Home / Platform

How Monsoon works

Monsoon connects to engagement data, grounds it in your firm's methodology and runs agreed tasks through AI agents. Professionals set the scope, review the results and sign off. The trail from source to output is kept.

  1. You set the scopeConnectBring in the data
  2. GroundLink it to your methodology
  3. You review exceptionsReasonAgents run the task
  4. You sign offDeliverOutputs prepared for review
  5. VerifyChanges can be detected

One procedure, from data to review

An illustrative example based on a type of engagement Monsoon has supported: matching commission income to provider statements across a group. confirm: example approved; use prepared data in any visuals

  1. 01

    The engagement team sets the task.

    The procedure, population, period and tolerance are defined and approved by the manager.

    Decided by: engagement manager

    Kestrel Holdings / FY2026 / C7 Commission incomeSample data

    Task set-up

    Defined by the engagement team. Agents run only what is approved here.

    Procedure
    Match commission income to provider statements
    Population
    All statement lines, 3 group entities
    Period
    1 Apr 2025 to 31 Mar 2026
    Tolerance
    R50.00 per line
    Exceptions to
    S. Pillay, senior

    Approved by N. Dlamini, engagement manager 3 Apr 2026, 08:41

    Sample data. Task set-up, approved before any agent runs.
  2. 02

    Monsoon loads the data.

    General ledgers for each entity and a year of provider commission statements, in the formats received.

    Checked by: engagement team

    Kestrel Holdings / FY2026 / C7 SourcesSample data

    Sources loaded

    6 sources, 3 formats. Anything that cannot be read is reported.

    • KBR_GL_FY26.xlsxExcel41,206 linesLoaded
    • KFA_GL_FY26.csvCSV18,940 linesLoaded
    • KLA_GL_FY26.xlsxExcel9,512 linesLoaded
    • Northgate_Life_FY26/PDF12 filesLoaded
    • Umbra_Assurance_FY26.csvCSV2,880 linesLoaded
    • Baobab_Invest_Mar26.pdfScan4 pagesPage 3 unreadable
    1. Read 3 general ledgers, 69,658 lines
    2. Extracted 6,214 statement lines from 3 providers
    3. Reported 1 page that could not be read. Sent to the team.
    Sample data. What was loaded, when, and what was incomplete.
  3. 03

    Agents perform the procedure.

    Each statement line is matched to ledger revenue; unmatched and partly matched items are flagged with the reason.

    Within the scope set in step 1

    Kestrel Holdings / FY2026 / C7 ResultsSample data

    Match results

    Every statement line in the agreed population, matched to ledger revenue.

    LineProviderStatement (R)Result
    NL-0311Northgate Life24 310.001Matched
    NL-0312Northgate Life7 905.502Matched
    UA-1187Umbra Assurance18 420.003Partly matched
    UA-1188Umbra Assurance3 116.754Matched
    BI-0045Baobab Invest11 760.005Unmatched
    6,214Lines
    6,171Matched
    31Partly
    12Unmatched
    Source for 3
    Commission statementMar 2026
    ProviderUmbra Assurance
    LineUA-1187
    Commission18 420.00

    Umbra_Assurance_FY26.csv, row 1187

    Ledger shows R16,980.00. Difference of R1,440.00 is above the R50 tolerance.

    Sample data. Each result links back to the line it came from.
  4. 04

    The team reviews exceptions.

    Reviewers open each flagged item beside its source documents, record their conclusion and escalate where needed.

    Decided by: reviewer on the engagement team

    C7 / Exceptions / UA-1187Sample data

    Exception 14 of 43

    Flagged by the agent. The reviewer decides.

    Entity
    Kestrel Brokers
    Statement
    R18,420.00
    Ledger
    R16,980.00
    Difference
    R1,440.00
    Agent's reason
    Amount differs beyond tolerance. Closest ledger entry posted 4 Apr 2026.
    Reviewer conclusion

    March commission posted in April. Timing difference; agrees to the April entry. No escalation.

    Statement, page 2
    Umbra AssuranceMar 2026
    BrokerKestrel Brokers
    PeriodMar 2026
    Total due18 420.00

    Umbra_Mar26_statement.pdf, p.2

    Sample data. The flagged item and its source, side by side.
  5. 05

    Outputs are prepared for professional review and sign-off.

    A working paper with the population, results, exceptions and reviewer notes, linked to the underlying records.

    Decided by: engagement partner

    Kestrel Holdings / FY2026 / C7 Working paperSample data

    C7 Commission income: matched to provider statements

    Prepared by: agent Reviewed by: S. Pillay Approved by: awaiting partner

    Population
    6,214 lines
    Results
    6,171 matched
    Exceptions
    43, all concluded
    Reviewer notes
    43
    Prepared by agent Senior reviewed Manager reviewed Partner sign-off

    Sources linked: 6 files, 6,214 linesIntegrity record made on sign-off

    Sample data. Monsoon prepares the working paper; it never signs off.

What happens at each stage

Five stages, each with something you can check. The technical detail sits underneath, for those who want it.

1Connect

Brings in ledgers, trial balances, spreadsheets and documents, structured or not, from the systems and files you use.

What you can check

Which sources were loaded, when, and whether any data was incomplete.

Sources · 6 loaded · 1 page unreadable

2Ground

Organises that information and links it to your firm's methodology, policies and prior knowledge, so agents work from the right context.

Technical detail: graph retrieval

Ground uses a knowledge graph combined with retrieval (graph-based retrieval-augmented generation), so answers cite the records and documents they draw on.

What you can check

The references an output relies on.

References · 3 cited

3Reason

Agents carry out the defined task, singly or in sequence, within set boundaries. Each action is logged.

Technical detail: orchestration

Reason uses orchestrated agents with defined task scopes and permissions. confirm: wording on bounds and permissions

What you can check

The log of actions taken and the evidence each step used.

Log · 14 actions

4Deliver

Produces working papers, schedules, reconciliations and draft reports in agreed formats, prepared for professional review and sign-off.

What you can check

Who prepared, reviewed and approved each output.

Prepared · Reviewed · Approved

5Verify

Records each output with an integrity check so later changes can be detected, and keeps the link back to source.

Technical detail: hashing

Verify applies cryptographic hashing to outputs and inputs. A hash shows whether something has changed; it does not on its own show that the output is correct. Correctness is established by professional review. confirm: technical description

What a hash does and does not prove

What you can check

Whether an output has changed since review.

Hash · unchanged since review

Chips show sample values. A technical brief is available on request. confirm: brief exists

Who decides what

People hold the decisions. Monsoon executes, flags and prepares.

Decision points, who holds them and what Monsoon does
Decision pointWhoWhat Monsoon does
Scope and procedureEngagement manager or partnerExecutes only what has been set up and approved
ExceptionsReviewer on the engagement teamFlags and explains; does not resolve on its own
Judgement areas and estimatesProfessional staffProvides analysis and the supporting evidence
Final outputsEngagement partnerPrepares for review and sign-off; never signs off

Users can see the actions an agent took, the decisions it recorded, the evidence it used and the explanation it gave. confirm: exactly which elements users can see We do not offer unrestricted access to the internal reasoning of the underlying AI models.

Named controls: evidence receipts on findings, verification marks on working papers, and accountability gates where a decision requires human sign-off. confirm: these feature names are current and available

Read more in the Trust centre

C7 Working paper / Sign-offSample data

Sign-off required: Partner

The workflow stops here until a partner approves or returns the working paper.

Output
C7 Commission income
Prepared
Agent, 3 Apr 2026
Reviewed
S. Pillay, N. Dlamini
Open items
None
Sample data. An accountability gate waits for a person.

Works with the data you already have

Monsoon reads from accounting systems and files. Where a source cannot be read directly, exports can be used.

Accounting systems

  • SAP
  • Sage
  • Xero
  • QuickBooks
  • NetSuite
  • Microsoft Dynamics
  • Oracle

Documented connectors. confirm: which connectors are built and supported today

Files

  • XLSXExcel
  • CSVDelimited text
  • PDFIncluding scanned documents

Data that is incomplete or cannot be parsed is reported rather than silently skipped. confirm: supported formats and size limits

Operating limits

confirm: any limits on volumes, languages, file types or systems to state publicly

Deployment

Cloud, hybrid or on-premise options. confirm: options available and any conditions Hosting and data handling are set out in the Trust centre.

Adoption questions, answered

Set-up, training and how Monsoon fits with what your firm already uses.

How long does set-up take?

It depends on the procedures, data sources and your firm's methodology. We agree a plan during scoping. confirm: typical range, if any can be stated

Does it replace our audit software?

Monsoon works alongside existing tools and working practices. confirm: integration with audit file software, e.g. CaseWare

Who needs training?

Authorised users of the core platform complete Monsoon user certification: operator training covering execution, supervision, exception handling and review. confirm: duration, cost, provider (RAiN?), validity Packaged solutions require basic user training.

User certification is separate from the ISO certifications, which apply to Monsoon's management systems. What the ISO certifications cover

Is our methodology used?

Yes. Monsoon is configured to your firm's methodology, policies and authority matrix. That configuration belongs to your firm. confirm: ownership terms

Can we try it first?

Yes, through a bounded pilot. See Work with us, or read use cases from engagements delivered with Monsoon.

See the workflow for yourself

We will walk through a procedure from data to review on prepared data.

Demonstrations use prepared data. No client files needed.