VAUG Agents
AI agents that answer, qualify and reconcile, while your team does the real work.
We design agents around your workflows, connect them to your CRM, inbox, WhatsApp and ERP, and run them for you. Every action is logged, every risky step goes to a person, and they get better every month.
- Support
- Sales
- WhatsApp & voice
- Finance
- Operations
- Documents
VAUG Agent Console
Support Agent · WhatsApp
Hi, my order #4821 hasn't arrived. Can I get a refund?
Illustrative · not real customer data
Answer customers, Qualify leads, Book appointments, Match invoices, Chase payments, Update the CRM, Read documents, Route tickets
- typical time to a first agent in shadow mode
- 2 wkstypical time to a first agent in shadow mode
- of agent actions logged and reviewable
- 100%of agent actions logged and reviewable
- across chat, email, WhatsApp and voice
- 24/7across chat, email, WhatsApp and voice
- improvements included in every plan
- Monthlyimprovements included in every plan
Agent types
Agents We Build.
Six families of agents cover most of the repetitive work in a business. Pick one to see how it runs.
Customer support
Answers questions, tracks orders, handles returns and escalates what needs a person, in your tone of voice.
- TriggerCustomer message arrives
- Step 1Identify customer and order
- Step 2Search policies and history
- Step 3Draft or send the answer
- Step 4Escalate refunds over the limit
- OutcomeTicket resolved and tagged
What it handles
- Order status and tracking
- Returns and refunds
- Account and billing questions
- Policy and FAQ answers
- Ticket triage and tagging
Where it works
- Web chat
- Zendesk
- Intercom
- tickets resolved without a human
- 60–70%tickets resolved without a human
- first response
- < 30sfirst response
Typical targets, agreed per project.
Process
How an Agent Project Runs.
Five stages, a demo every week, and nothing goes live until it has proved itself on your real work.
Week 1
Audit
We shadow your team, measure where hours go and pick the workflow with the clearest payback.
Weeks 2–3
Build
Agent, knowledge, integrations and guardrails, demoed to you as they grow.
Weeks 3–4
Shadow run
The agent drafts, your team approves. We measure accuracy on real cases.
Week 5
Launch
It goes live on the tasks it has earned. People keep the edge cases.
Monthly
Improve
Monthly reviews, new skills and new workflows, reported every Friday.
Human in the loop
Humans in the Loop. Guardrails by Design.
An agent is only useful if you can trust it. These controls come as standard.
Approvals
3 waiting- Waiting Approved
Support Agent
Refund on order #4821
Above auto-refund limit
- Waiting Approved
Lead Qualifier
10% discount offer
Enterprise lead, needs sales sign-off
- Waiting Approved
Finance Agent
Mark 6 payments duplicate
Low-confidence match
Approve · Edit · RejectEvery decision logged
Approval thresholds
Refunds, discounts, payments and anything above your limits wait for a person.
Grounded answers
Agents answer from your approved knowledge and systems, and say so when they don't know.
Full audit trail
Every message, tool call and decision is logged with who approved what.
Scoped permissions
Each agent can only read and write what its job needs. Nothing more.
Evaluations before release
Changes are tested against real past cases before they reach customers.
Instant off-switch
Pause any agent or skill in one click, and fall back to your team.
Integrations
Works Where Your Work Already Lives.
Agents plug into the tools you use today. No migration, no new inbox to check.
- CRM & sales
- HubSpotSalesforcePipedriveZohoClose
- Messaging & voice
- WhatsApp BusinessTwilioIntercomSlackTeams
- Email & calendars
- GmailOutlookGoogle CalendarCalendlyCal.com
- Helpdesk
- ZendeskFreshdeskGorgiasHelp ScoutFront
- Finance & ERP
- XeroQuickBooksNetSuiteSAPStripe
- Commerce & data
- ShopifyWooCommerceAirtableGoogle SheetsPostgreSQL
Responsible AI
Responsible AI, in Plain English.
We decide with you, in writing, what the agent does on its own and what always stays with a person.
What the agent does
- Answers routine questions from approved knowledge
- Looks up orders, bookings, invoices and records
- Drafts replies, journals and summaries
- Updates your CRM, helpdesk and spreadsheets
- Books meetings and sends reminders
- Flags anything unusual for review
What stays with your people
- Refunds, credits and payments above your limits
- Complaints, sensitive or emotional conversations
- Discount exceptions and contract terms
- Medical, legal or financial advice
- Hiring, firing and anything about staff
- Final sign-off on month-end and reporting
Case studies
Agents in Production.
- reconcile.traxpay.internal/exceptions?status=open&sort=confidenceReconcile
Exceptions
Close 30 Sep · run 06:00 CET · 38 open
Search reference, merchant…CSVAll38Split settlement12Fee at source9Timing7Duplicate6Other4Sort: confidence3 selected€15,033.62ApproveOverrideReassignReferenceMerchant · sourceAmountReasonConfidenceAgePSP-88240Nordlicht Mode GmbH · Adyen€12,005.12Split settlement64%4hWLT-10923Kaffeerösterei Lang · PayPal−€89.90Duplicate refund22%4hSTR-55190Velo Werk Berlin · Stripe€640.00Fee at source71%4hSEPA-40188Alpen Outdoor AG · Deutsche Bank€3,118.40Timing (T+2)83%1dKLN-20471Haus & Hof Online · Klarna€1,249.99Fee at source58%4hPSP-88262Studio Feinkost · Adyen€7,402.66Split settlement76%4hSEPA-40203Brettspiel Kontor · Commerzbank€412.07FX rounding91%2dWLT-10930Grünwerk Naturkosmetik · PayPal€58.35No ledger entry12%3dPSP-88277Nordlicht Mode GmbH · Adyen€2,871.30Split settlement69%4hSTR-55204Lindenholz Möbel · Stripe€19.00Fee at source88%4h1–10 of 38Rows per page: 10Fintech & Insurance · Frankfurt, Germany
AI as a Service
A Reconciliation Agent That Closes the Books in Hours, Not Days
An AI agent that matches bank, PSP and ledger records, explains every mismatch and hands finance a clean exception queue each morning.
- transactions matched automatically
- 92%transactions matched automatically
- month-end close (down from 3 days)
- 4 hrsmonth-end close (down from 3 days)
- manual review time
- -70%manual review time
- 10:16Summerhill HealthBusiness accountTodayThis business uses a secure service from Meta to manage this chat. Tap to learn more.Hi, I need to book the knee MRI my GP referred me for. I'm with Bupa10:14Thanks Joanne, I've found your referral SH-58213 (MRI left knee). Earliest slots:
Salford Quays · Thu 1 Oct, 14:30
Stockport · Fri 2 Oct, 09:1010:14salford thursday please10:15Booked for Thu 1 Oct, 14:30 at Salford Quays (ref A-2291). Please fill in the MRI safety form before your scan.10:15Open form10:17MRI safety questionsSection 2 of 5 · Implants and metal
Do you have a pacemaker or implanted defibrillator?
YesNoNot sureHave you ever had metal fragments in your eyes, e.g. from welding or grinding?
YesNoNot surePlease give details
Welder 2006–2011. Had a bit of metal taken out of my eye once
ContinueManaged by Summerhill Health. Learn more
Healthcare · Manchester, UK
AI as a Service
A WhatsApp and Voice Agent That Answers Every Patient Call
An AI agent that answers calls and WhatsApp messages, books scans, collects pre-scan questionnaires and sends reminders across 9 imaging centres.
- patient contacts resolved without staff
- 78%patient contacts resolved without staff
- answer time (down from 6 min)
- <5sanswer time (down from 6 min)
- missed appointments
- -38%missed appointments
How we start
Three steps to your first live agent.
Start with one workflow. The agent proves itself on your real work before it ever answers a customer on its own.
- This week
Step 01
Agent discovery call
We map where your team loses hours and pick the workflow with the clearest payback.
What you get
- Ranked list of agent opportunities
- Expected accuracy and payback
- The integrations you'd need
- 3–4 weeks
Step 02Where most start
Shadow-mode pilot
One agent drafts replies and actions on live work while your team approves each one. We measure everything.
What you get
- A working agent on your channel
- Accuracy report on real cases
- Go-live recommendation
- Monthly subscription
Step 03
Managed agents
Live agents hosted and watched by us, improved every month and extended to new workflows as they prove out.
What you get
- Hosting, monitoring and alerts
- Human approval on risky steps
- Monthly improvements
- Proposal within 48 hours
- Every action logged
- You own your data and prompts
- NDA on request
Questions We Hear Often.
What's the difference between a chatbot and a VAUG agent?
A chatbot answers questions. An agent also acts: it looks up the order, issues the refund (with approval), updates the CRM and books the call. It works across your systems, not just in a chat window.
Which AI models do you use?
Whichever fits the job: Claude, GPT, Gemini or open models like Llama and Mistral. We choose per task for quality, cost and data residency, and can switch without rebuilding.
Is our data used to train models?
No. We use enterprise model APIs that don't train on your data, or host open models privately. Data stays in UK or EU regions by default.
What if the agent gets something wrong?
Risky actions need approval, low-confidence answers go to a person, and every action is logged. When something slips through, we fix the cause and add it to the test set so it doesn't happen again.
How long until an agent is live?
Most first agents run in shadow mode within two weeks and go live around week five, once they've proved themselves on your real work.
Do we need technical staff?
No. We build, host and maintain everything. Your team just approves, reviews and tells us what to improve.
Can you build agents inside our own product?
Yes. Through Custom Development or AI Engineering we build agent features into your SaaS or app, which you then own and run.
Get in touch
Tell us which work you'd hand to an agent.
Describe the workflow in a few lines. A senior AI lead replies within one business day with a first view on what's possible.
- Free 30-minute strategy call
- NDA on request
- Proposal within 48 hours
Every repetitive task is a job
an agent could do today.
Free workflow review · NDA on request · Proposal within 48 hours