Leading organizations expect to double the number of artificial intelligence (AI) projects in place within the next year, and over 40% of them plan to deploy AI solutions by the end of 2020, according to the Gartner 2020 CIO Agenda Survey (registration).
But most organizations struggle to scale their AI pilots into enterprise-wide production, which limits their ability to realize AI’s potential business value. So what do organizations need to do to ensure successful AI roll-out?
What Should You Ask Before Deploying AI?
Vaclav Vincalek, tech entrepreneur and partner with Future Infinitive said the most serious challenge companies will face — as with previous efforts to implement big data and machine learning initiatives — is at the strategic level. Put simply, they need to start by asking the right questions.
Does the CEO and other non-technical stakeholders understand what outcomes they are hoping to achieve for the business (with or without this new technology)? Do they have a good understanding of how AI might be deployed to help them achieve that objective? Can the CTO explain it in plain English?
Closer to the deployment stage, ask questions like: what is the process for using AI, aligned with other business processes? How will they know when the AI is wrong because it’s using a bad or incomplete set of data? Will they have the expertise in-house, among the CTO and data scientists, to tell when they need to increase or upgrade their data set to provide more meaningful results?
“To ensure a successful roll-out, all of these questions will need to be considered in turn. Otherwise, you’re just adding more technology at great expense, without the ability to really use it,” Vincalek said.
Related Article: Avoiding Potholes on the Road to Effective AI Deployment
Are Your Goals, Data Quality and Processes in Place?
According to Chris Heineken, CEO and co-founder of Atrium, organizations need to ask three key questions to ensure a successful AI roll-out. They are:
Do you know what problem needs to solved?
With traditional IT projects, you seek to address existing pain points, but with AI you are looking for the questions that if you could answer them, it would drive the needle in your business. An AI project will not be successful if you do not know what question you are trying to answer.
How good is my data?
Most companies are hesitant with AI projects because they think their data is not good enough. But really the best way to assess is your data is by using it. Start searching for a signal. Until you do, you will never be able to improve it. AI is a continuous journey of discovery with consistent adjustments.
Am I set up to implement insights into my workflow?
AI can surface a lot of helpful insights for a business, but too many companies get started without knowing how they will utilize the insights they gain. AI models are only helpful if the insights turn into action. Embedding insights into the business workflow is the key to AI success.
Related Article: What Data Will You Feed Your Artificial Intelligence?
'AI Is Broken for Business'
However, AI is broken for business, Arijit Sengupta, founder and CEO of Aible told us. He cited a 2019 MIT-Sloan/BCG survey, which found that 65% of companies report seeing no value from the AI investments they’ve made in recent years. It’s basically a last-mile problem. You can throw all the money you want at AI, but you won’t get any value from it unless it’s adopted company-wide, and that means all the way to the end user.
The business end user needs to see the potential business impact of the AI models and have a say in their creation. You can’t treat all end users the same. One salesperson might want a more conservative AI model because they’re good at closing deals by spending a lot of time focusing on a few key deals. Another salesperson might want a more aggressive AI because they excel at pursuing a lot of deals at once. AI must give the end user the ability to customize the AI so that it matches their individual reality. That way, you ensure that AI gets operationalized and delivers business impact.
Related Article: AI, Bots, Collaboration and Teams: The Dynamics Shaping the Workplace
Questions Around AI Deployments
While there's truth in that, Anthony Macciola, chief innovation officer of ABBYY argued that enterprises could use AI if they introduced three components — or answered three fundamental questions about their deployments.
Do you have a center of excellence?
The first thing companies need to do is establish a Center of Excellence (COE) to oversee their digital transformation process. According to research from IDC, two-thirds of senior global influencers undergoing automation had set up a COE, with the number rising to as high as 78% for leaders in transformation such as banking and investment firms.
“It’s important for the enterprise and early adaptors to know that it is no longer the IT department that is responsible for automation but a collaborative company approach to include specialists like data scientists, the business analysts, the tech guys and, of course, the office ‘citizen developers’ who will be working alongside new digital workers,” he said.
They all need to be part of the COE and the automation of everything within the company from simple invoice processing to AI-led process intelligence technologies to improve customer engagement. Too many projects have failed because leaders have taken a siloed approach and not had a proper roadmap to expand across the entire organization.
What processes are you targeting?
It’s important to decide which processes within the organization you wish to automate as the right processes to start with are not always selected. That’s because many companies are compartmentalized in organizational process knowledge. Additionally, top management is not involved in the day-to-day workflow and lack process documentation, making it increasingly difficult to truly discover what processes are ready.
Many organizations think they know how their processes work, when in fact, it is far from the truth. Incorporating process intelligence before a project will give you a comprehensive understanding of where to apply RPA and AI solutions by seeing your workflow in an “as is” state. It will determine the expected value and savings to the organization based on data, not opinion or bias. A few qualifications make a process a good candidate for automation. These include:
- The process follows rules-based logic, rather than human judgment-based decisions.
- The process is repetitive and may be prone to human error.
- The process follows a clear set of instructions.
- If there is input data, it is digitized, or can be with methods such as OCR.
Have you established KPIs for processes?
Finally, organizations must not take a “one and done” approach after deployment. Having clear KPIs is always an ongoing process. It is critical to enable continuous improvement by monitoring and measuring your automation's up and downstream impact to ensure ongoing protocol compliance and prevent the ‘bottleneck shift’ and the possibility of negatively impacting other processes in place.
Monitoring the digital workforce as well as the entire end-to-end process post-implementation is just as essential as the initial planning and execution. All too often, AI projects die because of political backlash due to lack of process and planning, Rajeev Dutt, founder and CEO of Dimensional Mechanics said. To ensure a successful AI rollout, businesses need to understand that AI should never exist in isolation, rather it needs to be integrated into broader business processes.
A final point Dutt shared: ensure you aren’t building a centralized AI team, as they will likely be too disconnected from the business to be successful. For a satisfactory AI rollout, you need to construct a team that connects the business process with a business architect, machine learning engineer, and an enterprise SE.