Sometimes It Isn't Easy to Automate Simple Things

Even in a future with powerful agentic customer service agents, some seemingly simple transactions will be very hard to automate. Recently, I had a simple need: I wanted to extend my stay at a conference for one night. Just extend the hotel reservation for a day. What could be simpler than that? But suddenly I found myself at the nexus of three different organizations playing the "Where's my reservation" game.

I was attending a vendor's conference, and the vendor wass paying for my room, so I needed to talk to their travel agent who oversees the room block. I called the travel agent and spoke with Angela, who was very friendly and supportive. The challenge was that the room block had just been given back to the hotel and any changes would need to be made with them directly. Angela went out of her way to help me understand who to call and how to talk to them and gave me the room block number so the person at the hotel could easily find the reservation. Just before I hung up, Angela asked me to wait a minute, she returned to tell me that it appeared that they had found a way to make the change without requiring me to call the hotel. She told me to keep my notes and that she would call me if they were not able to make the change.

This simple transaction had an extra twist: Any changes would need to be approved by the owner of the room block, which is the vendor. So, regardless of whether it was the hotel or the travel agent who made the actual changes to the room block, they would need to coordinate with the vendor for permission.

I didn't hear from my friend Angela again, so as far as I knew I'd have a bed to sleep in that extra night. This simple transaction took a human agent with some help from folks in the background about 20 minutes to solve; the challenges came from federating an interaction across three different organizations.

For those of you playing "Where's My Reservation" along with me at home, this is the score:

• Three different organizations—the vendor, the travel agent, and the hotel. There could be scenarios where my company was a sponsor of the event and there would be a fourth organization involved.
• Two separate approvals required—the hotel or travel agent to confirm there is availability for that extra night, and the vendor who needed to agree to pay for my extra night's stay.
• Three separate infrastructures, three different telephone systems, who-knows-how-many back-end systems (email, maybe chat, call logs, booking systems, who knows, maybe a fax was thrown into the mix).

Is this a transaction that could reasonably be automated in the next five years? I posted this question on LinkedIn to see what folks thought. One said that if organizations don't figure out how to automate scenarios like this, someone with better approaches and systems will take their business. Others said when we get our own bots, they will be able to federate across multiple organizations.

Call me skeptical. No one is arguing that this is anything but a very simple transaction. The challenge is building a bot that can act across multiple companies to complete one transaction.

Let's look at the wants and needs of the various players here:

• For the vendor—From their perspective, the details of the room reservations are their vendor's problems, be it the travel agent or the hotel. All they want to know is that they are not paying for anyone they don't want to pay for, and their customers are being treated reasonably well. This transaction is a one-off for them that will generate five or six transactions a year, not something that is cost-effective to automate.
• For the travel agent—Their job is to make sure that every traveler who is approved for the hotel block has a reservation there and that costs are managed. Theoretically, once the block returned to the hotel, the travel agent had no skin in the game for my extra night, and Angela was just going above and beyond for me. This is probably a situation that happens reasonably often with large events, but it would certainly be a very low-value thing to automate.

• For the hotel—I could see there being enough traffic across dozens of events every quarter to make this practical if this was solved at the hotel brand level, not at each hotel separately. I'm not sure hotels are set up to handle this sort of situation globally.

• For my personal bot—These reservations are within a group reservation. It is possible that the system could enable access to rooms in the block, but the approval process with the vendor would likely limit the ability for any external entity to self-serve reservation changes.

I guess the above is a long-winded way of saying I don't see a reasonable way that these interactions could be automated in the next five years. I know this is an odd one-off scenario, but it made me realize that there are simple things that will not lend themselves to automation anytime soon. There are a lot of simple activities that smart, agentic agents will start taking care of, but there are going to be limits for years to come.

I see a lot of automation over the next five years or so, just not for everything, and not even everything that should be simple.

Max Ball is a principal analyst at Forrester Research.