Now understands Saudi, Emirati, Qatari and Omani Arabic voice notes.See how it listens →
Sizmic Agentic CX

Agents that resolve.
From voice note to finished task.

Autonomous AI agents that understand your customers in Gulf Arabic and English, answer from your own knowledge, and complete the transaction in your ERP, CRM or clinic system. Your team is brought in only when a request is outside your limits or can't be understood. Tuned for Saudi, Emirati, Qatari and Omani dialects. Built, tested and run for you.

Autonomous within your limitsSaudi · Emirati · Qatari · OmaniKeep your WhatsApp number
agent livehosted in-regionOdoo connected
AN
Al Noor TradingWhatsApp Business · example
Operator console · what happened behind the replyconversation 4821
    Connects to the systems you already run
    • Microsoft Dynamics 365
    • SAP
    • Odoo
    • Zoho
    • HubSpot
    Why it's different

    Chatbots answer. Agents resolve.

    What to compare when you evaluate an AI agent for customer service.

    What to checkTypical chatbotSizmic Agentic CX
    Voice notes and dialectText keywords. Voice notes and Gulf dialects usually fail.Sizmic's proprietary two-stage speech pipeline, tuned with Saudi, Emirati, Qatari and Omani speech data and your product vocabulary. Unclear audio goes to staff with the recording attached.
    Source of answersA script or FAQ that drifts out of date.Your live data and documents: stock, account prices and policies, with the source shown for every answer.
    What it actually doesSends a link or a form, and leaves the work to your staff.Completes the transaction: checks stock, books the slot, posts the order in your ERP and sends the statement, on its own.
    Commercial riskRules live in the prompt, where a persistent customer can argue past them.A policy check in code before every action. Discounts, credit and clinical limits cannot be talked around.
    Who runs itSelf-service software. Your team maintains it.Fully managed: monitored, re-tested on your messages before every change, supported by the engineers who built it.
    How it works

    Five layers, built around your business.

    Every agent we deliver has the same five layers. Your language, knowledge, systems and rules shape each one.

    01 / 05LISTEN · DIALECT SPEECH ENGINE

    Understands how your customers actually speak.

    Our proprietary two-stage speech pipeline: the first stage transcribes the voice note, the second corrects it with a language model tuned on Gulf Arabic speech data (Saudi, Emirati, Qatari and Omani) and your own vocabulary, including product names, brands, branches and trade words. When a voice note is still unclear, the agent sends it to your staff with the audio attached.

    Word error rate, Gulf Arabic business voice notes[12%] pending benchmark
    Requests understood correctly[96%] pending benchmark
    Voice note · 0:05 · wholesaletranscribed
    0:05
    heardعندكم سيم؟ كم آخر سعر لـ ٢٠ لفة
    meaning"Do you have tie wire? Best price for 20 rolls?"
    requestPrice quote
    Dialectdetected: Omani
    SaudiEmiratiQatariOmani ✓
    Stage 2 · taught from your catalogue
    سيم = tie wireكم آخر = best priceلفة = rollGhalaRiyadh 2nd Industrial
    02 / 05KNOW · VECTOR RAG KNOWLEDGE BASE

    Answers only from your approved sources.

    We build retrieval (RAG) modules from your policies, price schedules, catalogues and service terms: each document is split, indexed as vector embeddings, and searched by meaning, in Arabic or English. Every answer cites the document it came from. If the sources don't cover a question, the agent says so and hands it to your team.

    Question · clinic"Can I see a female doctor on Thursday evening?"
    XLSDoctor roster · Novembersource 1
    PDFClinic hours and booking policysource 2
    DOCPublic holiday schedulechecked
    Dr. Sarah Al Balushi sees patients on Thursday from 4 pm to 9 pm. Would 6:30 suit you? [1][2]
    03 / 05ACT · ERP/CRM CONNECTORS AND CUSTOM AGENT TOOLS

    Custom tools that act in your ERP and CRM.

    We connect the agent to your ERP, CRM or clinic system and build it a set of custom agent tools for your workflows, each limited to the actions you approve: check stock, read an account price, book a slot, post an order, send a statement. Within your limits the agent completes these on its own, and every action is logged with what, when and why.

    Actions · retail orderShopify · WhatsApp
    READFound order #20417Shopify · customer verified by numberok
    READCourier status: out for deliveryCourier APIok
    WRITERepeat order placed for the same perfumeShopify · within limitsposted
    WRITERefund for damaged bottleabove your limitto staff
    04 / 05GOVERN · POLICY CHECK IN CODE

    Your commercial limits, enforced outside the AI.

    Your rules are written as a policy contract and checked in code before every action. Inside those limits the agent acts on its own. A person is brought in only when a request is outside your limits, when the agent can't serve it, or when it can't understand the message, and the right person gets it in the supervisor console.

    policy-contract.yamlsigned off by you
    discount:max_without_approval: 5%approver: commercial_managercredit:max_days: customer_limitclinic:medical_advice: never # bookings and admin onlyunclear_message: send_to_staff
    Above your limit12% discount on 200 bags
    ApproveCounter at 8%
    05 / 05IMPROVE · TESTED AND MANAGED

    Run by us. Approved by you.

    We monitor every conversation, trace what went wrong and propose fixes. Each change is re-tested against your own messages before release, and nothing ships without your approval. Prices, products and policies stay current.

    Weekly review · week 46monitoring
    fixed"Rebar 12 mm" added to the vocabularyre-tested
    fixedFriday delivery answer now reads the holiday schedulere-tested
    for youAllow repeat orders under OMR 50 without a staff check?needs approval
    Architecture

    What happens between the message and the reply.

    Every request runs the same path. The policy check sits in code before any action, and anything outside your limits goes to a person first.

    أبغى ٥٠ كيس أسمنت · voice note, 0:07

    Message

    Voice note or text arrives on WhatsApp.

    ✓
    Received

    Listen

    Transcribed and corrected for Gulf dialect.

    ✓
    Saudi Arabic

    Understand

    Request read; your stock, prices and policies checked.

    ✓
    Order · 50 bags

    Policy check

    Your limits, enforced in code before any action.

    ✓
    Within limits

    Act and reply

    Order posted in your ERP; confirmation sent.

    ✓
    SO-10482 posted
    Where it runs: in your region or in your own cloud accountPrivacy: personal details removed before any external AI modelAudit: every step logged with what, when and why
    Control

    Rules that hold in every conversation.

    An instruction in a prompt can be argued with. A rule in code can't. These controls sit around the agent, not inside it.

    01

    Policy contracts

    Discounts, credit terms, delivery promises and refunds follow written limits you sign off.

    02

    Human in the loop

    Exceptions go to the right person with the full conversation, for one-click approval or takeover.

    03

    Stays on topic

    The agent only discusses your business. Clinic agents handle bookings and admin, never medical advice.

    04

    Impersonation defence

    "The boss already approved 20% off" is caught and sent to your staff, not acted on.

    05

    Identity before account actions

    Statements, balances and results are shared only after the customer is verified by registered number or one-time code.

    06

    Personal data handled carefully

    Personal details are masked where staff and engineers don't need them, and kept only as long as you decide.

    Release gate

    Nothing reaches your customers until it passes your test suite.

    1. A test suite from your own messages

      With your permission, we build a test set from your WhatsApp history and voice notes, and label the right outcome for each.

    2. Every request, action and handoff checked

      Each test ends as passed, sent to staff, or blocked by the policy check. Wrong answers without a flag are the number we drive to zero.

    3. A release threshold you agree

      No change reaches production below a [99.8%] pending benchmark regression pass rate on your suite. A version that misses it doesn't ship.

    4. Versions compared side by side

      Every change runs the same suite, so you see exactly what improved before you approve it.

    Proactive

    The agent messages first, when it helps your customer.

    Reminders, confirmations and follow-ups on Meta's approved message templates, only to customers who've agreed to hear from you.

    24 hours before

    Appointment confirmation

    Your cleaning with Dr. Al Habsi is tomorrow at 10:30.
    ConfirmReschedule
    6 months later

    Recall reminder

    It's been six months since your last check-up. Shall we book one?
    Book
    Usual reorder day

    Reorder prompt

    You usually order 50 bags of cement on Sundays. Same again?
    Yes, sameChange
    Invoice due

    Payment reminder

    Invoice INV-3381 is due on Thursday. Here's your statement.
    View statement
    On dispatch

    Order on its way

    Your perfume order left our Doha warehouse and arrives this afternoon.
    Track
    Insights

    See what your customers are really asking for.

    What people ask, what was completed, why the rest went to staff, and what they wanted that you don't offer yet.

    Start here

    The Conversation Audit. See where your customers are waiting.

    Before anyone builds anything, we analyse your recent WhatsApp and phone enquiries and show you what an agent could take on.

    Enquiries analysedLast 90 days
    Arrive outside working hours31% example
    Routine and repeatable64% example
    You receiveA business process assessment and AI automation plan
    Get your Conversation Audit →
    Enquiries by day and hour · example datahover a cell
    Staffed hoursNo one on shift
    01

    Business process assessment

    Your Conversation Audit plus a review of the processes behind it: orders, bookings, approvals and the systems involved.

    02

    AI automation plan

    A written plan: which processes to automate first, the tools the agent needs, and its policy limits.

    03

    Build

    Speech tuning on your products, vector RAG modules, ERP/CRM connectors, custom agent tools and your policy contract.

    04

    Release gate

    The agent runs your test suite. You see the results first.

    05

    Go live

    On-site or remote onboarding with our engineers. One workflow first, staff trained on the console.

    06

    Managed

    Monitored and improved every week by the engineers who built it.

    Integrations

    Connects to the systems you already run.

    Your system stays the record of what happened. If it has no way to connect, we cover the options in the automation plan.

    Official platform · keep your number
    ZohoCRM · Books · Inventory
    OdooSales · Inventory · Invoicing
    SAPBusiness OneOrders · stock · pricing
    Microsoft Dynamics 365Business Central · orders · customers
    TallyPrimeTallyPrime, via a local connector
    Clinic systemsCalendars · patient records
    ShopifyOrders · carts · customers
    Estimate

    What could an agent take off your team's plate?

    Pick your industry for illustrative starting values, then adjust them to match your business.

    Estimated staff time freed
    hours a month
    full-time staff equivalent
    after-hours messages answered right away

    Starting values are illustrative, not benchmarks, and this is not a quote. Your Conversation Audit gives you the real figures.

    Data and security

    Answers for your IT and legal teams.

    Agreed in writing before the pilot starts.

    Hosting

    Processed and stored in your region

    Voice notes are transcribed and stored in data centres in your region, or in your own cloud account. Personal details are removed before any text reaches an external AI model.

    Control

    Limits enforced in code

    The agent cannot change price lists, grant credit beyond your limits or share account data without verification.

    Privacy

    Designed to support Gulf data-protection laws

    Personal data is masked where it isn't needed, retained only as long as you decide, and covered by a processing agreement.

    Channel

    Official WhatsApp Business Platform

    Built on Meta's official platform, on your existing number. No unofficial connectors that put your number at risk.

    Engineers

    One team, start to finish

    The engineers who build your agent onboard your team and support it after go-live. No hand-offs to a separate support desk.

    Service levels

    Written into your contract

    [99.9% uptime] and [15-minute response] for critical issues pending decision, with monitoring reports every month.

    Questions

    About the agent.

    Does the agent act on its own?

    Yes. It is autonomous within your policy contract: it completes the request and posts the transaction in your systems without waiting for staff. A person is brought in only when a request is outside your limits, when the agent can't serve it, or when it can't understand the message. You decide where the limits sit, and you can widen them as trust builds.

    How accurate is it with Gulf Arabic voice notes?

    We measure it on your own customers' messages and show you the results at the release gate: how many passed, how many went to staff, and how many were wrong without a flag. We don't quote a general accuracy figure, because it depends on your customers and your products.

    Can it make up a price or a stock level?

    Prices and stock come from your systems at the moment of the request, not from the AI's memory, and the policy check blocks any action outside your limits. If the agent can't find an answer in your approved sources, it hands the conversation to your team.

    What do you need from us to build the knowledge base?

    Your current price lists, catalogue, delivery and payment terms, and the answers your team gives most often. We organise them, and you approve what the agent may use.

    Which AI models do you use?

    We choose the model for each task, such as speech, understanding and answering, and can run models in your region or in your own cloud where your data rules require it. You're not locked into one AI vendor.

    What does "fully managed" include?

    Monitoring every conversation, a weekly review of anything that went wrong, fixes re-tested against your messages, updates when your prices, products or policies change, and support from the engineers who built it.

    Start here

    Start with a Conversation Audit.
    See where your customers wait.

    We'll analyse your recent enquiries and show you what an agent could take on, before you commit to anything else.