Customer Service
John Lim
How We Make Sure Alfie is Safe & Secure
Agents are the new frontier of AI, and with it comes an understandable level of caution and skepticism about trust and safety. After all, no one wants an AI agent that hallucinates and says something it shouldn’t.
Those are valid concerns to us as well. Before we ever let Alfie, the Experience Agent, near a real guest, we ensure he adheres to the best AI standards in place today. Here’s how we make sure Alfie gets it right before he ships, after he’s live, and every day in between.
Before he ever talks to a guest
Thousands of simulated conversations, across every surface he lives on, before anything ships.
Every change to Alfie runs through simulation first: real scenarios, real guest history, entirely separate from your actual guests. If it doesn't hold up, it doesn't ship.
After he's live
Real conversations are reviewed, and low ratings or overrides are flagged quickly.
We review conversations that didn't go the way they should have and flag fixes for low ratings or moments when a team had to step in. If something's not working, we see it fast.
Every single day
Alfie keeps getting retested against your own configuration, not a shared, generic model.
This isn't a one-time check. Alfie keeps getting retested as your team makes changes, and every live conversation feeds back into how he improves, so next month's version has already learned from this month's. And it's tested against your own policies specifically, so as your business changes, Alfie is held to your rules, not a one-size-fits-all standard.
What we actually hold him accountable to
Graded by a machine and a person, on accuracy, resolution, and how often he says something he shouldn't.
Every run gets graded, and not just by a machine grading itself. Real people review a sample of conversations too, not just an automated score, checking for the same things: accuracy against the real scenario, guideline adherence, how often he actually resolves the issue, how often he recovers a guest who was upset, how often that turns into a loyalty signup.
Fraud and abuse guardrails
Recovery limits you set yourself, verified against real order data.
You stay in control how recovery is carried out through Alfie. By default, Alfie limits recovery actions to once per order, and verifies the order actually happened using your POS data, online ordering records, or a receipt a guest uploads and matches against what's on file.
Beyond that default, you can set your own limits on top:
How often the same guest can receive a recovery offer (no more than two or three times a month, for example)
A total recovery budget per brand per period, in dollars or loyalty points (say, $300 or 300 points a month)
If a guest hits either limit, Alfie doesn't push through and will escalate to a person instead.
You'll see it too, not just hear about it
Regular, plain-language performance reports, the same metrics we hold Alfie to internally.
For clear visibility and accountability, pilot customers get regular, plain-language reports on how Alfie is actually performing, average conversation grades, and the same metrics we hold him to ourselves.
And on the data side
Secure infrastructure, protected guest data, and recovery that runs exactly as configured.
We put strict guardrails over your guest data.
Trusted infrastructure. Alfie is built on Momos' own battle-tested components: the same Customer Data Platform, Offers, and Loyalty systems already running in production.
Guest data protected. The AI providers behind the scenes never see a guest's name, email, or phone number. They only ever see what's actually needed to resolve the conversation, like order history where it's relevant, and nothing more.
Recovery runs exactly as configured. Alfie identifies the scenario, then runs your policy exactly as specified, nothing improvised. He can't change offer values or loyalty points on his own.
Alfie also doesn't collect anything beyond what's needed to do the job, no IP addresses, and no keystrokes. And where the law requires it, guests are told upfront that their conversation is being handled by Momos, not left to guess who they're actually talking to.
You’re in good hands
This is the work that happens behind the scenes so Alfie holds up when it matters. If you're evaluating Alfie, or already running him and want to know more about our AI guardrails, reach out to us for a personalized walkthrough here.
