Assistants can now read a client’s request, pick the e-Form and hold a conversation to fill it. The interesting question for a CS practice is where that help should stop.
When we show AutoROC to Company Secretaries, the first question is rarely “what can it do?”. It is “what will it do without me?”. That is the right instinct. ROC filings carry a professional’s name and a company’s legal record, and a mistake can mean a rejection, additional fees or an awkward conversation with a client. So we want to be explicit about the lines we drew.
Most of the time in a filing is not professional judgement. It is classifying what the client wants, finding facts already on record, asking for what is missing and chasing it, and checking formats. These are the steps we hand to the assistant:
None of these steps commits anything to the registry. They produce a draft and a complete file for a professional to look at.
| The assistant should not… | Why |
|---|---|
| File on the MCA portal | Filing is a professional act; a person reviews and submits on MCA V3 |
| Decide that a form is legally correct | Applicability, timing and consequences are matters of professional judgement |
| Invent a missing value | If a field is unknown, it should ask, not guess a plausible answer |
| Override a failed check silently | A DIN or date that fails validation should block, not be waved through |
| Mark a filing complete without an SRN | The record should show what was actually filed and when |
These are the reasons the product stops where it does. The professional reviews the pre-filled e-Form, prints it to PDF or exports JSON for the MCA V3 portal, files it, and records the SRN. The ticket moves to done after ROC approval, not before.
We also think a practice should be able to choose how much language-model behaviour it wants. AutoROC has two modes that write to the same fields and run the same checks. The rule-based mode classifies by keywords and asks every missing field in order, which is predictable and easy to audit. The language-model mode reads the client’s words and documents, holds a natural conversation and records fields through structured tool calls. Some firms will prefer the conversation for clients who write long, messy requests. Others will prefer rules for the forms they file most often. Both end at the same professional review.
Classification is the step most likely to be wrong in an interesting way. A client who says “we took a loan” may need a charge form, or may not, depending on the facts. Showing a confidence level next to the suggested form is a small detail with a large effect: it tells the reviewer where to slow down. A request the assistant cannot place should become a general request, with details still collected, rather than being forced into the nearest form.
The time an assistant saves should be spent on review, not removed from it. If associates stop typing and chasing, they have more attention for checking applicability, timelines and documents. That is the trade we think is worth making. Statutory questions, from due dates to consequences of delay, remain for the professional and the client’s adviser to confirm.
For how clients feed requests in without endless follow-ups, see our playbook on stopping client chasing. A side-by-side view is in manual vs assisted ROC filing, and more is in the AutoROC guides.
In AutoROC it does not. The assistant classifies the request, pre-fills, collects missing fields and checks formats; a CS or CA reviews the e-Form, files it on MCA V3 and records the SRN.
It shows a confidence level with the suggested form, and anything outside the catalogue is raised as a general request so the professional decides.
Yes. A rule-based mode classifies by keywords and asks every missing field in order; a language-model mode holds a conversation. Both run the same checks.
See it on your own data. AutoROC — MCA e-Forms, pre-filled. Book a 30-minute working session with an engineer.
