How to define your AI roadmap

Many organisations have woken up to the potential of AI over the last couple of years. Every second CEO wants their own version of ChatGPT. With this surge in interest, teams across the world are rushing to deploy AI solutions on a whim, without having a long-term strategy, rushing to claim they’re using generative AI. This article will point out the risks of a knee jerk reaction to AI deployments and show you how to properly define your AI roadmap in a way that’ll set you up for short-term and long-term success.

Why knee-jerk AI deployments fail

Firstly, let’s talk about what can (and does) go wrong when companies deploy AI on a whim, without a proper plan.

1. Pilots don’t go to production. The first risk is that you spend untold time and effort on your pilot, only to shelf it before going live. This is usually because the wrong use case is selected and you bite off more than you can chew. Selecting something too broad or too complex to deliver first means that you don’t have the opportunity to learn the ropes first.
2. Wrong technology selection. When you’re thinking purely about deploying generative AI at all costs, you’ll wind up building a specific solution for the one use case you have in mind to start with. Even if your first deployment goes well, your next use case is highly likely to have different requirements, as will the rest of your roadmap. This means you’ll eventually need to take a step back and consider your holistic technology strategy, which often means the solution you just built needs to go to or be heavily altered to cater for what’s next. You’re then at risk of boxing yourself into a corner with a complex technical deployment that doesn’t scale and is a nightmare to manage, update or extend.
3. Horsepower overkill. Generative AI isn’t always required to deliver what you’re aiming to achieve. Sometimes it’s like bringing a gun to a knife fight. And rarely is generative AI the ideal end-to-end solution. This means that teams shoe horn or even mandate generative AI for sake of it, rather than because it’s the best solution. This introduces needless risks and costs into your project that could be avoided.

Everest Group also found that unclear success metrics and cost concerns are also common reasons for generative AI project failures.

How to avoid AI failures?

So how do you avoid these failures? How do you make sure that you not only find short term success, but that you also plan for how to scale and develop your roadmap, maturity and value?

This all starts from developing a proper understanding of the use cases that your users and business need to tackle. Using those use cases to guide what you do and how is the key. And to get a handle on those use cases, you should conduct a contact channel analysis.

Conducting a contact channel analysis

A contact channel analysis should be the first place to start with your AI strategy. In a nutshell, this is where you analyse the conversations that you’re having with your customers across all of your touch points to determine what use cases would bring you the most value, in what order and what your approach to resolving those issues should be.

To conduct this analysis, follow these steps:

1. Data gathering. Gather all of your conversational data from across all of your contact channels; phone calls, emails, tickets, social media messages, SMS, and any other inbound or outbound channel where you’re having conversations with customers.
2. Topic modelling. Conduct a topic modelling analysis to determine what conversations are happening with what frequency, on aggregate and by channel. There are plenty of tools you can use to do this these days. We have our own tools at VUX, but you can also use commercial tools such as DashBot or HumanFirst, platforms such as AWS Lex, Genesys, Kore.ai and many others are beginning to incorporate this stuff into their product suites.
3. Score and rank. From here, you can begin to score each use case according to the criteria that matters to you. We recommend building a matrix that includes factors such as:

• Business impact: what impact will the use case have on the business from a revenue, cost or efficiency perspective?
• Contact volume: how much of a need is this for customers? Is it a high or low volume issue?
• Conversational complexity: how challenging or diverse is the conversation itself? Is it a short, few-turn issue or a longer, more complex need?
• Technical complexity: does it require system integrations? How many? Do you have APIs available?
• Business support and alignment: how supportive are other key stakeholders in other business units that this use case affects? And how aligned is this use case to your wider strategic goals?
• Effort to build: how much time and effort would each use case take to build and deploy? Do you have the required skills and resources?
• Self-service availability: do you already have a self-service journey for this use case on another channel? If so, is it effective?
• Existing precedent for use: Is there a working example elsewhere in your industry or another industry where this use case has already been automated by AI?

Once you understand this picture, you can start to determine the short, medium and long term use cases to build into your roadmap, as well as your approach to solve them.

Selecting your short-term roadmap

Ideally, if you’re starting out on your AI journey, you want to find use cases that are low conversational complexity, low technical complexity, high business support and alignment, low effort to build, low current self service availability and high existing precedent for use.

Contact volume and business impact are both measures that will play a factor depending on your current level of maturity. For example, if you have some experience in AI, and resources to boot, then you might aim to tackle higher volume, higher value use cases. If you do this, though, you should still aim for a slow, phased rollout. Your proof of concept should be released to a small number of users, rather than everyone initially. This will give you the opportunity to learn and iterate your solution and prove the potential value it can provide.

If you do prioritise volume and business impact in the short-term, however, make sure you still prioritise low complexity, low effort and high business alignment otherwise you’ll be in development for a long while and your time to value will increase substantially.

For those starting out, with low levels of maturity, prioritising high volume and high business impact use cases represents a risk. Instead, search for those use cases that will be able to demonstrate sufficient business value that it validates the approach, technology and strategy, but that stops short of putting the solution in front of every customer.

Taking your short term roadmap to proof of value

The next stage is to define your approach, bring together your team and find the right technology, before refining your long term roadmap. We’ll cover each of these in a following article. Subscribe to VUX World to be the first to read it.

About Kane Simms

Kane Simms is the front door to the world of AI-powered customer experience, helping business leaders and teams understand why AI technologies are revolutionising the way businesses operate.

He's a Harvard Business Review-published thought-leader, a LinkedIn 'Top Voice' for both Artificial Intelligence and Customer Experience, who helps executives formulate the future of customer experience ad business automation strategies.

His consultancy, VUX World , helps businesses formulate business improvement strategies, through designing, building and implementing revolutionary products and services built on AI technologies.

1. Subscribe to VUX WORLD newsletter
2. Listen to the VUX World podcast on Apple , Spotify or wherever you get your podcasts
3. Take our free conversational AI maturity assessment