Auto-routing public grievances with AI

A citizen filing a complaint does not usually know, or care, which department is responsible for a broken streetlight versus a blocked drain versus a stray animal. They just want it fixed. When that complaint lands in a general inbox or a portal with the wrong category picked, it sits until someone manually reads and re-routes it — often the slowest step in the whole process.

Signs you need this

• A meaningful share of complaints in your system are miscategorised at the point of filing.
• Your staff spend real time each day just reading and forwarding complaints, not resolving them.
• Citizens re-file the same complaint because they cannot tell if the first one reached the right office.

What a good system does

• Reads the complaint text, in Tamil or English, and suggests the right department and category — the citizen still files in plain words.
• Flags severity, so a burst water pipe is not queued behind a routine pothole report.
• Learns from staff corrections, so repeated mistakes in routing get fixed rather than repeated.
• Keeps every original complaint and its routing decision visible for audit, especially for escalations.

What it costs you

The system itself is a smaller cost than getting your department categories and routing rules clearly defined first — many offices discover their own categories overlap or are inconsistently used once they try to automate against them. That cleanup is worth doing regardless of which tool you choose, and it usually surfaces problems in the manual process that were costing you time long before any automation was involved, quite apart from anything the AI itself adds to the picture once the categories are actually clean and consistently used across every counter and every shift.

Questions to ask any vendor

• Does it read Tamil complaints as accurately as English ones, not as an afterthought?
• What happens when a complaint genuinely fits two departments — does it pick one and log the ambiguity?
• Can a staff member correct a routing decision, and does the system learn from that correction?
• Is the original complaint always visible, or only the AI's summary of it?

How we can help

We build systems that read and tag text in Tamil and English at scale, using Sol and Thedal, our language and matching models, the same technology behind our district-level news tagging. See how we approach this for government and public bodies — AI for government — or talk to us.