Write the rule once (anyone who mentions a bereavement, a health problem or nowhere to stay goes to a person) and Switchboard puts those callers straight through to your team. It follows the rules you give it; the judgement stays with your trained handlers.
Straight through. Human judgement. Plain-English rules. Calls on record.
The agent's rules can tell it to stop and put the caller through to the number you set, before it asks anything else. It's a straight transfer, so your handler starts by listening.
Switchboard does not mark a customer as vulnerable on the case. That remains a human judgement, made by your trained vulnerable-customer team. The agent puts the caller through; the handler assesses and decides.
There are no sensitivity dials. You write the rule in plain English, in the language your vulnerable-customer policy already uses, and change it whenever the policy changes.
Set the agent to record and you can play calls back in Switchboard. At the end of each call it writes a short summary and an outcome, and can email them to you.
Four Things to Put in Its Rules
You write these into the agent's rules in plain English, from your own vulnerable-customer policy. The agent follows them; it doesn't score anyone.
Imminent harm, unsafe accommodation, threats, accident-with-injury, or health crisis disclosures during the conversation.
Examples
Disclosed medical needs, mental-health conditions, recent bereavement, or impacts of disability on the customer's ability to manage the claim.
Examples
The things a distressed caller says. The agent works from the caller's words, so tone of voice on its own won't trigger the rule.
Examples
Indicators that the customer's circumstances change how the claim should be handled: children, elderly relatives, sole income, accessibility needs.
Examples
Putting Them Through
Write the rule so the agent stops and puts the caller through, without finishing its questions first. Then test it on recorded calls.
The caller mentions something your rules cover: a bereavement, a health problem, children in the car, nowhere to stay tonight. The agent follows the rule you wrote for exactly this.
Following its rule, the agent stops asking questions, tells the caller what's happening and transfers the call to the number you set for that agent.
The handler picks up a live caller with no briefing on screen. After the call, Switchboard writes a short summary and an outcome and, if you've set a notification email or webhook, sends them with the transcript and the details collected.
Trained vulnerable-customer handler picks up the call. They make the assessment, follow your VC procedure, and update the case record. The AI's role ends; human judgement starts.
Screenshots
What a customer tells you about their circumstances is recorded where each handler will see it, and a rule passes the complaint to someone trained to help.
The sun and moon switch a screen between light and dark, and a click opens it full size.
See all 77 screens in the product tourConsumer Duty Alignment
Consumer Duty, FG21/1 vulnerable customer guidance, FOS evidence: the same rule and the same records serve all three.
Consumer Duty requires firms to avoid causing foreseeable harm to customers. Telling the AI to put these callers through to a person, before it asks them anything else, is foreseeable-harm avoidance written into its instructions.
FCA guidance on vulnerable customers expects firms to identify vulnerability, understand the customer's circumstances, and adapt service accordingly. The AI puts them through; your people understand the circumstances and adapt.
When a FOS file is opened on a vulnerable-customer case, you have the call recording (if the agent records), the summary and outcome Switchboard wrote, and the handler's actions on the case in Timeline.
Switchboard reports conversations by channel, status and outcome, so you can see how calls ended. What the handler did next sits in SwiftCase's own reports.
FAQ
No. The AI follows the rules you write and puts certain callers through to a trained human handler. The vulnerability assessment itself (and the decision to flag the customer's record as vulnerable) is a human action by your trained vulnerable-customer team. Switchboard has no vulnerability detector of its own.
Whatever you put in its rules. A sensible start is the four categories above (life-safety language, health disclosures, distress, and capacity or dependency), written in plain English from your existing vulnerable-customer policy. The agent works from the caller's words; it doesn't analyse tone of voice.
It can. The agent follows instructions; it isn't a detection system with a measured miss rate, so it shouldn't be your only safety net. Tell it to put through anyone who asks for a person, and keep your handlers' own responsibility for spotting vulnerability.
False positives (callers put through who didn't need to be) are operationally cheap. The receiving handler completes the conversation either way; the only cost is a handler's time. You tune the wording of the rule during the pilot to balance the two. Insurers tend to err on the side of more transfers early on.
On the phone, the agent puts the caller through. A text conversation can't be put through to a person, so on texts the safer rule is to give the customer your team's number. On email, the email agent can hand the message to a person: it becomes a high-priority task in SwiftCase with the reason and the start of the email.
The call recording (if the agent is set to record) plays back in Switchboard, Switchboard keeps each call's summary and outcome, and anything the agent or your handler changes on the case is logged in Timeline. A transfer isn't written to the case on its own, so have your handler note it.
30-day pilot. We write the agent's rules from your existing vulnerable-customer policy. You review the recordings and the handler outcomes. No platform migration. No long lock-in.
