Backed byYCombinator
The operating system for financial services.

Autonomous agent
teams that get work
done in customer
facing ops

Cozmo does the full job of your ops team running your customer work.

Onboarding
KYC / KYB
Collections
Claims
Servicing
Tax filings
Loan applications
Customer support
Replace a workflow this month
CozmoAgent Run · #4521
Live
See01
Policy №SF-84729
Speak02
CustomerSpeaking0:02
I need to file a claim for yesterday's accident.
Think03
Identify claim type
Verify policy coverage
Extract incident details
Assess next-best action
Do04
Claim #4521 filed in Salesforcejust now
Notified adjusterSMS sent
Live with Fortune 100s,
global banks, insurers,
and the BPOs serving them.
Nestlé
Discovery Bank
SDC
U
01Takesthe call.
02Readsthe documents.
03Decidesagainst your policy.
04Closesthe case in your system.
10M+
regulated interactions in production
78%
lower cost per interaction
800
compliance checks per exchange
4 days
contract to live calls

Why we are building Cozmo AI

01The thesis

Every seat on your floor costs $3,000 to $8,000 a month. We’re accountable for the outcome, not the seat.

An ops seat running customer work costs $3,000 to $8,000 a month fully loaded, whether it sits on your own floor or on your outsourcer’s. For that, you get one human who works one shift, speaks one language, needs six months of training, churns in eighteen, and produces an outcome your QA team samples at the end of the week.

Cozmo replaces the seat. Not assists it, not augments it. Replaces it. You stop paying for a human’s time and start paying only when the case actually resolves.

BY THE MONTHONE BPO SEATONE COZMO AI EMPLOYEE
Fully loaded cost$3,000 to $8,000~1/5 of that
Shifts coveredone24 / 7
Languages spokenone100+
Training before productive6 months4 days
Annual attrition40%0
Compliance QAsampled after the fact800 checks live, every exchange
Documents read on callwhat a human can skimfull ingest, extracted, attached
System of record updatetyped by a second personwritten before the call ends

The shift your CFO has been waiting for. The shift your P&L was built to demand.

02What an AI employee actually does

A seat’s job.
Not a chatbot’s job.

The reason voice AI vendors, RPA tools and horizontal agent platforms haven’t replaced the ops seat is that they each do one piece of the seat’s job. An ops seat has to do four things, in order, on every interaction. Miss one and you still need the human.

01.

Hear the customer.

Sub-300ms voice. English, Spanish, Arabic dialects, Hindi, 100+ languages. Handles tone, interruption, ambient noise, accent.

voice.v4 · 287ms
100+ langs
barge-in
tone + intent
02.

Read what the customer sends.

IDs, trade licences, bank statements, claim forms, dashcam footage, collateral packs, tax documents. Read live on the call, structured data extracted, attached to the right record.

dashcam.v2
docs.ocr
form.extract
video.analyze
03.

Decide against your policy.

Your credit box, your authority matrix, your KYC/KYB rulebook, your refund thresholds, your regulatory obligations. Applied inside the agent's reasoning, not as a post-call filter.

policy.runtime
ape · 800/exchange
authority matrix
deterministic
04.

Act in your system.

Core banking, policy admin, LMS, claims platform, CRM, case management. The case closes in your system of record before the call ends.

Guidewire
Temenos
Salesforce
Finastra LMS
INBOUND RINGHEAR·READ·DECIDE·ACTCASE CLOSED · SIGNED

One agent, all four, every time. That’s what an ops seat does. That’s what Cozmo replaces.

03The numbers a replacement actually moves

Every workflow we replace. Every number on the other side.

Live customer results. Same thesis, replace the seat, across different workflows. Pick the one on your P&L.

#WORKFLOWTHE NUMBERCUSTOMER 

The pattern is the same across every workflow. The seat gets replaced, the cost line drops, the throughput metric climbs, the audit trail gets cleaner. Your outsourcer doesn’t have an answer to this. Your in-house floor doesn’t either. We do.

04Why Cozmo, not the alternatives

There is no other category that owns the whole customer workflow.

Every other approach forces you to stitch together two or three vendors and hope nothing breaks at the seams between them. The seams are where the regulator finds you.

ALTERNATIVEWHAT THEY ACTUALLY DOWHERE THEY LEAVE YOU EXPOSED
Voice AI platforms
Vapi, Retell, Bland

Voice infrastructure. You bring the agent logic, the document processing, the system integrations and the compliance layer.

✕ GAPYou are the systems integrator. You own the FDCPA risk. You build the audit trail yourself across four vendors.

Horizontal agent platforms
Sierra, Decagon, Ada

Customer support agents for unregulated SaaS and ecommerce. Strong on chat. Weak on voice, documents and regulated decisions.

✕ GAPNo deterministic guardrails for TCPA, FDCPA or BSA/AML. Built for the SaaS help desk, not the regulated call center.

Offshore BPOs
Concentrix, Teleperformance, regional partners

Trained humans on the phone. Per-seat pricing. QA by sampling. Compliance by training and hope.

✕ GAP40% attrition. 6-month training cycles. You see the compliance breach in next month's QA sample, not in the moment.

RPA + bolt-on AI
UiPath, Automation Anywhere

Scripts that act inside systems on structured inputs. Bots break when forms change or customers go off-script.

✕ GAPCannot hold a conversation. Cannot read an unstructured document. Falls back to a human queue at the first edge case.

Building in-house
Your ML team + a dozen APIs

Your ML team stitches together LLM APIs, voice infra, OCR, RPA and a homegrown compliance layer.

✕ GAP18 months and $2M+ before the first regulated call. Then you own the model risk, the SOC 2, and every compliance update forever.

Cozmo AI
SEE · SPEAK · THINK · DO · AUDIT

The only platform that does all four jobs inside one audited state, with TCPA, FDCPA, GLBA and BSA/AML built into the agent itself, not bolted on after.

05Compliance by design, not by policy document

Your QA team samples 2%. APE checks 100% of every exchange.

APE is Cozmo’s governance layer. It runs 800 deterministic compliance checks on every exchange the AI employee has with a customer. The full regulatory stack, enforced inside the agent’s reasoning, not bolted on as a post-call review.

APE · DETERMINISTIC CHECKS PER EXCHANGE
0
FIRED · PER EXCHANGE · P50 2.1MS
2%
What QA samples today
100%
What APE checks live
TCPA✓ LIVE

Outbound dialing with consent state tracked per customer per channel.

quiet-hoursconsent.v4channel_opt_in
FDCPA✓ LIVE

Mini-Miranda, validation notices and dispute handling enforced on every call.

mini_mirandavalidation_30ddispute_cease
GLBA✓ LIVE

Identity verification before any account information is disclosed.

auth_tier_2pii_redactionneed_to_know
BSA/AML✓ LIVE

Thresholds applied live during KYC, with SAR-relevant flags routed for review.

ctr_thresholdsar_indicatorsofac_screen

Every decision logged with the input, the rule it followed, and the policy clause behind that rule. Ready for the regulator.

06What customers say

Live in production.
Not a pilot.

Cozmo is running regulated customer operations today across insurance, banking, lending and collections, for Fortune 100s, US carriers, and digital lenders.

We replaced an entire collections floor. Cost per resolved account dropped 70%. The AI employee negotiates, processes the financial documents on the call, and produces an FDCPA audit trail that is cleaner than what our human team was doing.

COO
Leading US Insurance BPO
70%
↓ cost per resolved
· COLLECTIONS & RECOVERY

Dash-cam footage is analyzed instantly, the call is handled in the customer's language, document follow-ups happen automatically. One AI employee runs what used to be our entire FNOL desk.

Chief Product Officer
Motor Insurance Brokerage
1
AI employee replaces desk
· FNOL & CLAIMS

Same AI employee handling voice, messaging and documents 24/7 across English, Arabic and Spanish. No shift changes, no handoffs, no quality variation between agents.

CXO
Global Fortune 100
24/7
3 langs · 1 agent
· UNIFIED CUSTOMER OPS
07Deployment

Live calls in under 4 days. Not a pilot. Not a sandbox.

Your written compliance policies, your authority matrix, your call scripts, your system integrations. Loaded into the AI employee, dry-run against your historic calls, then live.

1
DAY 1
Workflow scoping

Pick the workflow. Hand over policies, scripts, system credentials.

2
DAY 2
Configuration

Compliance rules loaded into APE. Agent persona and authority matrix configured.

3
DAY 3
Shadow run

AI employee runs against historic calls. You QA the transcripts and decisions side by side.

4
DAY 4
Live

First live customer call. Phased ramp from there at the pace your risk team is comfortable with.

PRICING

Outcome-based.
You pay for resolved
cases, not time on
the line.

Onboarded customer. Collected account. Closed claim. Funded loan. Handled support case. Pricing is tied to the outcome the AI employee delivers, defined and measured against the same SLA your ops floor is on today.

UNIT
Resolved claim
per closed case
Collected account
per promise-to-pay
Funded loan
per disbursal
Closed support case
per CSAT-valid resolution

See what your ops floor does
in a day happen in minutes.

Pick a workflow from your operation. Send us the policy, the script and a representative document set. We’ll run a live AI employee against it on the call, with a clean audit trail at the end.