Insight of the Week: AI is not a strategy

By Kerry Robinson

A senior executive recently shared his frustration with the long line of AI vendors vying for his business:

"They keep telling me I need their AI... then they ask me what I want to do with it!"

He lamented.

"It reminds me of that Henry Ford quote... if you ask me what I want, I'm just going to ask for a faster horse!"

And then he said something that I think really illuminates the struggle that many senior execs are experiencing right now:

"AI is not a strategy. It's a capability that supports a strategy. So what's our strategy?"

Good question. So what's your strategy? Or rather, what was your strategy? And how should it change?

I've argued before that AI is a general-purpose technology like steam power, electricity, computers, the internet, and mobile. It will birth new companies - even whole industries - and kill others.?

If you believe that, then your strategy almost certainly needs to change. Let's start by considering how AI will impact your Customer Service strategy:

Customer service used to require a lot of human agents answering calls and chats from a bunch of more-or-less frustrated customers. But AI changes that. It has the potential to handle the majority of those calls and chats.

Businesses used to set expectations of superior customer experience by guaranteeing a human would answer the phone. Or that an on-shore agent would. But AI changes that. AI has the potential to deliver a better customer experience. Because it can constantly improve and doesn't walk out the door for a better job just as soon as it starts getting good.

Customers used to have to call or chat with your customer services function if something went wrong. But AI changes that. AI has the potential to call or chat with you on behalf of your customers. Which means you might end up getting a lot more calls from AI, than from humans!

So as the senior exec I quoted above said: AI is a capability that supports a strategy. But it's also a general-purpose technology that changes the game.

If you carry on doing what you were doing, it's almost certainly a bad strategy.

If you try to come up with a new strategy, based on what you think AI can do, it's almost certainly going to be a bad strategy. Because we still don't fully understand what today's AI models can do, let alone what GPT5 might be capable of.

So you need a new strategy. But you don't have enough information on which to build one.

I think this leads us to only one viable AI strategy: exploration.

You don't need to know everything that today's AI tech is capable of. You need to understand what it can do for your business. And your customers.

You don't need a long-term strategy, nor do you need to understand how AI will impact every part of your business. But you need to find some quick wins that will begin to inform your strategy.

And the contact center is the ideal playground to explore the capabilities of AI. Because it's full of conversations - something that Generative AI is extremely good at. There's lots of repetition - so we can test and learn. And we have established quality assurance and escalation processes - to pick up the pieces if the AI gets stuck, or strays out of process, or compliance.

So what can we do with AI in the contact center? How should we get started?

I think the best way is to begin with automation of a contact type that your junior agents handle, but is beyond what simple scripted automation (think IVR, simple chatbots) are capable of. Build a gen AI-powered chatbot (easy) or voicebot (harder, but more valuable). Test it with internal users, or do a small pilot to see how well it performs.

Scheduling is a good use case to consider. Or explaining a bill. Negotiating a payment arrangement is another one I'm seeing a lot. Status of an order, delivery or a claim is also worth a look - it seems like a simple use case to start with, but quickly leads to questions (why's it late?) and requests (I want to cancel!) and explanations (it's already left the depot so it can't be canceled, but we could arrange a return?). Also, make sure you find a use case that means you can reasonably separate out that contact type for treatment with Generative AI. For that reason outbound use cases are attractive. As are contact types that have a dedicated phone number, or that you can reliably route based on an existing IVR menu or Natural Language Understanding (NLU) routing system.

Build and test a prototype. Let your team and agents interact with it. Maybe do a small-scale usability test where you get a handful of actual customers to interact with your prototype and provide feedback.

You're going to find problems, many of which you can fix with better prompting (the instructions you put into the AI along with the response from your customer) or tools (small pieces of software that let the AI do things like query a knowledge base, CRM or API). Some problems might lead you to try a different AI model that's better, faster, cheaper, or maybe specifically fine-tuned with your data to perform better on your use cases. Once you've proven out a use case, and you're comfortable with the impact on costs, customer experience, revenue, and operations, it's time to walk the path from prototype to production. I covered the whole end-to-end process here.

Getting from prototype to production is a big lift. Don't underestimate it. But this is where you'll learn even more: how susceptible is your solution to bad actors? How often does it make mistakes? Can your Infosec team get comfortable with the security posture of the AI platform you're using? Can legal get comfortable with the contractual implications for your customers? Can leadership get comfortable with the risks of doing what you're exploring. And equally as importantly: the risks of not doing it. A regulatory compliance: "No" is easy, but if it means you're one of the companies that die in the face of this general-purpose technology, that's a rather risky "No"!

All of this will inform your strategy too. Maybe you can get away with a simple SaaS solution but you might need to roll your own fine-tuned Open Source models?in your own private cloud environment to deliver on production performance, speed, cost, or security requirements. Maybe you'll need to change your brand promise - or contracts - to allow the use of AI in customer service or other scenarios.

So getting back to the quote that got me started writing this email:

"AI is not a strategy. It's a capability that supports a strategy. So what's our strategy?"

The real issue here is that the 'capability' of AI is not well understood. We keep discovering new capabilities of old models. And new models demonstrate new capabilities. The same models given access to different tools, and data, can achieve wildly different results.

So your strategy should be exploration. Explore the capabilities of AI in your organization. With your customers. Your contracts. Your brand promise. Your unique contact types, and volumes. Your unique capabilities. And those of your partners.

Kerry Robinson is an Oxford physicist with a Master's in Artificial Intelligence. Kerry is a technologist, scientist, and lover of data with over 20 years of experience in conversational AI. He combines business, customer experience, and technical expertise to deliver IVR, voice, and chatbot strategy and keep Waterfield Tech buzzing.

Subscribe to Kerry's Weekly AI Insights