I was very fortunate to be invited to speak at the AI Implementation Summit in Toronto yesterday, and it was an excellent conference -- it wasn't about buzzwords, it was about how to actually use AI as a tool. Chris Warburton and Simon Howard brought together some amazing people who knew what they were talking about. There were several speakers, a few roundtables, and to me, it sounded like a lot of us were on the same page.
The recurring message at the conference -- AI is not a magic wand. It is not going to make everything better. In fact, John Sanders from Bridgeforce said it best -- AI laid on top of a bad process will make it far worse, with brutal efficiency.
You can use ChatGPT or Grok in your homes, right now -- it can make a video of you juggling, it can write you a song, it can randomly pick a number between 1 and 100, or you can ask it what is the best collection agency in Canada, but what you are doing is just using computing power -- huge servers somewhere are making a kitten video for you, and you may or may not pay a small fee for that computing time.
Credit and collections can use AI to analyze a spreadsheet, look for trends in data, it can analyze call recordings and transcribe them, or it can partner with other IT services such as VOIP phone services, SMS gateways and text to voice programs to simulate a virtual agent and call people and speak from a script. It can do very powerful things, using those servers for a fee.
Collections has always been about economy of scale -- nothing about collecting past due accounts is complex, but it does take manpower, focus, and tools of the trade. Those tools over the decades have been Vernon's Directories, database programs, auto dialers, mailing machinery or vendors, credit bureau data scrubs and segmentation, SMS and email services -- and now AI.
I'm using an AI agentic agent -- I named him HAL. It calls low balance or low probability files, and can do so with a lot more manpower than a single human (or group of humans) can do. It is not meant to replace my team, it's meant to augment it. 70% of right party contacts ask to be transferred to a human being, anyway ... but it scales very efficiently and gets my team talking to people more often. It can be 'brutally efficient', making thousands of calls a day, leaving perfectly crafted personalized answering machine messages, offering self-cure options, and reaching people. But it's only as good as it's data (what we ask it to call), and it's process (how it links with our phone system, our SMS gateway, and our team of people).
And when we use these tools, we need to think about how real people and real companies who owe money will respond to an AI outreach. When we first started using HAL, 80% of consumers hung up on it ... we got that down to about 12% by changing voice, tone, what it said ... if people don't want to interact with it, they won't. At another conference, someone said 'all collection agents will be out of work in 10 years' ... I responded 'how are those self-checkout machines doing at Walmart and McDonald's'? They've been in place for decades and people still resist them. People will do what people will do.
We've been using our agentic agents for over a year now, and it is constantly being evaluated, and only working a small segment of what it could, because we want to get it right. We look at transcripts daily, reviewing how people interact with it, what data it brings back, and ultimately, it's revenue over it's cost.
The world of AI is changing constantly ... it's going to look very different in six months or a year, costs of using it will either go up or down, government regulators are going to set boundaries, and we might all have head chips .. who knows? But we all need to worry less about buzz words and magic wands, and more about tools at our disposal, and how we are going to use them.
Got thoughts? Want to pick my brain on how we use AI? Drop me a note at blair@receivableaccounts.com.
Thanks kindly,
Blair DeMarco-Wettlaufer
KINGSTON Data & Credit
226-946-1730
No comments:
Post a Comment