives:
- Increasing throughput
- Reducing production interruptions
- Reducing costs
- Safety
5. We got out of the building to test our first value proposition Week
1
Value Proposition
Increase throughput in plants and factories
using AI
Oil & Gas, Chemicals, Food & Bev,
Pharma, etc etc.
Customer Segment
5
6. … and we received good signals, but we hit a roadblock Week
1
6 agreed 3 agreed 2 agreed
“This sounds really exciting…”
(Engineer @ Shell)
“...I would hesitate to approve it
because of safety risks…”
(Engineer @ Woodside)
“We have done this… it takes a lot
of engineering hours…”
(Executive @ Woodside)
Consulting model
(difficult to scale)
6
7. Many interviewees suggested we apply our AI expertise to
predict when equipment fails
Week
2-3
“Predictive maintenance would be super
useful and has far fewer implementation
risks.”
(Engineer @ BASF)
8 agreed
7
8. So we pivoted to apply AI to a different problem! Week
2-3
“Predictive maintenance would be super
useful and has far fewer implementation
risks.”
(Engineer @ BASF)
8 agreed
8
They have four objectives:
- Increasing throughput
- Reducing production interruptions
- Reducing costs
- Safety
9. We got out of the building with our new value proposition Week
2-3
Value Proposition
Predict when equipment fails to
reduce plant downtime using AI
9
10. … and we received good signals,
but we hit another roadblock
Week
2-3
DEMAND FROM CUSTOMERS
FEASIBLE TECHNOLOGY
AVAILABLE DATA
“We had three separate contractors try to
build predictive models ... there simply isn’t
enough data”
(Maintenance Manager @ Nihar)
2 agreed
10
11. We learnt to embrace a problem-first approach...
not a technology-first approach...
Week
4
“I have no visibility on how
we are going”
“SAP’s user interface sucks”
“Our data is kept in different
places, and they don’t talk to
each other”
“I’m hearing a lot about
predictive analytics!”
11
13. Identity Crisis Interlude
Our divide and conquer interview
style was creating confusion
Competitors
Equipment
Manufacturers
Oil and Gas
Operators
Chemicals
Other
Manufacturing
13
Week
4-6
Week
4-6
14. We heard that a new risk-based way of managing maintenance
has emerged to improve on the classical approach
Week
7
CLASSICAL
MAINTENANCE
PARADIGM
RISK-BASED
MAINTENANCE
PARADIGM
“I maintain all pumps
every 6 months”
“ I maintain all pumps
based on their
current condition & the
consequence of failure
for each”
14
15. We discovered that risk-based maintenance teams
have no modern software solution!
Week
7
(spreadsheet hack)
X
CUSTOMERS :
SOFTWARE :
OPPORTUNITY
CLASSICAL
MAINTENANCE
PARADIGM
RISK-BASED
MAINTENANCE
PARADIGM
15
16. So we pivoted to software for risk-based maintenance.
We got out of the building with our new value proposition
Week
8
Plants in the Oil & Gas industry which
are already hacking solutions
Customer Segment
Software for risk-based maintenance
activity selection to improve
production and safety performance
Value Proposition
16
17. Our interviewees got really excited! Week
8
“This is exciting… I’d like to test it when it’s up
and running.”
(Plant Manager @ Chevron)
“I am actively looking for this!”
(Maintenance Manager @ DuPont)
“We had to develop our own tool in Excel”
(Maintenance Engineer @ ExxonMobil)
17
18. We developed other elements of the business model,
here’s our potential sales process for a plant…
Week
7-9
IDENTIFY
CHAMPION
GATHER
INFORMATION
ENGAGE
USERS
ENGAGE
IT FUNCTION
ADDRESS
DETRACTORS
CLOSE THE
DEAL
OPERATIONAL
INTEGRATION
Org Chart
18
19. The company will continue, with Dane taking the lead Week
10
ALISSA
ORLANDO
DANE
RENNER
FERDINAND
LEGROS
MAXIME
VOISIN
IAN
TAYLOR
CONTINUING MOVING ON TO NEW PROJECTS
19
21. We learnt to solve one problem for a specific customer
subsegment, before solving more problems for them!
Week
8-9
TODAY
+PREDICTIVE
MAINTENANCE
+REPLACE
SOFTWARE
ADD-ONS
+CHEMICALS
+NUCLEAR &
POWER
+ETC.
NEW INDUSTRIES
NEWPRODUCTS
Bottom-up
TAM
$300M
Bottom-up
TAM
$6-8B
21
22. We learnt that sales cycles are 9-12 month-long.
This impacts our fundraising & operational plan!
Week
9
Q1 Q2 Q3 Q4
2019 2020 2021 2022
Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4
Cashreserves
5M
10M
20M
30M
Seed
$2M
Series A
$5M
Start sales
process
Finalize
product
Complete first
sale
22
Editor's Notes
[same script as the video?]
Team
Maxime + Ferdinand, Andrew Ng’s lab, AI for traditional industries [NOT DEFINED]
Alissa [future of work] + Taylor [tech investor] GSB
Dane [industry expert]
Idea Genesis:
spoke to people in traditional industries
produce commodities. it’s crucial to make production efficient
sparked an idea: squeeze as much out of plants as possible
Plant… equipment count, budget, authority.
@Steve: adapt diagram with equipment count / budget / authority
What is a plant / what does it do / industries [emphasize on diversity of plants?]
We will look mostly at these specific plants: [oil and gas?]
How is it set up [nb of pieces of equipment…]
E.g. Pluto
There are X plants like Pluto in the world
Who are the key people running a plant? What are their problems? What are their authority [budget…]
[Put a picture of us in a plant?]
Plant… equipment count, budget, authority.
@Steve: adapt diagram with equipment count / budget / authority
What is a plant / what does it do / industries [emphasize on diversity of plants?]
We will look mostly at these specific plants: [oil and gas?]
How is it set up [nb of pieces of equipment…]
E.g. Pluto
There are X plants like Pluto in the world
Who are the key people running a plant? What are their problems? What are their authority [budget…]
[Put a picture of us in a plant?]
We start with technology.
Looking for a problem!
....... So we got out of the building…..
We were keen to apply AI to industries that are largely untouched by innovation
An early interviewee [use the same title as the character we introduced before?] explained how they [were using Noodle.AI to] adjust inputs to his process to get consistent output quality using AI (tobacco plant)
[tobacco story -> keep it grounded in real life example]
[ great, makes sense to use AI and historical data to optimise production…. but it’s tobacco…. we’re never getting into LLP… then we realized it applies to a lot of industries!]
...... And here's our insight!
- the problem is valid: plants want to optimize production
- the solution is feasible
- but difficult to turn it into a product
Good signals
Some push back on technology “blackbox”
But customization
[are we allowed to use the word “process control”?]
Good signals
Some push back on technology “blackbox”
But customization
[are we allowed to use the word “process control”?]
We keep our AI expertise.
We focus on a different problem: reducing downtime in plants!
We were super excited, we had found a problem where we can apply our solution!
Good signals
Some push back on technology “blackbox”
But customization
[are we allowed to use the word “process control”?]
We were keen to apply AI to industries that are largely untouched by innovation
An early interviewee [use the same title as the character we introduced before?] explained how they [were using Noodle.AI to] adjust inputs to his process to get consistent output quality using AI (tobacco plant)
[tobacco story -> keep it grounded in real life example]
[ great, makes sense to use AI and historical data to optimise production…. but it’s tobacco…. we’re never getting into LLP… then we realized it applies to a lot of industries!]
Have impact today + start building trust with clients who don’t trust startups
Here’s our insight:
So….. we have a problem (reduce downtime)
We have a solution: predict when equipment will fail
But we cannot build the solution!
There was good tech & demand, but insufficient data
We couldn’t add value in the short term so parked this opportunity
So we zoomed out of AI technology, gathered our interview notes,
and realized there are many ways outside AI to reduce plant downtime!
Here’s our insight:
We hit a roadblock with our “where can we apply AI” approach
However, we have found a good problem: reducing downtime in plants!
We learnt that AI is not the solution
That’s fine, because we learnt that there are many other ways to reduce downtime in plants
→ We decide to build whatever it takes will solve the problem (reduce downtime in plants), even if it does not involve AI…
→ We shifted from technology-first to problem-first
We hit a roadblock with our “where can we apply AI” approach
Began considering where else problems were being experienced in plant maintenance
There were a lot of them!
Here we are going to step out of the facts and into our feelings for a moment
We were feeling frustrated and fatigued by two big pivots and a string of micro-pivots
Many on the team were feeling disillusioned by the path that we were on - we’d moved so far away from AI & the industries were hard work!
We came together and spent three hours talking about our feelings on the matter,
We found that there was little overlap …
…
...
We were each coming back from interviews with different insights and were struggling to reconcile
We eventually realized that there was little overlap between the types of
Many people in maintenance follow a classical paradigm:
“Currently their model for maintenance is: I replace this pump every 6 months because that’s when it breaks on average”
Our insight: in a perfect world, people would do risk-based maintenance:
“ TODO describe”
They have in-house hacks
They are looking for software!
Predict when equipment fails
to reduce plant downtime
using AI
So, from AI to optimize throughput at the outset,
we’ve landed on software for Risk-Based Maintenance!
Mention that we had validation from 5-7 plants
And that 2 people wanted to quit their job to build this software with us
We learnt X…
We learnt X…
We learnt X...
We were keen to apply AI to industries that are largely untouched by innovation
An early interviewee explained how they were using Noodle.AI to adjust inputs to his process to get consistent output quality using AI
This triggered a thought… we could explore AI for process optimization… there’s a lot of underutilized data in this space!
We were faced with skepticism and push back from many interviewees
We were keen to apply AI to industries that are largely untouched by innovation
An early interviewee explained how they were using Noodle.AI to adjust inputs to his process to get consistent output quality using AI
This triggered a thought… we could explore AI for process optimization… there’s a lot of underutilized data in this space!
We were faced with skepticism and push back from many interviewees
We were keen to apply AI to industries that are largely untouched by innovation
An early interviewee explained how they were using Noodle.AI to adjust inputs to his process to get consistent output quality using AI
This triggered a thought… we could explore AI for process optimization… there’s a lot of underutilized data in this space!
We were faced with skepticism and push back from many interviewees