Published January 18, 2020 co-authored with Alex Bratton https://medium.com/@alexbratton/why-your-ai-project-is-going-to-fail-f03972f5b9f9
It seems like whenever we talk with business executives about technology these days, the first thing on the agenda is AI. While enthusiasm for AI abounds, we are also seeing considerable apprehension and uncertainty about how it is used, the readiness of the organization to embrace it, and the jobs it will impact.
We see companies trying to make an impossible leap from manual data work and traditional batch data warehousing to full machine learning in one step. Unfortunately, these scenarios typically end in early failure, setting back the entire, well-intended initiative. The good news, though, is that needless struggles can be avoided if you follow our playbook.
This is the start of a new series by Alex Bratton and Kevin Glynn on a pragmatic approach to data and AI success.
How AI Goes Wrong
Truth is, AI might not be the right place to start for your employees. One senior leader we know jumped into an AI pilot back in late 2017. He found a reliable small AI company to partner with. He personally wrote the use case and sat in on many of the weekly project meetings. After two months, the vendor told him the project was at a standstill — the data was too dirty and his use case was a poor choice for a pilot project. Even worse, his executive team became suspicious of AI and didn’t approve any new data initiatives for over a year.
What early project failure looks like:
1. Gartner forecasts that there could be up to an 85% failure rate in early AI projects. This includes projects that were successful, but had a lower ROI than expected or resulted in huge change management problems.
2. An 2019 IDC survey confirms that companies jump in too quickly, with only 25% having an enterprise AI strategy. McKinsey, too, cautions for poorly constructed use cases.
Getting AI Right Starts With Data
Many companies jump into AI, ignoring the basics … starting with how their data is organized.
For example, a company we know had customers asking for more complete status information. The IT team worked diligently for a year to provide the necessary data sets to meet the needs — very basic, simple work. Yet, their customers were delighted by the quality insights they were getting. This work added huge value; and was the foundation necessary for future AI initiatives.
The benefits of organizing data sets to allow answering questions that span across multiple systems and sets of information are large.
Unifying Data From Multiple Sources
Does any of this sound like your team?
Employees manually merge multiple reports and spreadsheets to get to the answers they need. Teams create their own trackers to try to answer questions. Lots of spreadsheets and PowerPoints are emailed around to communicate status up and down the chain. Significant time is spent in arguments over how the data was collected or calculated rather than the story it tells. Everyday employees are frustrated as they deal with multiple, inconsistent sources of truth.
In order for employees to quickly get the answers they need, data from multiple systems typically needs to be combined and simplified by stripping away the extra data. The ability to pull data from multiple sources and merge it together implies a healthy integration layer that makes the data available (which is typically a gap in enterprise). Another challenge that arises when combining data is the use of operational (typically more real time) and reporting (usually weekly or monthly) data in the same conversation. This requires business users to think differently about the information at hand as it typically is not used in the same conversation, let alone on the same screen.
Once this effort is completed, employees will be able to answer questions by quickly accessing multiple data sources in a single query rather than suffering through multi-report/multi-system user workflows to get the information they need.
While eager to use AI, most groups fail to recognize that there are many benefits to enjoy along the road to AI. Simply creating easy to use applications that merge data and allow employees to see a bigger picture can bring huge benefits. Plus, the ability to merge and correlate data sources is a key foundation to successful AI.
“If You Don’t Have Good Data, AI is Over”
The ROI of showing merged data to employees can be quite large — maybe even bigger than your early AI use cases. Plus, showing broader data to employees requires significantly less change management than installing AI to automate processes.
We need to apply the same thinking to all of our vast amounts of data. How can we organize it and simplify it to make it easier for our teams to find what they need? This is where the employee experience becomes critical — a focus on what the people using the data need, not with the data architects who set up the data repository need.
Using the Employee Experience as a Lens
Combining systems creates even more data to bury our users with. How can we simplify and enable them to find what they really need when they need it?
To get that value we have to understand what our employees are trying to do and the questions they are trying to answer. The data by itself doesn’t have intrinsic value. The real organizational value comes from what the data enables our employees to do.
Most organizations dive into their oceans of data and think about the possibilities of all the things that could be solved with that information. It is typically approached from an organizational or operations team lens rather than from the perspective of your end users. Your employees are the users of this data and your focus must be on what they are trying to accomplish, where they are when they’re doing it, what value we can provide to them and what success looks like.
“Defining our data experience is no different than defining the user experience for our applications.“
Using the employee experience as a filter to identify the subset of information that needs to be exposed (or fixed to provide a single source of truth) simplifies the effort. Start by identifying what answers employees need and what data is required to support them. This effort enables us to differentiate between what is needed for automated reporting tools that directly provide answers versus analysis tools that allow discovery inside the data. Plus, it simplifies the work required to surface the right data. This cuts out 99% of what we think we have to create and what the IT and operations teams have to struggle with, and keeps our efforts focused on what really matters.
An organization we know was executing a large data migration and systems upgrade. At the same time, a new initiative was spinning up that would require data from the new system. Rather than waiting 18 months for the data systems updates to complete, the team was able to identify the employee experience data needs for the new initiative and reprioritize the data work to make those elements available from the start. This provided much faster value to the employees.
“Glance-able answers simplify processes.”
This simplified view can provide glance-able answers which will start to simplify processes by cutting down on the number of reports to run or spreadsheets to create. We have seen this frequently deliver time back to employees by taking low value work out of their way and helping them surface direct answers much more quickly.
Most Organizations Are Not Ready for AI
Just like the best ERP implementations or mobile app developments, a successful AI solution requires collaboration from teams across the organization: IT, project management, finance, sales, operations and so on. Moving your organization from a line hierarchy to collaborative teams is hard. Your team will likely struggle with the broader cross-functional decision making, mindset shifts and new skillsets required to for success.
One good example was a recent data planning meeting where the senior leader stood at the white board and wrote a probability function. The four engineers in the room all looked uncomfortable until thankfully the youngest said “I just graduated college two years ago and you have just hit the limit of what I remember from statistics. I bet everyone else is nervous about this.” The senior engineer in the room laughed and said he was correct. It took an active acknowledgement that they were in new territory to help the team move forward.
In other words, skillsets need to change or be refreshed. Data science and AI are skills that can be developed. You can hire for it, but without focused attention on growing skills internally, you can’t simply assume a successful jump into AI. The good news is your focus on the kinds of basic data work described in this article allows you to grow your team’s capabilities while delivering results to your employees.
Yes, There is Hope
We’re not saying AI isn’t for you — but there are critical steps you need to take before jumping in. While most organizations today struggle with the issue of data maturity and how to evolve their use of data, it’s a straightforward effort to continually add to your team’s capabilities and deliver more value to employees.
To support your journey to AI success, we will soon be sharing our Data to AI Playbook. Based on our work with dozens of enterprise organizations, the Data to AI Playbook will help organizations understand where they are in their AI transition compared to the market and how to move forward. Is your organization stuck at manual data processing? Are you suffering from death by dashboards? Is your data telling a story? Is intelligence emerging in the organization? Have you transformed how you deliver value to customers?
That’s where we’ll be headed next in our discussion of the data maturity journey. Stay tuned for our next article in this new series on the pragmatic approach to data and AI success.
Alex Bratton & Kevin Glynn
About the Authors:
Alex BrattonAs the CEO and Chief Geek of Lextech, Alex leads a team that creates employee experiences to empower people and achieve business results. He is passionate about refining the future of work, helping organizations transition from a set workplace to work-any-place. Alex is the author of Billion Dollar Apps and an adjunct professor of computer science at Northern Illinois University.
Kevin GlynnKevin is a 3x CIO and led early work in AI, bringing new tools and processes to data teams. Kevin was recognized as Chicago CIO Innovator of the year in 2019. He chairs the board of advisors for Northwestern University’s MS IT program.