per·vi·gil /pɜːˈvɪdʒɪl/ Latin, adjective. Wakeful throughout the night; ever watchful.
We build real-time intelligence for regulated customer conversations, so that the people least able to ask for help are the ones most certain to be heard.
Most contact centres check fewer than two calls in a hundred for signs that a customer is struggling. Regulation now expects every case to be found, supported and evidenced. That distance is the whole of our work.
Every conversation is analysed as it happens, not a chosen few after the fact. There is no sampling gap for a vulnerable customer to fall through, because there is no sample.
Our software never decides anything about a customer. It notices, it tells the agent what it noticed, and it recommends what your own vulnerability policy says to do next. The person on the call decides. Always.
Every flagged conversation leaves behind a complete record: what was detected, what was said, what consent was given, what support was offered. When a regulator asks how you treated someone, the answer takes a moment rather than a month.
When a customer is confirmed as vulnerable, Sentinel says so to your dialler and CRM in a form they can act on: take this account out of the campaign, route it to the specialist team, do not auto-dial. Someone who has just told an agent they are struggling is not rung again on Tuesday about the same debt. That is where the cost of a repeat call, and the distress of it, both stop.
Sentinel listens alongside a live call, recognises the signs that someone is in difficulty as they surface, and puts that in front of the agent while the conversation is still happening. By the time the call ends, the evidence pack has already written itself.
Not agent assist. Not QA scoring. Not sales flagging. One job, done during the call.
Generated automatically during the call. Confirmed by the agent before filing.
We are looking for three firms willing to build this with us.
Design partners shape what Sentinel becomes. It costs nothing, it starts with no customer data whatsoever, and it only goes further as trust does.
Sentinel is real-time vulnerability detection software for regulated contact centres. It listens alongside live customer calls, identifies signals that a customer is in difficulty as they emerge, prompts the agent during the call, and generates an FCA Consumer Duty evidence pack automatically.
Consumer Duty requires firms to identify customers in vulnerable circumstances and evidence that they were treated fairly. Sentinel supports both: it scores every call rather than a sampled subset, and produces a timestamped record of what was detected, what consent was captured, what support was offered and what happens next. Compliance itself remains the responsibility of the regulated firm.
No. Sentinel never classifies a customer, denies a service, or takes action on its own. It surfaces what it detected and recommends next steps drawn from your own vulnerability policy. The agent on the call decides and acts. That is deliberate, because decisions about vulnerable people should not be automated.
Call audio, a live transcript, the vulnerability signals detected, and the evidence note the agent confirms at the end. Sentinel acts as a data processor on your documented instructions and you remain the controller. Processing and storage take place in the UK and EU. We keep the transcript, not the recording: audio is transcribed and then discarded, and recordings stay on your own telephony platform.
Two things. The evidence pack is generated for the compliance record, and a disposition signal goes to your dialler and CRM: suppress this account from outbound campaigns, route it to the specialist team, do not auto-dial. Without that signal, a vulnerable customer wrapped with the wrong code stays in the campaign and gets called again, which means more distress for them and the cost of a repeat call covering old ground for you. Sentinel publishes the signal as a structured event. Your own systems consume it.
No. Sentinel is not general agent assist, not QA scoring, and not sales or opportunity flagging. It does one thing: it finds vulnerability while the call is live, acts on it before the next call, and evidences it before the agent hangs up. That narrowness is deliberate.
At two separate layers. First, speech recognition accuracy across regional accents, non-native English speakers, older voices and speech differences, because a degraded transcript degrades everything after it. Second, detection accuracy on clean transcripts, to isolate the classification itself. We treat under-detection as the graver harm, because a missed signal means somebody who needed help did not get it. This is our published testing method. Results go to design partners as a one-page fairness summary alongside the security overview.
All of them. Manual quality assurance in most contact centres reviews fewer than two percent of calls for vulnerability, and finds problems weeks after the conversation that mattered. Sentinel scores every call as it happens, without adding anybody to the floor.
FCA-regulated firms running contact centres: debt collection agencies, consumer lenders, banks, energy utilities, and the outsourcers who run collections and servicing on their behalf. United Kingdom first, with the equivalent regimes in Ireland, Germany, France, the Netherlands and Spain to follow.
Yes. The live demo runs on simulated call data with no real customer information. We are also taking on three design partners, and the first phase uses realistic scripted scenarios with no customer data and no integration, so your compliance team can look at it without needing a data protection review first. Running Sentinel on your own recorded calls happens inside that programme under a signed agreement, not through a public upload.
Spent his career in UK collections and recoveries. Runs the company and everything commercial: customers, partners, funding, and keeping the product honest to what actually happens on a call.
Builds the platform, from the speech and detection pipeline through to the evidence packs a regulator will read. Owns the product roadmap and the technology.
Pervigil is based in London. We started it because we had both watched, from different sides, how easily somebody in real trouble can pass through a five minute phone call without anybody noticing.