Enterprise Connect AI: How AI May Develop in the Short TermEnterprise Connect AI: How AI May Develop in the Short Term

Enterprise Connect AI ended on Wednesday afternoon with a ‘locknote’ panel session moderated by Eric Krapf, General Manager and Program Chair for Enterprise Connect. The panelists included:

The session provided a wrap-up of the topics covered across many of the conference sessions and each of the panelists offered some forward-looking thoughts on how AI might develop in both the short and longer term. What follows is a transcript of the locknote. It has been edited for clarity and brevity.

Krapf: From your own research, from what you've heard at the event, and people that have talked to you, with respect to AI, what's working right now? What's not been so successful and maybe needs to be put on hold for a little bit?

Gajwani: Mostly, AI is not working in the way we're used to computer software working, meaning it'll work 80% to 90% of the time – while at the same time we're learning how to make reliable enough to use it. We're also learning what the user experience can be, and that it is better to deploy AI in a place where it might fail and consumers don't get upset at you. As a result, the use cases today are largely internal, because that's where you can tolerate a 90% effectiveness.

What I see [working] is with content creation and content manipulation, and then software production with some big improvements in software productivity. And then there are horizontal use cases where your employees are using ChatGPT, or using Copilot in Microsoft Office, etc. So, I think some variation of those three things are where the primary productivity gains are today.

Kerravala: I’ve seen the same thing – a lot of internal use cases. I was talking with an HR officer from that a large company last week about this topic. That is, with people being more productive because they are using these AI tools, what do you do? She speculated that maybe this is a step toward a four-day work week or maybe you raise people's goals? I think, too, that even as we're getting better at using AI, it is also opening a bunch of questions that we aren’t sure how to answer, such as where – or where not – we should use it.

Gareiss: I think we're going to see is a lot of changes with contact center agents. I think the agent role is evolving very quickly – actually, evolving is not the right word even, because it's going so fast. What we're finding is that when companies bring agent assist into the contact center, the average handle time typically drops by about 28%. So, say it took 10 minutes on a call to solve a problem; with agent assist, it takes seven minutes. Now you have an extra three minutes.

Up until recently, we've seen companies [use those extra minutes to] just send more calls to agents so they get more productive. And at the same time, they have AI automating a lot of these interactions there are fewer calls going to the live agents and those live agents are becoming more productive. So what happens? We get more idle time. And when you have idle time, what comes next? Layoffs.

What we see a lot of companies doing now is take those extra three minutes and have agents upsell and cross sell – and use AI to help agents do that. About 54% of companies right now have a sales quota attached to their service representatives on in the contact center. You can start generating a lot of revenue in the business that way, especially when you can use AI to help coach these agents along. Maybe you just give them a screen prompt that says, ‘Okay, here's the best offer you can possibly make to this particular customer.’

Now, the idle time problem is addressed and the agents become hugely valuable because they’re bringing in revenue. Now, the contact center is a value center, not a cost center.

Kerravala: But you might need a different type of person in there [to do that selling].

Gareiss: We see about 71% of companies right now are hiring more experienced agents and agents with different skill sets. So that's absolutely happening. But even those agents who say, ‘You know what, I'm a service agent. I don't want to sell. I'm leaving,’ if you can get them to stick around for a little while and at least try AI – which is very well scripted well thought out – all of a sudden, they're getting commissions, they're making more money, they're getting bonuses. So, it's a win-win for everybody.

Leaden: To complement what you're saying here, Robin, conversational AI for basics is really working – for appointment confirmations, scheduling or even paying a bill. Another piece that we're growing into now is knowledge management, meaning: How accurate is that data that they're using for AI? We're constantly telling our clients: ‘you have to embrace AI. If you don’t, you're a laggard.’ And when you do embrace it, start slow. Start internally, because that's easier. But you have to get in front of the customer at some point. We're in a world now where customer experience is paramount, and I think that AI is critical to CX.

Gajwani: I'm not as close to the call center world as you guys are. I'm more in the technology world, the AI world, and I think one of the things that's striking to me is that the rise of call centers has reduced the expectation of quality of a call center interaction because we expect them to be low complexity and high volume. So, when I call and I say, ‘I want to talk to a human,’ it's because I have some question that can't be addressed by a wiki [article]. The standard has been I can either do self-service or I can have a human repeat the self-service answer to the customer. As the self-service becomes so good that you don't feel the need to have a human repeat it to the customer, then when I want to talk to a human it is for the more complex issues.

I think that tier one customer service will begin to go away, so you really need to be training people up towards tier two, tier three, tier four, so that by the time you reach a person, I'm talking to a human, because I need to, because that person is going to have human level knowledge and empathy and flexibility and expertise. So, I think this is a call to action to get rid of the low complexity work and shift your humans to high complexity work.

Krapf: Robin, you've done some research directly on consumer attitudes towards AI. What did you find?

Gareiss: Right now that what we're seeing with consumers is that they're not thrilled with customer service. If you ask them, ‘how has customer service changed in the last 12 months?’, about 30% will say it's gotten better. A little over about 35% will say it's gotten worse, and the rest will say it's stayed the same. If you ask that same question to the IT-CX and business unit leaders, 78% say that it's gotten better.

So there's a huge disconnect. The businesses thinks it’s doing great while the consumers aren’t happy with it right now. And honestly, a large part of that is because of the early generation of bots that were not trained properly or were not applied to the right problems or were poorly designed. That's really put a bad taste of the mouths of consumers for using AI.

And then on top of that, you have 30 plus percent turnover rates among [contact center agents]. It's hard to keep staff in the contact center that's up to speed and fully trained and knows what they're doing.

In our most recent round research that’s just completed and not even published yet, I'm starting to see a little bit of a turnaround – a few percentage points – but it's notable at least that consumers are starting to say, okay, my most recent chatbot experience wasn't as bad as it was before. It's getting better.

Krapf: Okay, so we have some incremental improvements and quick wins. As you look out six months, what should enterprise technology decision makers and businesspeople expect and be ready to act accordingly on? Will there be more of a transformational change in 2025, or will it be a long-term series of incremental improvements?

Leaden: It won’t be transformational over the next two to three years, but it will eventually become transformational. I just saw some stats that in the next eight years (2032) the UC industry and contact center industries combined will be $132 billion – which is very nice, but it will be dwarfed by AI, which will be Gen AI, which will be $1.3 trillion. That alone suggests we must transform, but I think it's going to start small, and we’ll adapt over time.

Kerravala: I remember the early days of the Internet – I think Cisco put out a report trying to calculate its economic value. Whenever anybody did that, they were grossly wrong. But the fact is, Gen AI will change every aspect of everything we do. You'll see Gen AI interfaces built into all the applications we use, all the appliances we use. I think the leading brands will set the things in motion.

Gajwani: I agree with what you were saying, Steve, about how it won't be transformational in the next couple of years. I think I'll take an even harder stand on that, which is, I think the vast majority of early applications are going to fail. The vast majority of early deployments are going to fail. Most of these are experiments, right?

The reason why Silicon Valley is so excited about AI is because we made a fundamental computer science discovery like: Did you know you could make a computer do this? I had no idea. That's amazing. Wow. This is going to change the world.

It's not changing the world yet, because the [stuff] doesn't work that well.

In every infusion of a new technology, in addition to the core technology, you need infrastructure around it, right? It reminds me a lot of the late 90s, where everyone said, Internet everything, right? You're going to order dog food online. You'll have friends online. You'll live your life online. You'll chat with people around the world by video. All those old AT&T commercials from the mid 90s, and all of that was absolutely ridiculous [over] the next five to 10 years – that 1995 video of the guy talking on a video phone to his kid was ludicrous. But today, we live all of that. All of that has completely come true, but it was wildly optimistic in the 10-year time frame, much less the two-year time frame. So, I think you have to have a little bit of patience.

Gareiss: I would say transformations have already taken place, although I agree with you guys from a high-level perspective that more advancements will be made in the future. If you look at CX, it's really focused in on AI. And when you look at some of the use cases, some of them are very simple, call summarizations and call transcripts already. We're seeing agents save 35% of their time on after call work because they have summarization. That's transformational for companies.

Kerravala: I was talking with some of the folks at the Mass General and they're using AI for MRIs. When they first started it, the doctors were dead set against it. They did not want machines doing their job. And then the doctors found out that it could spot brain bleeds and other things they really couldn't see with the naked eye. So now, instead of spending 80% of the time diagnosing, 20% treating, they're spending hardly any time – 10% of the time diagnosing, because the machines are doing it – and 90% of the time treating. But they had to get over that mental hurdle of having AI do part of their job.

Krapf: The thing that people really seem to be worried about is the AI governance, hallucinations, security, data privacy, etc. What should enterprises do to impact that governance situation within their own enterprise?

Gajwani: I think that all of the large vendors are racing to knock down those objections one by one, because they know it's such a big deal to adoption. So all of these early problems with models or problems with these use cases are being addressed now. I think most of the large vendors are addressing it through indemnification, meaning, ‘we'll keep your data secure, we promise, and you can sue us if not.’ But there's a new class of vendors who are going to take the existing models – and I'm speaking specifically of foundation models, the LLMs – and are building versions of those that can be run locally.

Kerravala: If you are a financial services firm or law firm, and you use AI tools that can join calls -- and I love those things because they transcribe and summarize – but now that's a digital asset that becomes discoverable. So from a governance perspective, should I allow it to be used at every call? Should I not? That's not a vendor issue. The company needs to set those rules.

Gareiss: We do see companies setting up Centers of Excellence for AI and security compliance is one of the big components of those centers of excellence. But, just under 40% of companies have already set one up, and those are usually larger companies – but a lot more are planning to across all sizes of companies. You have to do that with AI.

The other thing I would say is about call reporting is: you have to record calls for a variety of reasons, including compliance. But we're seeing a lot of companies now saying AI has huge value in call recordings, because now I can go back and look at compliance issues historically over a certain period of time. Obviously if it’s just a recording, you have to transcribe it, but because of AI and its ability to classify and summarize a lot better than humans can, you have the ability to detect if there were PII violations, or insider trading or whatever else.

Krapf: Over the next six months or so, what do you think are the top things that enterprise decision makers ought to be doing to position themselves for where AI is likely going in the next six to 12 months?

Gareiss: Make sure you're measuring success. When you're rolling out a new technology, figure out what metrics you'll measure. Take your baseline. A lot of people forget to take their baseline, and then they can't really measure their incremental improvements. Also, make sure you roll out AI conservatively, rather than trying to boil the ocean.

In the early days, when we looked at our research ‘success groups,’ the companies who tried to boil the ocean were not successful. Those which embarked on more conservative, very deliberate projects showed a lot of success. And, because they were measuring the results, they could go back to the budget holders and say, ‘Okay, here's what we did this time. Here's my next project.’ They were able snowball that.

Leaden: It's all about a moment in time where, if we don't go after this, we're going to be left behind. A new term, agentic AI, has now come out in the last couple of months. It’s maybe the next level of AI. My point is, from March 2024 to now, the acceleration of AI feels like it's three years not six months.

Gajwani: I agree, and I think the pace of change is going to increase. When the pace of change is so fast, large investments constitute basically a large bet on where the world is going to go. That can be very risky, because the world can change so fast. My personal belief is that the order of the day for enterprises is incremental experimentation and then hygiene.

Experimentation means you need to get some ‘reps in.’ Your team needs to learn how to use this stuff. Your IT department needs to figure it out. Hygiene means, as Robin was saying, things like KPIs and measurements and getting your data house in order, the quality and the scale of your data and the cleanliness of it matters more now than ever before.

We've all heard garbage in, garbage out. Well, now we need, truckloads of data, and you have truckloads of garbage, you'll get a bad outcome. So, focus on hygiene and a little bit of incremental experimentation will set you up really well, because two years from now, we're going to have crazy new capabilities.