AI Best Practices
There is no one way to implement artificial intelligence solutions. But there are AI best practices that apply repeatable methods to minimize investments, save time, reduce risk and improve results. Here are some of those lessons.
Start with Strategic Objectives
I've been an AI consultant for about two decades. My experience has been that many companies approach artificial intelligence with curiosity more so than strategy.
They adopt piecemeal pilots designed to test technology rather than the technology's ability to achieve significant business results.
Most of these pilots tend to adopt the easiest use cases that deliver the least payback.
It's not that this approach cannot work. It's that it creates significant delay and tends to miss the most significant technology-enabled business opportunities.
There's a better way.
Start with your company's strategic priorities. Then architect AI technologies to achieve and accelerate those goals.
AI is transformational. So, use it to transform business growth strategies, market expansion opportunities or product innovation. Use it to improve staff performance, customer engagement and company differentiation. In short, bypass incremental advancements and use it to achieve order of magnitude improvements or competitive advantages.
Otherwise, you will end up spending way too much money for too little return.
Design Use Cases that Deliver Business Results
You need to define the business use cases to ensure your technology choices achieve your planned results. Otherwise, you risk having to make unplanned purchases to get what you originally wanted.
Use cases include techniques to increase customer acquisitions, grow customer lifetime value or improve customer retention. Or from a cost takeout perspective, they may include methods to improve employee productivity or automate business processes.
It's helpful to start with use cases that make staff more successful in their roles. Then seek cases that help managers make better decisions or business leaders forecast the future.
Some simple business development use cases are shown below.
However, to really maximize your payback, your use cases should be industry specific. An example for the professional services industry is shown below.
Weight your use cases by time to value, complexity and payback so that they can be easily prioritized and sequenced.
And one more thing. If you want to achieve business transformation, focus on the scenarios that most directly impact customer interactions. That means prioritizing the use cases that impact how customer facing staff work. If you improve how work gets done by the employees that sell, deliver and support your customers, your upstream objectives will be realized.
This bottom to top approach will also reduce change management obstacles and facilitate scale.
Create an AI Tech Stack
Executives seeking technology enabled business improvements know that Artificial Intelligence can help. What they may not know is which technologies and how they work together. An AI Tech Stack answers these questions.
Organizing artificial intelligence technologies pursuant to this approach creates several benefits.
For example, tech stacks identify fewer technologies to achieve the most business objectives in the least time. This lowers investment, simplifies IT management and future proofs technology decisions.
Without this approach, companies tend to acquire standalone or departmental systems to solve urgent but isolated challenges. That may help with the problem of the day, but quite often contributes to software sprawl, data siloes, lack of integration and temporary results.
Use an Implementation Methodology
Just about any AI consultant will advise an implementation methodology similar to the below five steps.
However, it's not those steps that will challenge your program. It's the culture and people that must be proactively managed to succeed. Integrating new technology into the company should be accompanied with the following AI Best Practices.
- Agile execution. Agile methods bring together 4 critical success factors. They include collaboration among cross-functional teams, close user involvement, iterative and adaptive deployment processes, and the frequent delivery and inspection of incremental releases. There are many types of agile methods. Use the one you have, and if you don't have one, consider starting with Agile Scrum.
- Risk mitigation. AI comes with some unique risk factors. They include things like data privacy, bias, hallucinations, data toxicity, ethical guidelines, and acceptable use policies. There are also data protection laws such as GDPR and the California Consumer Privacy Act. A risk management plan is needed to avoid problems and ensure systems are transparent, explainable, and accountable.
- Change management. AI brings new processes, automation, information, roles, responsibilities and control. It may also bring an actual or perceived loss of control. That's a lot of change. And the problem with change is that it causes anxiety for many staff. The change is endorsed by the few imposing it. However, it is not always so well accepted or is even contested by the majority receiving it.
To facilitate that difference, apply a change management program to systemically shift individuals, teams, and the company from a current state to a defined future state. This will mitigate productivity loss during the transition, create an environment for sustained change, and realize the benefits of change more quickly. It will also ensure resistance to change will not delay or derail your objectives.
- Program Management. This is needed to achieve forecasted objectives – on time, on budget, and on value. You will be best served with a framework that provides the governance, oversight and controls which monitor, measure and report on the effort's most important performance objectives. Your framework should also measure the underlying critical success factors which at the minimum will include scope, time, cost and quality. Without strong program management, it's easy to confuse activity with progress and difficult to separate the urgent from the important.
- Governance. Finally, you need executive oversight. Governance is really about 4 things, being transparency, inspection, adaptation and accountability.
Good governance must be proactive and forward looking. It must empower the steering committee to view the effort through the front windshield, and not just the rearview mirror. This then permits the committee to steer the effort to a planned destination, and not just be along for the ride.
Governance should include clear C-suite sponsorship, cross departmental leadership and multi-disciplinary teams. When you have this makeup, it sends a clear signal that the effort is strategic and holistic. This then in turn drives accelerated culture adoption and behavioral changes.
Measure What Matters
No AI Best Practices article would be complete without success measures.
But to know whether your solution is working, you have to ask the staff if it is making them more successful.
Our AI agency meets many new clients. In speaking with staff for the first time, we hear a story of two tales.
Participants who are underwhelmed with their company's AI rollouts advise they see the technology as trivial. It doesn't show them anything they don't already know. It doesn't perform tasks for them or save them time. We hear a repeated theme that the effort was a technology project hoping for business outcomes. That seldom works as the business outcomes need to be the starting point.
Participants that found AI effective advised that the technology shifts customer engagement from reactive to proactive. It elevates data from a byproduct to an information asset they actually use. It automates processes that save them time. Most advise they couldn't do their jobs as well without it. Some call it a game changer.
The only way to know if your program is working is to measure what really matters. The measures of success are things like improvements to staff productivity, customer satisfaction and company growth. If you are not measuring factors like this, your program is likely a technology-focused effort operating in a bubble that will inevitably burst.