"I see eyes in my soup": How Delivery Hero implemented the safety system for ...
ML4ALL Conference promoting ML
1. Breaking the barriers to ML in your organization
Rich Jolly
VP, Data Science
Hawes Group
May 29, 2018
@rich_jolly
2. Key Messages
• Proactively address
management’s concerns
• Use value stream mapping to
find key leverage points
• Focus – use 80/20 rule
• Create a virtuous cycle –
build on your wins
Luddites
@rich_jolly
3. But we don’t have the software or compute power to do that!
@rich_jolly
4. Recent advances transform Data Science
Compute availability
Rich, open source software
Powerful, open source
machine learning libraries
The capabilities that a few years ago only the big players could
achieve, are now within the reach of most companies!
@rich_jolly
5. BI, AI, ML, Analytics, Big Data –
I don’t even know how to think about it!
@rich_jolly
8. Value stream mapping
• From lean management
• Series of events - taking
a product or service
from its raw materials
through to the customer
• Material and
information flow
• Focus on areas that add
value
• To identify and remove
waste (muda)
• Focus on points of high
leverage
@rich_jolly
Cf. https://en.wikipedia.org/wiki/Value_stream_mapping
9. Example: VSM & ML in an Agile Process
• Collaboration – NLP checking team code for synergy
(duplication, expertise, etc.)
• Documentation –Automatically creating artifacts or
real time tracking of progress
• Automatic generation of simple code tasks
Source:
https://www.linkedin.com/pulse/making-machine-learning-part-your-future-agile-team-ashish-gupta/
http://www.ryantomlinson.com/6-engineering-organisation-anti-patterns/
https://www.slideshare.net/RichardJollyPhD/rich-jolly-executive-vp-corporate-ebitda-from-data-science
@rich_jolly
10. Pareto Rules!
• A great deal of machine
learning benefit can be
realized with reasonable
effort!
– Don’t forget regression
• Rank opportunities
– Pick low hanging fruit
• Monitor progress
– And results
@rich_jolly
12. Common pitfalls in data analysis
• Looking for silver bullets
• Searching for keys under the lamp post
• Conflation of effects and variables
• Bring me a rock
• Lack of domain expertise
For further reading:
- ‘Minding the Analytics Gap,’ Ransobtham, Kiron and Kirk Prentics, MIT Sloan Mgmt Review, Spring 2015
@rich_jolly
13. Even if we have some success, how can we sustain it?
@rich_jolly
14. Build a virtuous cycle
Successful
Analytics
Projects
More
organizational
confidence
More
resources
Building analytic capabilities is an evolutionary process
@rich_jolly
What are some of the areas to watch out for in data analysis projects?
First, don’t expect the team to deliver silver bullets. Usually, analytics projects deliver slow, steady, but measurable benefits.
Second, avoid the ‘Searching for keys under the lamp post’ effect. If you’ve never heard the joke, let me tell it. ‘Tell joke’
Sometimes the data team needs to move outside the comfort of the most well traveled databases to find what is sought.
Next, be careful not to conflate effects with variables. For example, your goal with a project may be to improve patient experience as measured by their response in a post appointment survey. You make it a goal to reduce their wait time to achieve this goal. However, you may find that once you have reduced the wait time, the survey response on patient experience has decreased. The ‘effect’ here, the patient experience, is a complex function of many variables, one of which is wait time. By reducing wait time, you may have impacted one of the other variables. Remember to monitor your key dependent variable.
Have you heard the parable of ‘bring me a rock?’ ‘Tell it’
Make the most efficient use of your data science team by being as explicit as you can as you outline your questions.
Another common pitfall is lack of domain experience. If you are hiring a consultant be sure to ask about their expertise in your area.