AI analyst workspace, Excel copilot and API for finance | NOVA

Product

One engine. Three places to put it.

NOVA connects to the sources you already pay for and the files you already own — filings, terminals, SharePoint, the VDR — reads across all of them at once, and hands back the model, memo or dataset with every number traceable to its page.

01 — WORKSPACE AGENT

Show it the work once. Ask for it on any name.

Drop in a report you already made. NOVA reads how it was built — sections, method, depth, tone — and produces the same thing for another company, off a universe it already holds.

  • 01Your own document is the brief. No prompt engineering
  • 02Every number opens the filing it came from
  • 03Out as PDF, PowerPoint, Excel or JSON
  • 04It tells you what it could not find
NOVA WORKSPACE
PDF Groww — credit rating report.pdf
Same report, same depth. Do it for Zerodha.

READING YOUR DOCUMENT

9 sections · 2 exhibits per section
Method: peer-relative, 4-year window
Rebuilding on Zerodha · 47 filings read
Ask NOVA…

THE DROP

Your best piece of work becomes the template.

Nobody wants to describe a thirty-page rating report in a text box. Give NOVA the report instead.

01 · DROP

The document you already made

Groww — credit rating report.pdf
Sector model · v6.xlsx
IC memo template.docx

02 · REVERSE-ENGINEER

Structure, method and depth, read back

9 sections
2 exhibits per section
Peer set of 11
4-year window
Tone: house, formal

03 · NAME A COMPANY

It already holds the universe

Zerodha Broking Ltd
Filings FY22–FY2547
Peers matched11
Alternative signals9

04 · SAME DEPTH BACK

In the format you need it

Zerodha — credit rating reportPDF
ExhibitsPPTX
Model and peer setXLSX
Machine copyJSON

WHAT IT DOES

Five things, and the reason behind each one.

01BUILT-IN SKILLS

The models your desk builds, already in the box.

DCF and three-statement builds, LBO and returns, credit rating reports, trading and transaction comps, IC memos, screens across the universe. Each one is a method, not a fill-in template — it argues its assumptions.

  • 01Skills adapt to your template rather than replace it
  • 02Assumptions surfaced as assumptions, always editable

SKILLS

DCF · three statement
LBO and returns
Credit rating report
Trading comps
IC memo
Screen the universe

DCF · APEX COMPONENTS

Revenue build · 5 drivers
WACC 11.4% · beta from 11 peers
Terminal g 4.0% · sanity checked
Three statements tie out
EV ₹6,340cr every cell traceable
3.0M registered in India
2.0M still active
2.8M structured by NOVA
LISTED · WHAT TERMINALS COVER ≈5,300
UNLISTED · WHERE THE WORK IS ≈2.0M
Listed + unlisted India100% of the registry
Customs, hiring, spend panelsweekly
Ratings, charges, litigationdaily
02THE UNIVERSE

You name the company. It already has the file.

Three million companies are registered in India; two million are still trading. NOVA holds the whole universe structured into queryable fields — assembled from 750 sources and a hundred alternative feeds. Terminals cover the listed 5,300. Nobody else has structured the rest.

  • 01Structured, not scraped — so one prompt can compare hundreds of names in a sector
  • 02Customs, hiring and spend panels for the private names
03YOUR SOURCES

It reads your side of the house too.

Point it at SharePoint, Drive, the VDR, your terminals. It works with the entitlements the person asking already has — deal-team walls hold, and what it skipped is stated rather than silently dropped.

  • 01Single-tenant. Your material never trains a shared model
  • 02Every read logged against the person who asked

YOUR SIDE OF THE WALL

SharePointonDriveonBloombergonCapital IQon Your VDR
09:41Read 14 documents you can already open
09:41Skipped 3 behind a deal-team wall
PDF Zerodha — credit rating report 31 pages
PPTX Exhibits, your master 18 slides
XLSX Model + peer set, native add-in writing…
JSON Every field, every source id schema v2

Same content, four containers. Nothing retyped between them.

04OUTPUT

Whatever the next person needs to open.

PDF for the committee, PowerPoint for the meeting, Excel for the analyst who will pull it apart, JSON for the system that comes after. The Excel copy lands in the native add-in, live.

  • 01Your master template, your section order, your tone
  • 02One artefact, four containers — nothing retyped
05AUDITABLE

Every number survives being questioned.

Click a figure in the finished report and you get the arithmetic, the inputs and the page each input came from. That is the difference between something you circulate and something you stand behind.

  • 01Provenance travels with the exported file
  • 02Where the record is silent, it says so
Client funds grew 31.4% year on year, against sector growth of 18.9%.
Numerator · FY25 client fundsAR FY25 · NOTE 9 · P.62
Denominator · FY24 client fundsAR FY24 · NOTE 9 · P.58
Sector growth11 PEERS · SAME BASIS
Computed, not recalled. The arithmetic is shown alongside the pages.

IN PRACTICE

How teams use the workspace

Same agent, same universe. What comes out looks like the work your desk already signs.

The pitch you built last month, on this month’s target.

Drop in the deck that won the last mandate. NOVA reads the argument you made, rebuilds it around a new name and sector map, and hands it back in your master template with every exhibit sourced.

  • 01Sector maps and target profiles off the full registry
  • 02Valuation range with the peer set justified, not assumed

OUT · PITCH DECK · SECTOR MAP · VALUATION RANGE

PPTX Pitch — Project Kiln.pptx

REBUILT FROM YOUR LAST PITCH · 18 SLIDES

Situation and mandate
2 sl
Sector map · 41 players
4 sl
Target profile · Apex
5 sl
Valuation range
4 sl
Process and timetable
3 sl
Your master template · every exhibit sourced

Read the data room the way a partner reads it.

Point it at the VDR and it works through the filings, the management pack and the registry together — reconciling what agrees and naming what does not, in the register format your firm already uses.

  • 01Findings ranked by what would change the price
  • 02IC memo drafted from the register, footnoted throughout

OUT · DILIGENCE REGISTER · IC MEMO · MODEL

PDF Apex — diligence register.pdf
Closing cash off ₹2.54cr vs its own cash flow AOC-4 · N14
₹1.14cr to a party on no register NOTE 34
Management revenue ties to filed accounts FY23–25 ✓
Two charges satisfied late MCA · CHG-4
22 MORE FINDINGS · WRITING THE IC MEMO
Your register format · 47 documents read

The Monday pack, without Sunday night.

Define the pack once — coverage, leverage, liquidity runway, charges filed, covenant tie-outs — and get it back across the book every week, with breaches carrying the clause and the arithmetic.

  • 01New charges and rating actions surfaced as they file
  • 02Same pack, 42 names, no re-keying

OUT · MONITORING PACK · BREACH LOG · COMMITTEE NOTE

XLSX Credit pack — 42 issuers.xlsxRUNNING
Kalyan Infra · net leverage 4.6x breach
Vertex Chemicals · DSCR 0.94x breach
Meridian Steel 2.9x pass
Sunrise Logistics 3.1x pass
Each breach carries the clause, the arithmetic and the filing page.
Refreshed weekly · same pack every Monday

Initiate on a name nobody on the street covers.

Unlisted comparables, customs and hiring signals, four years of filings and the last eight calls — assembled into your own note format, with each claim opening the page behind it.

  • 01Private peers most of the street cannot see
  • 02Estimate changes traced to what actually moved

OUT · INITIATION NOTE · MODEL · PEER SET

PDF Zerodha — initiation.pdf

Client funds grew 31.4%, against sector growth of 18.9%.

NUMERATORAR FY25 · NOTE 9 · P.62
DENOMINATORAR FY24 · NOTE 9 · P.58
SECTOR BASIS11 PEERS · RECONCILED
Footnoted to the page · 84 live links

Workpapers a reviewer can follow without you in the room.

Reconcile disclosures against the filed record, test cut-offs, tie comparatives, and surface what the client did not mention — every conclusion carrying the document and page that supports it.

  • 01Your firm’s methodology and file structure
  • 02Exceptions listed plainly, not buried in an appendix

OUT · WORKPAPERS · EXCEPTION LIST · REVIEW LOG

XLSX Workpapers — FY26 review.xlsx
Related-party disclosures reconciled 18/18 ✓
Two revenue cut-off exceptions Q4 · flagged
Prior-year comparatives restated note 2
Undisclosed charge on plant CHG-1 · FY24
Every tick has the document and the page behind it, in the file.
Firm methodology · review log exported

Board material that is current on the morning of the board.

Refresh the plan against actuals as they land, benchmark against listed and private peers through your own seats, and keep the pack pointing at live numbers rather than last quarter’s paste.

  • 01Variance explained line by line, with the source
  • 02Same pack quarter after quarter, in your format

OUT · BOARD PACK · VARIANCE ANALYSIS · PEER BENCHMARK

PPTX Board pack — Q2 FY26.pptx
Volume vs plan +4.0%
Realisation −2.4%
EBITDA −4.6%
Free cash flow +13.8%

Peer benchmark drawn from your own data seats.

Refreshed against actuals · exhibits still linked

THE SKILL LIBRARY

Six methods your desk already trusts.

DCF and three-statement builds

Revenue built from drivers, not a growth rate. WACC assembled from a peer beta set you can see. Terminal value sanity-checked against the implied exit multiple, and the three statements tie out before it hands anything over.

DRIVERS 5 · volume, price, mix, capex, WC
WACC 11.4% · beta from 11 peers
TERMINAL G 4.0% · implies 9.1x exit
CHECK BS balances, 24 periods

Rating reports and covenant packs

Reproduce your own rating format on any name: coverage and leverage history, peer positioning, liquidity runway, charge and default record, and the two or three things that would actually move the rating.

LEVERAGE 2.4x · 4-year trend
COVERAGE 5.1x · post-capex
CHARGES 3 filed FY25 · MCA
WATCH Two triggers within 0.3x

Sponsor cases and returns

Sources and uses, debt schedule with real covenants, base and downside cases, and an IRR bridge that tells you how much of the return is multiple and how much is operations.

ENTRY 12.6x · 4.2x leverage
BRIDGE 61% ops · 39% multiple
DOWNSIDE IRR 17.1% at −1.0x exit
COVENANT Headroom 0.4x in FY27

Trading and transaction comps

A peer set it justifies rather than assumes — same revenue basis, same period, same accounting treatment — with the outliers named and the reason they are outliers.

PEER SET 11 · basis reconciled
MEDIAN 12.6x EV/EBITDA
EXCLUDED 2 · different fiscal basis
DEALS 8 since FY23

IC memos and screening notes

Your section order, your tone, your length. It writes what the evidence supports, flags what it could not verify, and footnotes every claim to the page behind it.

SECTIONS 9 · your house order
FOOTNOTES 84 · all live links
UNVERIFIED 3 · listed, not buried
LENGTH Matched to your last memo

Screens across 2.8M companies

Ask in words — private manufacturers in Gujarat, revenue above ₹300cr, leverage under 3x, no charge filed in two years — and get the list with the arithmetic behind every filter.

UNIVERSE 2.8M · listed + unlisted
MATCHED 214 companies
FILTERS 6 · each auditable
OUT XLSX, JSON or a monitor

GOVERNANCE

The question after “can it do this” is always “can we use it”.

Six frameworks your risk, compliance and model validation teams will raise — and what the engine does about each of them. Nobody should have to take a diligence risk to try a tool.

A tool that assists a decision, with a person making it.

Regulation (EU) 2024/1689 turns on what a system is used for. NOVA produces analysis a qualified person reviews and signs; it does not score, rank or decide about anyone. The duties that reach a system like ours — oversight, traceability, record-keeping, transparency — are how the engine is built.

ART. 14 · HUMAN OVERSIGHT

A named person reviews and approves before anything leaves the workspace.

ART. 12 · RECORD-KEEPING

Every prompt, document read and output written is logged immutably.

ART. 50 · TRANSPARENCY

Generated text is marked as generated. Sourced figures are marked as sourced.

ANNEX IV · TECHNICAL DOCUMENTATION

Model, data sources, evaluation method and known limits documented for your file.

ART. 72 · POST-MARKET MONITORING

Defect, drift and incident process with a route back to you.

Govern, map, measure, manage — written down, not implied.

We run our AI risk programme on AI RMF 1.0 and its Generative AI Profile, because it is the language your risk function most likely already speaks. It is voluntary, which means the only thing that counts is the evidence underneath it.

GOVERN

Named owners for model, data and release risk; decisions minuted.

MAP

A use-case register that states intended use and out-of-scope use.

MEASURE

Evaluation on retrieval accuracy, citation fidelity and abstention behaviour.

MANAGE

Rollback, model-change and incident procedure, exercised not just written.

GENAI PROFILE

Controls for confabulation, prompt injection and data leakage across agents.

Built to the standards your risk team will name.

Our AI management system follows the ISO/IEC 42001:2023 clause structure, with per-engagement impact assessment along 42005 lines, and information security controls built to 27001:2022. We discuss certification and attestation status directly with your procurement team rather than on a web page.

ISO/IEC 42001:2023

AI management system: policy, roles, risk treatment, internal audit, review.

ISO/IEC 42005:2025

AI system impact assessment produced per engagement, shared with you.

ISO/IEC 27001:2022

Annex A control set mapped, with owners and evidence maintained.

THIRD-PARTY ASSURANCE

Attestation and audit position confirmed in writing during procurement.

CONTINUAL IMPROVEMENT

Control review on a fixed cycle, findings tracked to closure.

Personal data, insider walls and outsourcing, handled locally.

The DPDP Act 2023 governs the director, KYC and employee fields that sit inside company records. SEBI expects research and insider walls to hold. RBI expects outsourced technology to be auditable. Your deployment can be India-resident end to end.

DPDP ACT 2023 · PURPOSE AND MINIMISATION

Personal fields bound to the engagement; only what the task needs is read.

DPDP ACT 2023 · DATA PRINCIPAL RIGHTS

Per-record erasure with a receipt, across the instance.

SEBI · PIT AND RESEARCH CONDUCT

Deal-team and research walls enforced by entitlement, not by convention.

RBI · IT GOVERNANCE AND OUTSOURCING

Read-only inspector access and an auditable trail of every read.

DATA RESIDENCY

Single-tenant, India-resident deployment available.

Your validation team will ask three questions. These are the answers.

Is the arithmetic reproducible, can the output be challenged, and who owns the failure. Deterministic computation, complete provenance and a documented change process — the discipline supervisory model-risk guidance has asked of quantitative tools for over a decade.

REPRODUCIBILITY

Arithmetic and tie-outs computed deterministically, never generated.

EFFECTIVE CHALLENGE

Every figure resolves to its inputs, its arithmetic and the page behind it.

KNOWN LIMITATIONS

Where the record is silent, NOVA abstains and says so.

CHANGE CONTROL

Version pinning: the same question returns the same answer on the same version.

VALIDATION SUPPORT

Documentation pack and engineer time for your independent review.

Single-tenant, isolated, and never training on your work.

Your material stays inside your instance. Nothing you upload, and nothing NOVA reads on your behalf, trains a shared model or reaches another customer. Where a third-party model provider is involved, that is named in the contract on zero-retention terms.

TENANCY

Single-tenant deployment on an isolated network boundary.

TRAINING

Your data never trains a shared model — contractually and technically.

SUBPROCESSORS

Named in contract, on zero-retention terms, register available to you.

CRYPTOGRAPHY

Encrypted in transit and at rest, with customer-managed keys.

LOGGING

Immutable audit log, exportable to your own SIEM.

Framework mappings, our impact assessment and the subprocessor register are shared with your risk team under NDA.

Certification and attestation positions are confirmed in writing during procurement — never implied on a website.

Regulatory posture reviewed on a fixed cycle with external counsel across the EU, UK and India.

02 — EXCEL ADD-IN

The model, and the reason behind every cell.

NOVA builds and edits inside the workbook you were going to send anyway. Select a sourced figure and Trace names the filing, note and page — and the trace survives the export.

  • 01Fills your spreading template from four years of filings
  • 02Formula cells resolve back through the chain
  • 03Flags a figure two filings disagree about
  • 04Your template in, your template out
Apex_LBO_v11.xlsx
A
FY25
FY26E
12
Revenue
3,880
4,190
13
EBITDA
474
516
14
Closing cash
269
312
D14 =NOVA("APEX","closing_cash","FY26E")
NOVA
Carry closing cash into FY26E.
Wrote D14 — AOC-4 FY25, note 14.
Ask NOVA…

WHERE IT LIVES

A panel beside your sheet. Nothing new to learn.

You ask in plain words. It reads the workbook you already have, writes into it, and lists every range it touched. Your data providers switch on as connectors inside the session, so what stays behind in the sheet is your own licensed formulas.

Apex_LBO_v11.xlsx — Excel
A
B
C
D
11
FY
FY24
FY25
FY26E
12
Revenue
3,486
3,880
4,190
13
EBITDA
441
474
516
14
Closing cash
124
269
15
Net debt
1,914
1,842
C14 =NOVA("APEX","closing_cash","FY25")
NOVA Session · Apex
Spread Apex FY24–FY26E into my template and pull net debt from the filing.
Filled B12:D15 from AOC-4 FY25 and the Q3 call.
4 statements read · 2 years restated
!Closing cash differs from cash flow by ₹2.54cr
Want the same rows for the three listed peers?
on on off + connectors
Ask NOVA… Ctrl+K

01
Your workbook, your formatting. It writes into the file, never rebuilds it.

02
The panel docks beside the grid and stays with the file.

03
Connectors turn on per session, on your own licences.

04
Every range it wrote is listed, with the document behind it.

WHAT IT DOES

Five things, and the reason behind each one.

01CONNECTORS

Your terminals, switched on inside the conversation.

Turn Capital IQ on mid-question, sign in with your own seat, and from the next answer onward NOVA writes CIQ formulas into your cells — live, on your licence, still working when NOVA is closed.

  • 01Bloomberg, Capital IQ, FactSet, LSEG, CMIE, NSE — and your SharePoint and VDR
  • 02Where two sources disagree, you see both figures

CONNECTORS · THIS SESSION

Bloomberg on
S&P Capital IQ signing in
FactSet off
SharePoint on

Sign in to Capital IQ

Your seat · your entitlements

NEXT QUESTION → E12:E15
E124,190
=CIQ("APEX:IN","IQ_TOTAL_REV","FY2026")
E13516
=CIQ("APEX:IN","IQ_EBITDA","FY2026")
E151,284
=BDP("APEX IN Equity","PX_LAST")
Live formulas, not pasted values — they refresh with NOVA closed.
Add a downside case — volumes flat, realisation −4% — and keep interest on the live debt schedule.
01Copied Ops → Ops_Downside
02Assumptions written to C6:H8, kept as inputs
03Interest repointed at the live schedule
0484 dependents rechecked · none broken
Show me the diffUndo step 3Explain C6
Ask NOVA… Ctrl+K
02ASK AND EDIT

Say it the way you would say it to an associate.

It works inside the file you were going to send anyway, then tells you which ranges it touched. Nothing is a black box; every step comes back off one at a time.

  • 01Dependents rechecked after every write
  • 02Your assumptions stay marked as yours
03TRACE

Every number remembers where it came from.

Click the cell, read the page. When a figure is your judgement rather than the record, the cell says so — which is the difference between a model you can defend and one you hope nobody opens.

  • 01Provenance travels with the file you send out
  • 02Memo footnotes written from the trace
Net debt1,842
Interest168
Closing cash269
Leverage3.57x

AOC-4 FY25 · NOTE 14 · P.43

₹2.69cr closing cash, straight from note 14. The cash flow statement in the same filing says ₹2.54cr less.

Model audit 214 checks · 3.1s
Balance sheet balances · 24 periods
Cash flow ties to closing cash
Debt schedule reconciles to interest

Row 31 mixes a formula and a typed value

Ops!C31 typed 12.3% · D31:H31 computed

Stale external link

Comps_2024.xlsx · 11 months old · 6 cells

2 findings, 212 passed. Computed, not an opinion.

04MODEL AUDIT

The 11pm check, done in three seconds.

Balances, tie-outs, hardcodes sitting in formula rows, stale links, circular chains, sign and unit errors. Two hundred-odd checks that either reconcile or do not, each pointing at a cell.

  • 01Version against version, cell by cell
  • 02Findings export as a review log for the file
05RECORD AND REPLAY

Do Monday morning once. Then get it back every week.

Record yourself working across the terminal, the workbook, the deck and the email. NOVA keeps the sequence and the reasoning behind each step, then runs it on one name or the whole book.

  • 01Parameterise it — same routine, different issuer
  • 02Steps editable in plain words; every run logged
Weekly credit pack 8 steps
Open Credit_watchlist.xlsx
Spreads and CDS · 42 issuers
New charges filed at the MCA this week
Covenant tie-outs · flag breaches
Three linked exhibits refreshed
Monday note to the committee
▶ REPLAY 09:02 Mon · 42 names · 2 breaches · 6 min

FORMULA MAP

Precedents as a map, not a trail of arrows.

Walk upstream across sheets and see where a value stops being computed and starts being typed.

Exit EVOps!K40
Net debtDebt!H18
Multipletyped
IRR

HOUSE STYLE

Blue inputs, black formulas, one keystroke.

Your firm’s number formats, borders and print setup applied across the book from the keyboard.

Volume growth7.5%
Realisation / t4,910
EBITDA516
Revenue4,190

KEYBOARD

Ctrl+K from anywhere. Esc back to the cell.

Windows-first shortcuts for the dozen commands you use hourly, remappable across the desk.

Ctrl+Kcommand barCtrl+Alt+TtraceCtrl+Alt+AauditCtrl+Alt+MmapCtrl+Shift+RrecordCtrl+Alt+Preplay

IN PRACTICE

How teams use NOVA for Excel

Same panel, same workbook. A different job on every desk.

INVESTMENT BANKING

Re-spin the model and the pitch for a new target

Point NOVA at your comps template and a new name. It spreads the filings, pulls trading data through your own Capital IQ seat, and keeps the linked exhibits in the pitch in sync when a valuation assumption moves — no re-exporting tables.

  • 01Your firm’s template and formatting, not a rebuild
  • 02Trading comps refresh on your licence, in your cells

Output: filled template, comps set, refreshed exhibits

Comps_Apex.xlsx
Peer
EV
EV/EBITDA
Net debt
UltraTech
2,41,900
16.4x
18,420
Shree Cement
98,300
15.1x
2,110
Ambuja
1,12,700
13.8x
Dalmia Bharat
38,400
11.2x
6,940
Apex Cements
44,180
12.6x
1,842
Pitch_Apex.pptx · 3 linked exhibits refreshing
Wrote B4:E8 from 5 peers · exhibits relinked

PRIVATE EQUITY & VC

Build and edit an LBO without touching a formula

Ask for the case in words, then interrogate it: what is typed, what is sourced, what breaks if the exit multiple falls a turn. Diligence documents from the VDR feed the same sheet, and every figure keeps the page it came from.

  • 01Downside and management cases added as new sheets
  • 02IC footnotes generated from the trace, not retyped

Output: LBO with downside case, IC-ready footnotes

Apex_LBO_v11.xlsx
IRR · exit multiple × hold downside case added
3 yr
4 yr
5 yr
4.0x
14.2%
17.8%
21.4%
4.5x
17.1%
21.0%
24.9%
5.0x
19.8%
24.1%
28.2%
Exit multiple is typed (C6) · everything else computed. A turn off the exit takes base-case IRR to 17.1%.
Added Ops_Downside · 84 formulas rechecked, 0 broken

PRIVATE CREDIT

Covenant packs that run themselves on Monday

Record the weekly routine once — terminal pull, MCA charge check, covenant tie-outs, committee note — and replay it across the book. Breaches come back with the clause, the filing and the arithmetic behind them.

  • 01Same workflow parameterised across 42 issuers
  • 02Every run logged for the credit file

Output: watchlist refresh, breach log, draft note

Credit_watchlist.xlsx
Issuer
Net lev.
Covenant
Meridian Steel
2.9x
pass
Kalyan Infra
4.6x
breach
Sunrise Logistics
3.1x
pass
Vertex Chemicals
5.2x
breach
Orion Retail
2.4x
pass
Replaying “Weekly credit pack” · 42 issuers · step 5 of 8
2 breaches flagged · note drafted in Outlook

PUBLIC MARKETS

Auditable financials straight into your model

Pull structured financials into the model, move your assumptions, and compare the result against the street and the last four calls in a single prompt. When the numbers change, the thesis deck refreshes instead of being rebuilt.

  • 01Consensus read from the notes you are licensed for
  • 02Every estimate traceable to the transcript or filing

Output: updated model, street delta, refreshed deck

Apex_estimates.xlsx
Line
NOVA
Street
Delta
Revenue FY27E
4,510
4,428
+1.9%
EBITDA FY27E
598
571
+4.7%
EPS FY27E
38.4
36.9
+4.1%
Consensus read from 14 broker notes · every line traceable to the note it came from
Compared to consensus from 14 notes · 3 lines above street

ADVISORY & AUDIT

Review a workbook you did not build

Run 214 structural checks on a client model, read the formula map instead of chasing precedents across sheets, and hand back the findings as a review log for the file — with a cell reference against each one.

  • 01Version-to-version diff, cell by cell
  • 02Your firm’s review checklist added as house rules

Output: findings list, review log, version diff

Client_model_v4.xlsx
Reviewing Client_model_v4.xlsx 214 checks
Balance integrity · 24 periods
Depreciation ties to the fixed asset schedule
7 hardcodes in formula rows · Rev!D18:J18
Circular chain · WC!F12 → Debt!C9 → WC!F12
Tax rate differs between sheets · 25.2% / 34.9%
Review_log_Apex.docx · 3 findings written
3 findings · review log exported to Word

CORPORATE FINANCE

Board models that stay current between quarters

Refresh the plan against actuals as they land, benchmark against listed peers through your own data seats, and keep the board pack pointing at the live workbook rather than last quarter’s paste.

  • 01Variance explained line by line, with the source
  • 02Exhibits stay linked — no copy-paste cycle

Output: refreshed plan, peer benchmark, board exhibits

FY26_plan_actuals.xlsx
FY26 plan
Plan
Actual
Variance
Volume (mt)
12.4
12.9
+4.0%
Realisation
4,820
4,704
−2.4%
EBITDA
474
452
−4.6%
Free cash flow
188
214
+13.8%
Board_pack_Q2.pptx · 6 exhibits pointing at this workbook
Actuals loaded to Q2 · variance written to H4:H7

HOW IT FITS

Excel on Windows, Mac and the web.

Pushed once from your Microsoft 365 admin centre — no local installer, no desk-by-desk rollout. The shortcuts shown on this page are the Windows set.

Windows · 342 seatslive
Mac · 48 seatslive
Excel on the weblive
M365 admin pushv4.2

The file you open is the file you send.

A single-tenant instance, your template in and out, your formatting untouched. Nothing is copied into a shared workspace to be worked on somewhere else.

IN

Apex_LBO_v11.xlsx

your template

OUT

Apex_LBO_v12.xlsx

same template, filled

Nothing you already run has to come out.

Bloomberg, FactSet, Capital IQ and Macabacus keep working exactly as they are. Their formulas stay in the sheet, and the audit reads them alongside NOVA’s own.

Bloomberguntouched
FactSetuntouched
Capital IQuntouched
Macabacusuntouched

Someone will ask where the number came from.

Cell-level provenance and a run log your compliance team can read without a walkthrough: who asked, what was read, which ranges changed, which workflow ran.

09:41r.mehtaread AOC-4 FY25 · note 14
09:41novawrote D14 · sourced
09:44r.mehtaran model audit · 2 findings
09:02schedulereplayed Weekly credit pack

03 — API

One name in. Three depths out.

You keep your own tools. We supply the record from 750+ sources and the reasoning on top of it — as documents, as fields, or as a finished artefact.

  • 01Raw — the documents, as filed214 files
  • 02Structured — the record, as queryable fields38 groups
  • 03Intelligence — the artefact, with its arithmetic9 artefacts

REQUEST

Zerodha Broking
NOVA 750+ SOURCES
RAW 214 files
STRUCTURED 38 groups
INTELLIGENCE 9 artefacts

Pick your depth

Originals, with the text already pulled off the scans.

POST /v1/company/raw
{ "name": "Zerodha Broking", "since": "FY22" }
→ 214 documents · signed URLs or your bucket

AOC-4 / FY25

annual financial

118 pp · 41 tables

CHG-1 / 2026-08

charge creation

9 pp

Q3 FY26 call

transcript

· text layer

Rating rationale

ICRA · Jun 26

6 pp

+ 210 MORE · OCR TEXT LAYER · PAGE MAP

Typed, reconciled, four years deep. Join it to your own data.

POST /v1/company/structured
{ "groups": ["financials","directors","charges"] }
→ every field carries "source": { doc, page, line }

38 FIELD GROUPS INCLUDE

FinancialsDerived ratiosDirectors & KMPShareholding & pledgesCharges & securityBorrowingsContingent liabilitiesAuditor & opinionsRelated partiesLitigation & defaultCredit ratingsGST & EPFOImports & exportsGroup structureHeadcount
FINANCIALS · FY25 schema v2.4
revenue 3,880
ebitda 474
closing_cash 269 CONFLICT
net_debt 1,842

Cash flow in the same filing says 266.46. Both values come back — we don't pick one.

Our engine as a call. Arithmetic computed, never generated.

POST /v1/company/intelligence
{ "artefact": "dcf" }
→ drivers, method, checks and a replayable run id

ARTEFACTS

DCFRed flag registerPeer compsCredit assessmentCovenant testScreen the universeValuation rangeManagement qualityWorking capital
DCF run_9f2c…
volume 7.5%
realisation 1.9%
wacc 11.4%
EV ₹6,340cr statements tie ✓

Subscribe once. It reacts when the record moves.

04:12

A charge is filed

CHG-1/2026-08

04:12

NOVA re-reads the name

3 documents

04:13

Covenant test re-runs

headroom −0.18

04:13

Your system is told

webhook · run id

01Volume

One name live, ten thousand as a job.

02Freshness

Webhooks or polling, filings as published.

03Provenance

Source per field, replayable run ids.

04Schema

Versioned, additive, deprecation window.

The questions we actually get asked

Answered the way we answer them on a call, including where somebody else is the better buy.

Every sourced figure carries the document, the page and the line it was read from — AOC-4 FY25, balance sheet, Note 14. Click it and the filing opens at that page. Formula-derived cells resolve back through the chain to those sourced inputs, and the citation survives into the Excel cell, the memo footnote and the API response. Ask us to prove this live on a figure the system has just produced; it is the only demo question that matters.

It reports the break rather than reconciling it. Vouch runs a few hundred deterministic assertions across a company's filings — closing cash on the balance sheet against closing cash on the cash flow statement, open charges against secured borrowings, the eleven Schedule III ratios against the figures they are computed from. Each one passes or fails with both sides of the arithmetic shown. A language model would quietly pick the more recent figure. The contradiction was the finding.

Both, in clearly separated roles. Language models read, extract and draft. They never decide whether two figures agree — that is Vouch, which computes both sides and reports the arithmetic. This separation is what makes the output reproducible: version pinning means the same question on the same version returns the same answer, which is the first thing a model validation team will ask for.

GST turnover values — we see filing behaviour, not GSTR figures. Management accounts, unless you provide them. Unsecured and undocumented promoter funding. Unregistered charges. Anything a company simply never filed. Inside your own systems, anything the person asking cannot already see: deal-team walls and folder permissions hold, and every read is logged. A vendor who cannot name their coverage gaps has not thought about them.

Every figure is as-of dated and NOVA states which financial year it is reading. Indian private company accounts reach the Registrar six to eight months after year end, so between April and October you are largely reading the year before last — the product says that rather than letting an analyst quote FY24 figures in an FY26 committee paper. Charge filings are far fresher, because they must be registered within a statutory window of creation.

It says the filing does not exist and tells you the last year that does. Where a scan is illegible it says so rather than guessing. There is no configuration in which the system produces a confident number for a year that has not been filed.

Six frameworks are addressed directly on this page — the arithmetic is reproducible, every output is challengeable back to a primary document, version pinning fixes behaviour across a review period, and the access log shows who read what. Deployment is single-tenant inside your perimeter. We do not train on client data, and our model providers operate under zero-retention terms. Where our certification posture is incomplete we will tell you exactly where the SOC 2 audit stands and let you decide.

In your template. Hand it the workbook, credit note or IC memo format your firm already uses, plus a company name, and it returns that artefact populated with Trace on every sourced figure. This is a real capability and also the part of the product with the roughest edges — complex conditional formatting still needs review, and we would rather say so before you buy than after.

Nobody. It replaces the two days an experienced analyst spends holding six filings open at once and running arithmetic between them — work that gets skipped under a sanction deadline, which is the actual problem. What changes is portfolio-level: at two hours instead of two days you run the read on every borrower rather than the three you were already nervous about.

The licensed and internal layers are global, and EDGAR and Companies House are connected for cross-border groups. The statutory layer is deepest in India today — 2.8 million registered entities — because that is the market where the coverage layer is missing entirely. The assertions themselves are jurisdiction-independent: closing cash should reconcile under any accounting standard. What changes between markets is the connector, not the engine.

Three companies from your own book, chosen by you, including one you already know well and one you suspect is messy. Compare the register against what you know. A system that manufactures findings on a clean file is worse than one that finds nothing. A week is enough; longer and you are evaluating our patience rather than the product.

There are two of us and you get both numbers. Not a support queue — the person who can look at the specific finding and tell you whether the system is wrong or the filing is. That is a genuine advantage today and it will stop being one if we grow, so ask us again in two years.

It is more than an autocomplete inside a cell and less than a black box. The add-in works in your own workbook: you hand it your credit template or LBO shell, name a company, and it populates the input cells from filed documents — each one carrying the document, page and line it came from, visible on click and preserved when you email the file. Formulas stay yours; NOVA does not rewrite your model's logic. If what you want is a formula-writing assistant, Microsoft Copilot is the cheaper answer. If you want the inputs to be sourced and checkable, that is this.

There are three categories and they do not overlap much. General assistants (ChatGPT, Claude, Gemini) read and draft superbly but have no source of record for an unlisted Indian company. Coverage platforms (AlphaSense, Rogo, Capital IQ) are deep where a company is narrated by analysts and press — which excludes most of the 2.8 million entities on the Indian register. Document-intelligence tools extract from documents you already have, and stop when a document is missing. NOVA is built for the third case: it goes and gets the filed record, reconciles it against itself, and tells you where it disagrees. We publish the head-to-head comparisons, including where the other option is the better buy.

Tell us a bit about you.

We're rolling out NOVA carefully — quietly piloting with PE, IB and audit teams who care about getting it right. Drop your details and we'll be in touch within a day or two.

* Required

{{ dError }}

Thanks — that's with us.

We'll come back within a day or two with two slots and a short list of what to bring. Bring a company you know well and a template you actually use.