Page 28 - 10 Most Innovative AR VR Startups In 2019 up

FIVE HYPOTHESIS AS TO WHY ARTIFICIAL

INTELLIGENCE AND MACHINE LEARNING

PROJECTS FAIL

Hypothesis #1: Data Science Hypothesis #2: Data science/ less than optimal. Lack of a less
initial models don’t scale or ML models while brilliant complex model to fall back upon if
are too experimental to be and innovative don’t meet the data changes can also present
use d by internal or external the business requirements or a challenge. On the upside, the
customers. are too fragile to respond to research can be published in
change in the supporting data. academic journals read by other
Projects often start here as data scientists to further overall
companies hire on a few data At the recommendation of industry knowledge.
scientists who build their models consulting groups, some
in Python or R, only to discover companies make the decision In response to this non-adoption
quickly that there is a difference that in order to foster innovation, problem, some innovation/data
in mindset between data scientists “innovation teams” need to be science teams may add product
and engineers. In the short run this isolated from the “non-digital” marketers to their organization to
problem is often solved by having cultures which surround them. But “promote” their work internally
machine learning engineers or while isolated innovation teams and to try to market their concepts
other software engineers take the can lead to great opportunities to customers directly.
code, rewrite it and follow standard for experimentation and those Hypothesis #3: AI initiatives
dev-ops processes in order to teams can learn and develop are driven by the company’s
scale and deploy the application. interesting solutions, when the
Given the stochastic nature of resultant projects then need to be internal IT organizations and
results, Quality Engineering also “pushed” to the market or internal inherit “waterfall” challenges.
needs to scale to the task. During customers, there is often minimal As part of a “digital
this transition period, business adoption. While the solution may transformation”, some companies
users (and/or their proxy product meet the specific challenge, the see AI as part of an overall
managers) can be left out of the experience for the user is often
process and the requirements or
underlying data may change.

Organizationally, some companies
take the next step of hiring
engineers into the data science
group to help the data scientists
learn more about scaling for
production and deployments.
The goal with this approach is to
alleviate the handoff process. The
challenges are that this research/
engineering organization can
become siloed away from the
rest of the production support
workflow. “Data Scientists don’t
wear pagers on call”.

28 December 2019