Key takeaways
- AI is very good at the repetitive part of tax recovery: collecting documents, reading them and matching them to returns.
- It should not decide eligibility, sign claims or talk to the department. Those are judgment calls with legal consequences.
- The best results come from a clear hand-off: machines clear the volume, specialists review every exception.
Most tax recovery work is not difficult. It is large. A mid-size company can receive tens of thousands of airline invoices, vendor invoices and TDS entries in a year. Each one has to be found, read and checked against a return. Done by hand, the work takes months, so it is usually done once a year, if at all.
What machines do well
Software can now collect documents from supplier portals and mailboxes, read the key fields from a PDF, and match each document to the right line in GSTR-2B or Form 26AS. It does not get tired at invoice 30,000, and it applies the same rule to every record. That is exactly the work that used to stop companies from recovering small amounts at scale.
Who does what in a recovery engagement
| Step | Done by | Why |
|---|---|---|
| Collecting invoices and statements | Systems | High volume, rule-based, repetitive |
| Reading fields from documents | Systems, with sample checks | Fast and consistent; errors caught by review |
| Matching to returns | Systems | Thousands of comparisons a day |
| Deciding eligibility | Specialists | Needs law, facts and context |
| Approving a claim | Client, on specialist advice | A legal position the company owns |
| Replying to the department | Specialists | Argument and judgment, not data |
Source: TraCarta view of how recovery work divides.
Where machines should stop
Whether a credit is eligible depends on facts a document does not show: the purpose of a trip, whether a cabin class qualifies, whether a supply is really an export. A wrong call does not just lose money; it creates a position the company must defend. So a specialist looks at every exception, and nothing reaches a return or the department without a person signing it off.
The machine finds the gap. A person decides whether it is money you can claim.
TraCartaWhy this changes the economics
When the volume work is cheap, small claims become worth pursuing and monthly monitoring becomes affordable. Recovery stops being a one-off clean-up and becomes a routine, with a specialist spending time only where judgment is needed.
What to ask before you start
- Which steps are automated?
You should get a clear list, step by step.
- Who reviews exceptions?
A named person, with the right qualifications.
- What do you sign off?
Nothing should be filed without your approval.
- Can you see the evidence?
Every number should trace back to a document.
This is how our engagements run. Talk to us to see it on your own data.
Managing Partner of TraCarta. He started the firm in 2018 to recover airline GST credit for corporate clients and leads its three recovery practices.
View profileGeneral information only, not tax advice. Check the current law and your facts before acting.
