Dispatcher AI works through the customer emails in your Dynamics 365 dispatch queue the way your team does. It reads the email and its attachments, identifies the category, finds the customer and the contract, and opens the service case. When it doesn't have enough to go on, it hands the email back to your team with the reason.
Dispatcher AI is software that takes customer emails from your Dynamics 365 dispatch queue and does what a dispatcher does: it reads the email and its attachments, works out what the customer is asking for, finds the customer and the contract, and opens the right service case.
The AI part is limited to reading. Every decision is made by code, against rules your team can see and change, and every email ends in one of three ways: the case is opened, opened with a flag for review, or handed back to your team with the reason.
If your customers send requests by email with forms and documents attached, and someone on your team reads, sorts and keys them into Dynamics 365 by hand, Dispatcher AI is built for that work.
Built first for pension and life operations on PensionSoft. Contribution changes, exits, transfers, payment method changes and letter requests arrive with signed request forms and ID documents, and every one has to land on the right contract.
Policy, account and claim requests that come in by email with forms and identity documents, where every decision has to be traceable to a rule.
A shared mailbox that feeds a Dynamics 365 queue, recurring request types, and categories you can describe in instructions. You switch on automation one category at a time.
Fewer emails to sort by hand, and a reason for every one that comes back.
Start from the AI's notes on the case instead of a blank screen.
One connector to Dynamics 365, your servers, your choice of model.
Every decision is logged with the rule behind it; national IDs never reach the logs.
A representative example from a pension provider's service queue: about five seconds from the queue to an open case, with the AI reading and the code deciding.
Hello, I'd like to raise the monthly contribution on my pension plan from 700 to 1,000 TRY. My signed request form and my ID are attached.
The model only reads the email and its attachments and pulls out the information needed. Searching the CRM, running the checks and saving the record are handled by Dispatcher AI's own code, so you can see which rule every decision rests on.
Reads the email text and every attachment in one call. Returns the likely categories with a confidence score, and each piece of information with its source. Its answer must fit a predefined JSON schema.
Looks up the customer and the contract in the CRM, runs the 13 checkpoints in order, decides with the 139-rule matching matrix and creates the record. The model can't call any tool.
Claims the email, so two people or two servers never work on the same one.
Opens PDFs, scans, photos, Word and Excel files and tells their type from the content.
Reads the text and every attachment together and sets signature and logo images aside.
Searches by national ID, then contract or policy number, name, and sender address.
Checks the confidence score, required data and documents, and any open request on the same subject.
Opens the case, attaches documents by type and writes its findings into a note.
deterministic rules and checkpoints. Once the model has finished reading, only code makes the decision. The rules are laid out further down this page.
It attaches the documents by type and starts the workflow in the CRM.
The case opens with a [VERIFY] note that tells your team what to check.
It returns the email to the queue with the reason and writes what it read into a note, so your team doesn't start from scratch.
If one email asks for a contribution change and an address update, it opens two cases. If one gets stuck, the other still goes through.
It doesn't guess whether a number belongs to a pension or a life product; it checks which product table holds it in the CRM.
If a case on the same subject is still open, it adds the email to that case, provided the sender is the customer, the payer or an authorized intermediary.
The case can be opened under an unknown customer, held in the queue, or retried over a few days. You choose per category.
In sensitive categories, an email whose sender can't be verified isn't processed automatically.
If the CRM or the model can't be reached, it retries three times, 15 minutes apart. When a business rule rejects a request, it doesn't retry.
When the model has finished reading, the decision passes to code. Every request goes through 13 checkpoints, and a 139-rule matrix scores the match. The same findings always lead to the same decision, and each decision is logged with the code of the rule it rests on.
The tier alone doesn't set the outcome. Tier, category and contract status form a seven-part rule key, and the rule that fits the key best is applied.
From an 18-screen panel you follow every step Dispatcher AI takes and change its rules. Instructions, 11 behavior settings and the matching matrix are kept with their version history. The Live Test screen shows the outcome without writing anything to the CRM.
Seven of the 18 screens. The data is anonymized for this page; the interface is the product as it ships.
How every email ended: cases opened, attached to an open case, returned or out of scope, with daily and weekly comparisons.
Demo data · product interface
The queue, the service cases and the BPM process stay as they are. Dispatcher AI works next to your dispatchers and takes work from the same queue, like any member of the team. Every component is installed on your own infrastructure.
Only the connector API talks to Dynamics 365, and the business rules sit there too. If the CRM changes, only this layer changes.
Dispatcher AI calls the model through one endpoint: a model in Microsoft Foundry, such as OpenAI's GPT or Anthropic's Claude, or any OpenAI-compatible endpoint. Switch it in the settings.
Several servers are installed and one of them does the work at a time. If that server stops, a standby takes over.
Dispatcher AI only reads the live queue and compares its decisions with your team's.
You switch on automation category by category. A category can't be switched on until its instruction is published.
5% of the cases opened automatically are reviewed against four criteria.
You change the threshold, the matrix and the rules from the panel, with no new release.
Tell us about your service queue and we'll reply within one business day.
Soon you'll be able to book a 30-minute Dispatcher AI walkthrough directly in our calendar.
