AI in business: five tips for successful implementation

By David Ingham
While AI applications have become more mainstream, businesses must understand and apply these solutions in a smart, ethical, and economically friendly w...

Artificial Intelligence (AI) has long since outgrown its vague sci-fi connotations to become a central pillar of business operations. While AI applications have become more mainstream, businesses must understand and apply these solutions in a smart, ethical, and economically friendly way. 

No two areas of AI application are exactly the same; each must be assessed in its own right. However, here are five shared guidelines all decision makers should keep in mind when considering implementing AI into any business operation: 

1. Target quick-value delivery 

In the balance between return on investment (ROI) and speed, speed is critical. When selecting the first business challenge to solve with AI, IT decision makers often make the mistake of only choosing the challenge that will create the biggest ROI. ROI must of course play a role in decision-making – no solution should cost more than the value it generates – however, this will not necessarily yield the biggest or best result for the business as a whole in the longer term. Quick wins, while often deemed smaller tasks, will usually demonstrate value more rapidly and help with broader buy-in. One example of this is implementing AI for large data analysis.   

2. Connect with everyone 

Buy-in is crucial. It must come not only from quickly being able to demonstrate value, but also from bringing people on board and keeping them engaged and connected throughout the whole AI journey. This includes everyone right up through senior leadership. Generating agency with decision makers is a lot easier if everyone understands the goals and requirements from the start and if they receive regular updates on the challenges and performance metrics throughout the project lifecycle.  

This feeds into funding conversations too, which can often be tricky. Even when early results are positive, it’s essential to manage expectations of varying business leads. Don’t be held to unrealistic standards. For example, achieving 100% automation or accuracy immediately, while great in theory, is not an appropriate benchmark. Having a greater understanding and accommodation of what everyone wants to achieve will make it easier when setbacks arise. 

3. Use data properly 

AI is only as good as the data that is fed into its engine. Disorganisation is one of the biggest hindrances to effective AI initiatives. Data must be both high quality and trustworthy, and it needs to be fed into the AI system in real time to produce effective results. Old data only drives insights that no longer have value. Most importantly, the data should be prepared, categorised, and classified to enable easy analysis.  

4. Don’t be scared to fail 

The phrase “right first time” rarely applies to AI implementations, especially for those making predictions and forecasts. Refining to an acceptable level of accuracy will take time and will often follow a series of failures in order to determine what to further correct and refine. 

Starting with a smaller problem or a subsection of a large problem can help reduce the risk associated with the cost of failure. There is also no shame in dropping an idea and rethinking the approach – in fact, that willingness to rethink is vital.  

Persisting with a flawed system – and, by so doing, wasting time and money – is never the right way to go. Instead, changing course or even dropping the idea altogether helps accelerate progress. Once a smaller problem is resolved and the business can see its value and ROI, the solution can be scaled to solve a bigger problem. 

5. Measure outcomes, not outputs 

AI projects are inherently different from IT projects, which tend to proceed with a clear idea and a set target for the desired output from day one. By contrast, we mostly use AI to try to understand the unknown, thus we really can’t know what the output will be ahead of time. As a result, AI needs to be tuned, monitored, and modified continuously over time.  

Successful AI implementations may therefore require more iterations than originally planned, and the outcome might not deliver the same level of accuracy or automation as initially imagined. Determine success by the degree of impact it creates and how much value that yields.  

Implement AI with cause 

Organisations should not implement AI without purpose. However, because the technology offers so many possibilities, prioritizing them can be overwhelming.  

It’s important to take a collaborative approach and focus on what is required and doable. Above all, be fearless. This is much easier when working alongside partners with much more experience at implementing new technologies like AI. AI is a learning technology, and its learning process always comes back to the people instigating the programmes as much as the AI itself. It takes courage and patience to appreciate these lessons. 


Featured Articles

Jitterbit CEO: Confronting the Challenges of Business AI

AI Magazine speaks with the President & CEO and Jitterbit, Bill Conner, about the growing AI hype and how it can be integrated into a business successfully

Graphcore: Who is the Nvidia Challenger SoftBank Acquired?

SoftBank's acquisition of the UK startup Graphcore could accelerate development of the more efficient IPU AI chips and challenge chip giant Nvidia

Amazon Takes On AI Hallucinations Across Its AI Portfolio

Amazon is upgrading the memory capacity across a range of its services to improve the accuracy of responses Gen AI systems return to prompts

LG’s Athom Acquisition to Accelerate AI-Enabled Smart Homes

AI Applications

Why AI is Behind Samsung’s Expected 15-Fold Profit Surge

AI Strategy

AI Patent Race: What China’s Dominance Means for the Market