The Hidden Cost of Sticking with the Wrong AI
29 Apr 2025
6 min read
The sunk cost fallacy is one of the most famous phenomena of its kind. It’s the idea that after you’ve spent time and money on something, you are understandably reluctant to abandon it because of all the cost you’ve sunk into the project – even when the alternatives are clearly better.
Of course, if you’ve invested in a solution that simply doesn’t work or is falling well short of expectations then stepping away is more straightforward. But what if results are adequate, but could be better? Then stepping away feels much tougher. Let’s look at how this plays out in the world of AI for CX.
How the sunk-cost fallacy develops
The most likely scenario here is that you have partnered with a technology platform that gives you an AI for CX solution.
After you onboard, you spend weeks and months training the AI to work with your brand and follow your processes. You are making constant tweaks and building new prompts, and step by step you are starting to improve your automation rate to around 15-20%.
But then around the time that your contract renewal rolls around, you come across an alternative solution that looks like it should be better. So do you rip up what you’ve done and start again?
Why it can be hard to give up on an existing solution
Emotionally, it feels very painful to consider giving up on a solution after devoting that much time to it. That’s natural – it’s better the devil you know.
Plus it can be tough because it might appear like you made a mistake with your initial selection. However, you can feel reassured because this is such a new area of technology that there aren’t safe bets, and even companies that look like safe bets may not be right for your purposes.
It appears difficult because there will be a transition period if you do switch. During this period it seems that you either have to pay for both solutions until the new one is up to standard, or accept a decrease in automation.
This may not be as bad as it appears, as your new provider may be able to make strides faster, and you can learn from all the work you did already to build out a new solution that matches your old one in a matter of weeks, rather than months.
Getting past the sunk cost
The good news if you find yourself in this situation is that you have proved the concept with AI. You’ve already jumped the fence into the world of AI, and now it’s about creating the best situation.
So now you have to forget the past and look to the future, and the key question to ask yourself is:
“If we hadn’t picked our current supplier yet, would we pick them now?”
If the answer is a definitive no, then you are ready to move on. If you are not sure, you should look further into all the options.
Looking forward, not back
The next step is to build a business case for staying with your existing solution compared to making a switch. So rather than looking at what you’ve spent so far, take a look at what the comparative ROI will be going forward.
Here’s what you need to model:
The cost of each platform
Your current automation rate & how much this will increase over the next year
The quality of the automation
The expected automation rate from the new provider within the first year
How quickly you will be able to match your current automation rate
What the ceiling on automation can be with the current provider
Any other additional features
For simplicity, let’s assume the cost of the platform is the same, and the expected performance is credible. You should end up with a model that looks a little like this.
You may want to consider some conservative efforts of how long it’ll take the new platform to get up and running, in which case you’ll have something like this.
From there, it’s about managing the level of risk you are happy with.
The quality of the automation
For simplicity we've been comparing automation rate as though all automated responses are the same. We've written before about why this may not be the case, but to sum up:
There is no standard definition of "automation rate" meaning that it can be hard to compare apples to apples
The total number of tickets "automated" can include partial and full resolutions. Partial ones still need to be handled by a human.
"Automation rate" does not take into account how well the AI responded. For that you need to look at CSAT or another rating system.
Therefore it's possible that provider 1 offering 25% automation is fully resolving 0 tickets, but just gathering information from customers, whereas provider 2 is offering 20% automation rate with 80% full resolution.
We also see examples where there is no CSAT rating at the end of a AI conversation, so brands are unaware of how well the automations are being received.
De-risking the switch
Of course, the model above is predicated on the fact that the figures the new provider is giving you are credible. Switching from a supplier that is working (at an adequate, if not incredible rate) to a new one is always a risk.
Here are some ways to de-risk the decision.
Channels – one of the easiest ways to increase AI automation rate is for AI to be answering questions across channels. Can the new solution handle channels you are not currently automating?
Use cases – can you see live examples where the new provider is able to handle use cases that you are not able to currently automate?
Depth of integrations – can the new solution integrate with more of your systems, but not only that, go deeper and perform more actions
Referencable customers – does the new solution have customers you can talk to, in order to get realistic expectations of time to value
Support – how well will the new provider support you in your switch? How does that compare to the current level of support?
Pilots or proofs of concept – can the new supplier show you something working before you commit?
Other functionality – does the new provider help you with anything new, or can replace another existing solution?
Any decision to change is going to be a leap, but it doesn’t have to be into the dark if you can speak to as many relevant customers as possible.
Conclusion: are you falling behind with the status quo?
The easiest decision you can make is to do nothing. Sometimes it’s the best choice, but often it’s not. In the world of AI, there are constant innovations happening, and new developments coming out. If you stick with adequate performance you can easily be left behind.
Just as the sunk cost fallacy is real, so is the concept that the grass is greener on the other side. To take emotion out of the equation, you have to build out a business case for keeping things the same or changing.
If you are considering switching to DigitalGenius, our team can help you build out a business case and show you what is possible to help you make the best decision. Speak to us today.
