Tick if this was you

Six invoices appear beside what an "AI extractor" pulled from them. They look fine at a glance. Your job: find the ones where the numbers don't hold together — line items that don't sum to the total, a tax amount that doesn't match the rate, a date in the future. Each catch turns a field amber and asks you to write the rule that would catch it automatically ("sum of lines = subtotal, within a rounding tolerance"). By the end, you've written your first validation rulebook without noticing. The invoices are invented; nothing is uploaded.

Decide how many hours a week you can give: 5, 10, 15 or 20. Spread them across four blocks: learning one tool, building validation rules, running your sample set, and outreach. More hours means a shorter path to your proof; fewer hours is fine too. It's a planning aid, not a promise.

Card 1 — The reviewer-analyst (least code) You don't build the extractor; you run and check it. Design the review queue, set confidence thresholds with the team, track accuracy by field and vendor, and report weekly. Best if you liked reconciliation and reporting. Tools: a no-code document tool or a cloud demo console, Sheets or Excel, a dashboard.

Card 2 — The low-code builder (some setup) Connect an extraction service to a workflow: invoices arrive by email, get extracted, pass validation, and land in a sheet or accounting export. Exceptions go to a human. Best if you enjoyed the reporting automation idea. Tools: a cloud extraction API (application programming interface) plus n8n, Make or Power Automate.

Card 3 — The accuracy owner (SQL-friendly) Own the measurement: a labelled test set, field-level accuracy, error categories, and before/after comparisons when the vendor model changes. Best if you like analysis more than operations. Tools: SQL or Sheets, a labelled sample set, a simple dashboard.

All three need the same thing first: a clear idea of what a correct extraction looks like.

• Google Document AI — the pricing page lists Enterprise OCR (optical character recognition) with the first 1,000 pages a month free, then a paid tier; the Invoice parser is paid (pricing). Prices can change; recheck before quoting.
• Azure AI Document Intelligence — Microsoft's equivalent (pricing page). We haven't confirmed its free-tier details, so check the page yourself.
• Docling — open source (MIT) for turning PDFs into structured text and tables (GitHub). Needs some Python; useful if you want to try Card 2 without a cloud account.

Practise only with public sample invoices or ones you make up. Never upload real customer or employer documents to a free tool.

1. Line items sum to the subtotal (allow a small rounding gap).
2. Subtotal + tax = total.
3. Tax rate is one of the expected rates.
4. GSTIN or tax ID matches the expected format.
5. Invoice date isn't in the future or years old.
6. Vendor name matches your vendor master list.
7. Invoice number isn't a duplicate.
8. Any field below the confidence threshold goes to a human.

These eight rules are your portfolio's backbone. They show judgement, which is what employers hire for.

1. Collect 50 sample invoices (public samples or made-up ones in several layouts).
2. Run them through one extraction tool.
3. Apply your eight rules in Sheets or a small script.
4. Hand-check a sample and record field-level accuracy.
5. Write a one-page report: accuracy by field, the top three error types, and what you'd send to human review.

1. Days 1–2: Gather samples; write the rules.
2. Days 3–5: Open one tool and run the first 20.
3. Days 6–8: Run the rest; apply rules.
4. Days 9–11: Hand-check and measure.
5. Days 12–14: Report, headline update, and five messages to accounts teams or CA firms.

You've typed numbers from PDFs, scanned bills and email attachments into Excel more times than you can count. Document AI automation does that extraction with software, and you check and route the results. You'll start with the document types you used to key in by hand.

Assumes 2–3 hours a day alongside job applications, a laptop and a phone, and free plans only. Every step has a smaller version and a way to pause. Use them. It runs in three blocks of 30 days, each ending with a signal that you're on track. A signal is not a promise of a job or income.

What AI changed in your old tasks

Your first 7 days

Days 1–30: A checked extraction flow

Signal by day 30: An accounts or ops person says it would save them time.

Days 31–60: Second document type

Signal by day 60: A screening call or a pilot request.

Days 61–90: Pilot

Signal by day 90: A paid pilot or a second interview.

Job-hunt moves that matter for this route

• Search 'IDP', 'document automation', 'Power Automate', 'RPA (robotic process automation)' on Naukri.
• CA firms and logistics companies are realistic first users.
• Use consent and redaction; never store ID documents you don't need.

One search that fits your background

The full set of search strings for this move is on the route page. They work on LinkedIn, Naukri and foundit; Instahyre and Cutshort are worth a weekly look too. Add your old job title to any of them with AND, so recruiters who want both skills find you first.

The two worries we hear most about this move

Do I need to know machine learning? No. You'll use existing document AI services and spend most of your time on what analysts do well: checking extracted values, catching errors and building exception reports. Totals that don't add up are your best friend. Are these roles real in India? Document processing (invoices, KYC, claims) is common in Indian BPOs, banks and insurers, though job titles vary. Search for 'IDP', 'document processing' and 'intelligent automation' as well as 'document AI'.

Other worries, such as age, a career gap, no degree, English, family pressure and savings, are covered on the route page and the role page.

We haven't verified pay data for document AI roles or pilots. Compare live listings and ask people already doing the work.

Fake "PDF to Excel" jobs often ask for a registration deposit, then reject your work. Never share OTPs or UPI PINs. Official domains only. Report in India at cybercrime.gov.in or 1930; in the US at ReportFraud.ftc.gov.

Isn't this just data entry? No. Data entry types the fields; this route designs the checks and measures accuracy.

Do I need GST knowledge? Basic familiarity helps for Indian invoices. This isn't tax advice; for real tax decisions, a CA should advise.

Take the quiz · Back to Data analyst · Route: Document AI automation · Also: Reporting automation