INSIGHTS / TELEMATICS OPERATIONS
What is Agentra Telematics? One support case, step by step
In short: Agentra Telematics is an AI front desk for GPS tracking providers and systems integrators. It answers fleet clients on WhatsApp and checks their vehicles on the provider's Wialon account with read-only access. For an offline unit, it gives a "most likely" explanation from the message history. When a case needs the team, it goes on a ticket. The ticket goes to the client account's owner, or to the shared team group when no owner is set. The provider's staff decide when the case is closed.
A GPS tracking provider that puts AI on its support number has to decide two things. The first is what the AI may read and do on the tracking platform. The second is who owns the case after the AI has replied. Much of the work in a support case belongs to the second decision. That work includes the record of checks, the technician brief and the message that tells the client the outcome.
This page is for owners, heads of support and operations managers at tracking providers and systems integrators. It explains what Agentra Telematics is, and it follows one offline-vehicle case from the first WhatsApp message to the client being told the outcome. At each step, we show what goes onto the ticket. The Agentra Telematics solution page gives the product overview.
Why the provider desk is getting busier
Fleet tracking keeps growing in both of our markets. Berg Insight counts 4.0 million active fleet management systems in Southeast Asia in 2025, and it forecasts almost 6.4 million by 2029. In the GCC countries and Turkey, it counts about 6.4 million units at the end of 2025 and forecasts 12.5 million by 2030.
Each new unit is another vehicle that a client can ask about, and most of those questions arrive on WhatsApp. In Saudi Arabia, the regulator's 2025 internet report puts WhatsApp use at 92.7 percent of internet users.
Source: Berg Insight, Fleet Management in Southeast Asia, 2nd edition (2026)
Show the data as a table
| Item | Value |
|---|---|
| 2024 (about) | 3.6M |
| 2025 | 4.0M |
| 2029 (forecast) | 6.4M |
The cost of each answer also matters. Gartner's 2024 benchmark puts the median cost per contact at US$13.50 for assisted channels such as phone, chat and email, and US$1.84 for self-service. A desk that answers routine status questions from platform data can keep staff time for the cases that need a person.
How the AI agents work
An AI agent is software that uses a language model to decide which tools to call to finish a task. Anthropic's engineering guide Building effective agents recommends simple, composable patterns, careful tool design and checkpoints where the agent pauses for human feedback. OpenAI's practical guide to building agents recommends rating each tool by risk, and passing high-risk actions to a person. In customer service, that person is a human agent.
We follow the same approach. The desk has two assistants. A customer assistant talks with the client. An internal assistant opens and updates the ticket and briefs the right person. It is also designed to pass the team's update back to the client. The internal assistant gets only the tools that support work needs.
Wialon is the only tracking platform we connect to today. Each customer chat is linked to one account in the provider's account tree, and the chat sees only the vehicles under that account. An unlinked chat sees no vehicles. Our guide on how to identify the right account and vehicle from a WhatsApp message explains the identification order.
When a case needs the team, it goes on a ticket. The ticket goes to the client account's owner, or to the shared team group when no owner is set.
The customer assistant's tools are read-only lookups on the provider's Wialon account. Outgoing messages are screened, so that internal notes do not reach the client. Anything that needs a decision goes to a person on a ticket.
Most companies are still careful with AI agents, and the numbers show why. In a January 2025 Gartner webinar poll of 3,412 attendees, 19 percent had made significant investments in agentic AI and 42 percent had made conservative ones.
Source: Gartner press release, 25 June 2025
Show the data as a table
| Item | Value |
|---|---|
| Conservative investment | 42% |
| Significant investment | 19% |
| Wait and see, or unsure | 31% |
| No investment | 8% |
Gartner forecasts that by 2029 agentic AI will resolve 80 percent of common customer service issues without human intervention. It also predicts that over 40 percent of agentic AI projects will be cancelled by the end of 2027, because of rising costs, unclear business value or weak risk controls. Customers want a person within reach. In a 2026 Gartner survey of 3,566 customers, 87 percent said a company that uses generative AI for service must offer a way to reach a human agent. That is why we scope each pilot on a few case types, and why the provider's staff close every case.
One offline vehicle, step by step
The case below is illustrative. The plate, serial, times, readings and places are invented.
- 1Identify the vehicleCustomer assistantFinds the plate or serial under the one account linked to this chat.
- 2Check and explainCustomer assistantRuns a read-only Wialon check and gives a 'most likely' explanation from the message history.
- 3Open the ticketInternal assistantRecords the client's problem, each check and its result, and what the client was told. Then briefs the account owner.
- 4Decide the next stepAccount ownerAsks the client what the evidence leaves open and books a technician if needed.
- 5Confirm the causeTechnicianFinds and fixes the physical fault, and the unit reports again.
- 6Resolve the ticketStaffResolve from their own chat and note the closure evidence that the provider agreed for this case type.
- 7Tell the clientCustomer assistantIs designed to pass the staff outcome to the client in the same WhatsApp chat.
Step 1: The client writes
On Monday at 08:05, a fleet manager writes on WhatsApp: "Truck DEMO 311 has not updated since Saturday." The customer assistant looks for DEMO 311 under the account that this chat is linked to. It finds one match, with device serial 000000017. If the plate were outside that account, the assistant would give the same answer as for a plate that does not exist.
Carried onto the ticket at Step 4: the client's problem, the chat's linked account, the plate and the device serial.
Step 2: The read-only check
The assistant checks the unit on Wialon. The check returns the latest readings that the device and its configured sensors report:
- Last message: Saturday 23:12. Last position: Saturday 18:43.
- External power: 0.0 V. Internal battery: 3.6 V.
- Ignition: off.
- GSM signal: good. Satellites: 9 in the last position report.
- Place of the last position: Depot A, a client yard.
A reading that the device does not report shows as "no reading". If the lookup fails, the assistant tells the client that it could not check the vehicle, and the case goes to the team.
Carried onto the ticket at Step 4: the check and each reading it returned.
Step 3: The offline timeline
The unit is Offline. Agentra judges Offline on the last message, not on the last position. Checks, reports and the daily offline list use the same rule. The client and staff therefore see the same status for the same unit.
The system builds a short timeline from the unit's message history:
| Time (Saturday) | What the history shows |
|---|---|
| 18:31 | Ignition off at Depot A. External power 25.2 V. |
| 18:43 | Last position. External power 0.0 V. Internal battery 4.1 V. |
| 19:12 to 23:12 | Heartbeats every 30 minutes with no new position. Internal battery falls to 3.6 V. |
| After 23:12 | No messages. |
The assistant tells the client the most likely explanation and cites the evidence. External power was lost within 12 minutes after the truck was parked. The device then ran on its internal battery until it stopped reporting. If the history did not support a verdict, the assistant would say that it cannot tell.
Carried onto the ticket at Step 4: the timeline, the "most likely" explanation and what the client was told.
Step 4: The ticket and the owner
The internal assistant opens the ticket. It is designed to send the case to the person who owns this client account. When no owner is set, the case goes to the shared team group. The customer assistant tells the client that the team has the case. It is instructed not to promise a fix or a visit before staff confirm one.
Recorded on the ticket: the client's problem, each check and its result, what the client was told, and the owner.
Step 5: The owner decides
The evidence does not show why the power was lost. A cut wire, a blown fuse, a disconnected truck battery and a battery isolator switch can all give the same history. The account owner asks the client whether the truck has been used since Saturday or the isolator was turned off. The client says the truck is still in the yard. The owner books a technician visit. Our guide on what the desk must establish before sending a technician covers this decision in detail.
Added to the ticket by the owner: the question, the client's answer and the decision to send a technician.
Step 6: The technician confirms the cause
The technician finds a loose power connector behind the dashboard and fixes it. The unit sends a new message with external power at 25.0 V. A person confirms the physical cause. Until then, the assistant's explanation stays a "most likely" verdict.
Added to the ticket by staff: the technician's finding, the action taken and the first new message from the unit.
Step 7: Staff close the case and the client is told
The owner resolves the ticket from their own chat. The assistant confirms to the owner that the ticket is resolved. The assistant is then designed to pass the outcome to the client.
The closure evidence for this case type is the provider's own rule. In this example, the rule is a new message with external power present. Agentra does not check that rule automatically. Staff note the closure evidence when they resolve the ticket.
Recorded on the ticket: who resolved it and the outcome.
What the desk must record before closing an offline-vehicle case
Use this list before staff resolve an offline-vehicle ticket. Each item is something the ticket should show, so that the owner, the technician and the client work from the same facts.
- The client account and the vehicle, matched by plate or device serial within the account linked to the chat.
- The time of the last message and the time of the last position, recorded as two separate values.
- External power and internal battery readings before the silence, or 'no reading' where the device does not report them.
- The ignition state and the time of the last ignition change.
- The heartbeat pattern between the last position and the last message.
- The place of the last position, and GSM signal and satellites from the last reports.
- The 'most likely' explanation sent to the client with its evidence, or the statement that the evidence was not enough.
- The questions the account owner asked the client, and the client's answers.
- The cause confirmed by a technician and the action taken, or the remote step agreed with the client.
- The closure evidence agreed for this case type, the person who resolved the ticket, and what the client was told.
Our article on what an AI agent should accept as evidence of completed work explains how to define proof that a case is finished.
Where a platform AI assistant fits
Some tracking platforms now include their own AI assistants. These tools help staff with device questions, device configuration and platform features. They work in the platform's own help centre, website or panel. The provider's fleet clients still send their messages to the provider, and someone at the provider still owns each case.
| Support task | Platform AI assistant | Provider desk |
|---|---|---|
| Answer a status question about one vehicle | Depends on the platform. Some answer from documentation only. Some can read the account's device data. | Answers the client in the chat, with read-only checks |
| Answer a history question (trips, stops, sensors) | Depends on the platform. Some answer from documentation only. Some can read the account's device data. | Reads the history for a date range and reports what the data shows |
Tasks that stay with the provider's desk
- Match a WhatsApp message to the client account and the vehicle.
- Keep the client's problem, each check and its result, and what the client was told on the ticket.
- Give the case an owner: the account owner, or the shared team group when no owner is set.
- Brief a technician with the readings, the timeline and the client's answers.
- Close the case against the provider's closure rule, and tell the client the outcome.
Some platforms also offer AI agents that connect to a ticketing system. The provider still decides who owns each case and what evidence closes it.
Our article on telematics support automation covers the design rules for automated WhatsApp replies. Our article on what an AI agent should be allowed to change on a device explains why we start from read-only access.
What is in the product today
| Capability | Status |
|---|---|
| Read-only Wialon checks by plate or device serial, with the latest readings | Shipped |
| Offline timelines from message history, with a "most likely" explanation | Shipped |
| One rule for Offline, Stopped, Idling and Moving | Shipped |
| Trips, stops and sensor history in the chat | Shipped |
| Fleet status and trip reports as PDF files | Shipped |
| Daily offline list for each account owner in the portal (no automatic client messages) | Shipped |
| Tickets with a timeline, a ticket board by owner and a weekly report per owner | Shipped |
| Read-only provider portal with conversations and 30-day figures | Shipped |
| Screening of outgoing messages, so that internal notes stay internal | Shipped |
| Meta's official WhatsApp Business Platform, connected from the portal | Supported; set up with each provider |
| Replies in English, Arabic or Malay | Set up per provider; we test language behaviour in each implementation |
| Technician job offers and acceptance | Scoped implementation work |
| Escalation paths and priorities | Agreed with each provider |
| Closure evidence for each case type | The provider's rule, agreed before the pilot |
| Platforms other than Wialon, and ticketing systems | Assessed during scoping |
The Agentra Telematics solution page for tracking providers shows the full workflow that we scope with each provider.
What a provider prepares for a pilot
A pilot covers a few common case types, for example an offline unit or a request for a trip report. Our guide to choosing which device-support tasks to automate first includes a workbook that helps a team choose them.
- The provider's manager connects the existing Wialon account from the portal. Agentra does not create or change anything on Wialon.
- The provider tells us which client account each customer chat belongs to. We link each chat to that account.
- The provider names the owner of each client account. Accounts with no owner use the shared team group.
- The provider's team describes its common cases, and we write test conversations from them before the assistant goes live.
- The provider decides the success measures and the closure evidence for each case type before the pilot starts.
At Agentra, we turn a provider's support rules into the checks, records and handovers that the desk uses.
Questions providers ask
Which tracking platforms does Agentra Telematics work with?
Wialon is the only tracking platform we connect to today. We assess other telematics platforms and ticketing systems during scoping, because data access depends on each platform's API and the provider's account permissions.
Does Agentra Telematics change data on our Wialon account?
No. The provider's manager connects the existing Wialon account from the portal, and Agentra does not create or change anything on Wialon. Vehicle checks and history questions use read-only access.
Who closes a support case?
The provider's staff decide when a case is closed. A staff member resolves the ticket from their own chat, and the assistant is designed to pass the outcome to the client. Each case type has a closure rule that the provider agrees with us before the pilot. Agentra does not check that rule automatically.
Does the assistant decide why a tracker is offline?
No. The assistant gives a 'most likely' explanation from the unit's message history, and it cites the readings it used. When the history does not support a verdict, it says that it cannot tell. A technician or the provider's staff confirm the physical cause.
How is Agentra Telematics different from a tracking platform's AI assistant?
A platform AI assistant helps staff work with the platform. Depending on the platform, it answers from documentation or reads device data. Agentra Telematics works on the provider's own support channel. It matches each message to a client account and a vehicle, puts the case on a ticket when the team is needed, and sends it to an owner. The provider's staff decide when the case is closed.
Which languages does the assistant reply in?
The assistant can be set up to reply in English, Arabic or Malay. We test language behaviour in each implementation before the assistant goes live.
Sources
- Berg Insight, Fleet Management in Southeast Asia, 2nd edition (brochure, 2026)
- Berg Insight forecast for the GCC and Turkey, reported by TradeArabia (2026)
- Saudi Internet Report 2025 (CST), reported by Arab News
- Gartner, Benchmarks to Assess Your Customer Service Costs (2024)
- Gartner, agentic AI to resolve 80% of common customer service issues by 2029 (2025)
- Gartner, over 40% of agentic AI projects to be cancelled by end of 2027 (2025)
- Gartner, 87% of customers want access to a human agent (2026)
- Anthropic, Building effective agents (2024)
- OpenAI, A practical guide to building agents (2025)
To map one of your support cases with us, use the WhatsApp button on the Agentra Telematics page or write to info@agentra.my. Follow Agentra AI on LinkedIn.