Customer support AI agent
Answers customers 24/7 on your website, Zalo OA and Messenger using your own product data and policies. Tricky cases go to a human with a summary attached.
- Website
- Zalo OA
- Messenger
FreeFirst workflow assessment
Hand repetitive work to a team of AI agents
Cake Study builds AI agents and automated workflows that run for real inside your business — answering customers, processing documents, updating your CRM, writing reports — so your people can focus on work only humans can do.
We reply within 1 business day · founder@cakestudy.comSERVICES
From one AI assistant to fully automated workflows
Internal knowledge assistant
Ask about processes, contracts or company docs and get answers with sources. New hires ramp up faster.
Workflow automation
Connect scattered steps across email, spreadsheets, CRM and accounting into one automated flow — with human approval where it matters.
New orderCheckInvoiceNotify team
Document processing
Read invoices, contracts, CVs and stock receipts; extract the right fields and push them into your systems.
invoice-0912.pdf14 fields · importedcontract-ABC.docx3 clauses to review
Automatic reports & alerts
Pull numbers from many sources, send a summary every morning and alert you the moment a metric drifts.
RevenueOrdersInventory
FOR LEADERSHIP
AI strategy & training
Find which workflows to automate first, estimate the impact, pick the right tools and train your team to use AI every day.
- 1-day workshop
- 90-day AI roadmap
- Team training
HOW WE WORK
Four steps, each with something working for you to see
- 01
Assess
Sit with your team, map the current workflow and pick the bottlenecks worth fixing first.
1–2 sessionsYou getWorkflow map + priority list - 02
Pilot (POC)
Build a working pilot on real data. If it doesn’t pay off, we stop — no lock-in.
2–3 weeksYou getA demo running on your data - 03
Deploy
Integrate with the systems you use, set permissions, keep logs and put approvers on sensitive steps.
4–8 weeksYou getProduction system + training - 04
Run & improve
Monitor quality, update as your process changes and report on impact regularly.
MonthlyYou getImpact reports & next steps
BEFORE & AFTER
A day in a sales team, before and after AI agents
An illustrative example from a distributor of about 30 people — the most time-consuming work is usually the repetitive kind.
Done by hand
Copy-paste, verbal reminders, digging through files
- 08:00Open 40 quote emails, copy each line into Excel1.5 hours
- 09:30Check stock in 2 systems, call the warehouse to confirmError-prone
- 11:00Write quotes by hand, wait for the boss to approve by chatSlow reply
- 14:00Re-enter orders into the accounting softwareDouble entry
- 18:00Compile the daily report for the sales lead45 minutes
- 21:30Customer messages after hours — nobody answersLost deal
With AI agents
Agents do the work, people approve- 08:00The agent reads emails, extracts requests and creates CRM deals
- 08:01Stock is checked automatically across both systems
- 08:05Quotes are drafted; the sales lead approves in one tap
- 08:10Approved orders flow into the accounting software
- 18:00The daily report posts to the team chat at 18:00
- 21:30The agent answers after hours and books a call for the morning
Use cases by industry
Retail & e-commerce
PROBLEMInboxes overflow during sales; slow replies lose orders.
WHAT AI DOESAn agent recommends products, checks stock and drafts orders.
TARGET OUTCOMECustomers get answers in seconds, even at 2 a.m.
Education & training
PROBLEMRepetitive admissions questions and student reminders eat staff time.
WHAT AI DOESAn agent advises on courses, sends reminders and grades quizzes.
TARGET OUTCOMEAdvisors focus on the most promising students.
Finance & accounting
PROBLEMHundreds of documents keyed in by hand every month.
WHAT AI DOESRead invoices, reconcile statements, draft journal entries.
TARGET OUTCOMEAccountants only review and handle exceptions.
Real estate & services
PROBLEMLeads arrive from many channels; sales call back too late.
WHAT AI DOESAn agent scores leads, qualifies needs and books meetings.
TARGET OUTCOMEHot leads are contacted within the first minutes.
WHY CAKE STUDY
Systems that actually run — not just slides
We are engineers and product people who use AI agents in our own daily work before bringing them to clients.
- Data securityNDAs, least-privilege access, and the option to run on your own infrastructure.
- Built for Vietnamese businessesZalo, e-invoices, local accounting software and natural Vietnamese.
- Humans stay in the loopAgents propose; people decide on anything involving money or customers.
- Full handoverDocumentation, training and admin access all belong to you.
Tools & platforms we integrate
- Claude
- GPT
- Gemini
- n8n
- Make
- Zapier
- Google Workspace
- Microsoft 365
- Zalo OA
- Slack
- HubSpot
- Supabase
We pick models by task and cost — no lock-in to a single vendor.
ENGAGEMENT MODELS
Start small, scale once you see results
Assessment & advice
FreeFirst session, 60–90 minutes- Walk through your current workflows
- Recommendations on what to automate first
- Time and cost estimate
Fixed-scope project
By scopeTypically 4–8 weeks per workflow- Pilot on real data before rollout
- Integration with the systems you use
- Team training and handover docs
- 30-day warranty after handover
Monthly partnership
Monthly planFor businesses that keep scaling- A dedicated AI engineer as your contact
- New and improved agents every month
- Quality & model-cost monitoring
- Monthly impact report
Exact pricing follows the assessment. AI model API costs are billed at actual usage, itemised every month.
Frequently asked questions
Is my company’s data safe?
We sign an NDA before seeing any data, request only the minimum access each workflow needs, and use enterprise API plans that don’t train models on your data. For sensitive data, the system can run entirely on your own infrastructure.
How soon will we see the first results?
A pilot is usually ready in 2–3 weeks. A complete workflow in production typically takes 4–8 weeks, depending on how many systems need integrating.
How much does it cost?
It depends on scope: number of workflows, systems to connect and security requirements. After the free assessment you get a fixed quote per phase. AI model API costs are quoted separately and itemised monthly.
We don’t have an IT team. Can we still do this?
Yes. We handle all the technical work; your team only needs to describe the workflow, give feedback on the pilot and approve results.
How is an AI agent different from a chatbot?
A chatbot mostly follows a script. An AI agent understands the request, looks up data, operates tools (CRM, email, spreadsheets) and completes a whole task — with human approval on the important steps.
Who maintains the system after handover?
You can run it yourself with the handover docs and training, or choose a monthly partnership so we monitor, update and extend it for you.
What does your team repeat every single day?
Tell us about one workflow you’d like to stop doing by hand. We’ll reply within 1 business day with how AI can help.