Case Studies — ElevateIQ

Case study 01 · Non-profit

Golden Mouth — audit evidence from cheques, statements and chat

Situation A charitable trust faced its first statutory audit, and its chartered accountant needed transaction-level proof across two financial years.

Task Every rupee had to tie back to evidence — but the evidence lived in bank statements, loose cheque photographs and years of WhatsApp threads, none of it structured.

Action We built a structured transaction ledger from the statements, then used AI image extraction and categorization to read each receipt and match it to its bank entry, purpose and accounting category.

Result The trust reduced the time to prepare tax audit reports by roughly 80%, with a reconciliation that closes to zero and every entry carrying its own proof.

treasury.internal / dashboard — FY 2025–26

Total credits₹ 82,14,500

Total debits₹ 62,41,256

Closing balance₹ 19,73,244

Reconciliation diff.₹ 0.00

Date	Payee / Payer	Amount	Category	Evidence	Status
12 Feb	Fencing contractor	₹ 1,84,000	2B Architecture	Cheque no. 004182	Supported
03 Feb	Field labour — irrigation	₹ 46,200	2C Operating	WhatsApp thread + voucher	Mgmt evidence
28 Jan	Fixed deposit interest	₹ 38,290	Interest income	Bank advice	Supported
19 Jan	Transport — saplings	₹ 12,750	2A Phase-0	—	Pending

Date-range report → PDF Export to Excel Admin edit & correct Year-to-date / since inception

Illustrative mockup. Figures and parties are representative, not a real client record.

Case study 02 · Reinsurance

Governed-AI treaty renewals

Situation A reinsurance broker assembled every treaty renewal by hand, rebuilding the same figures across a cedent letter and a slip for each reinsurer on the panel.

Task Source figures arrived in emails, slips and bordereaux that openly contradicted each other, and every number in the output had to be defensible to a regulator years later.

Action We built a governed pipeline: AI extracts and flags conflicts, a human resolves only the exceptions, and a deterministic engine generates every document from version-locked rules.

Result One renewal letter and four reinsurer slips produced per treaty with five of six fields auto-validated — each hash-sealed, field-level traceable, and byte-identical on re-run.

renewals.internal / run / TRY-2024-001

source→ AI reads→ reconcile→ GATE · confirm data→ rules (locked v3)→ generate→ GATE · approve→ archive

5 of 6 fields auto-validated 1 exception routed to a human Determinism check passed

Exception — attachment point

Sources disagree

Email — “attaching at 5m”source-email.txtConfirmed

Slip PDF — “CHF 4,000,000”slip.pdfRejected

Generated · 1 renewal letter + 4 reinsurer slips

sha256:9f3c…a1d7

Reinsurer 1 · line 40%limit × share20,000,000

Reinsurer 2 · line 25%limit × share12,500,000

Reinsurer 3 · line 20%limit × share10,000,000

Reinsurer 4 · line 15%limit × share7,500,000

Append-only audit log Field-level provenance Version-locked rules Mandatory clauses verbatim

Illustrative mockup. Counterparties are anonymized and figures are representative.

Case study 03 · Insurance brokerage

Smart carrier matching

Situation An insurance brokerage matched every new client to carriers by hand, working through appetite guidelines binder by binder.

Task It took 30 to 60 minutes per client, only senior brokers knew which carriers wanted which industries, and small accounts were left unserved because the commission never justified the time.

Action We built an engine that parses a spoken or typed business description into risk attributes and scores every carrier on industry, state, headcount and revenue — exposed over both a web app and WhatsApp.

Result A ranked, explainable shortlist in under two seconds instead of an hour — making small accounts profitable to serve and putting carrier fit in reach of every broker, not just the senior ones.

Carrier Match Bot
online

Plumber with 50 employees in Florida, $2M revenue

Parsed: Plumbing · FL · 50 employees · $2M revenue

1. Carrier A — 100%
2. Carrier B — 86%
3. Carrier C — 84%

Carrier A: in-appetite industry, licensed in FL, headcount within band.
Tap for full breakdown.

match.internal / results

Voice or text < 2 seconds

1 Carrier A 100%

2 Carrier B 86%

3 Carrier C 84%

4 Carrier D 61%

Industry 40% State 25% Employees 20% Revenue 15%

One engine, two channels Transparent scoring Add carriers without rewriting logic

Illustrative mockup. Carrier names are anonymized; scores are representative.

Three problems, solved end to end.

Golden Mouth — audit evidence from cheques, statements and chat

Governed-AI treaty renewals

Exception — attachment point

Generated · 1 renewal letter + 4 reinsurer slips

Smart carrier matching

Have a process that looks like one of these?