Flagship · FlowCount WHT · Pilot available

The WHT Health Check

Three to five years of payments to non-residents, tied to your ledger, classified by substance, recomputed with effective-dated rules, and signed by a licensed professional. Ten steps, eight gates, and a sample input and expected output at every step.

Your ERP codes WHT by GL account. ZATCA assesses it by substance.

فحص ضريبة الاستقطاع: اكتشف المخاطر قبل الهيئة

Book a 4-week review Walk through the 10 steps

The sample at a glance

37 payments. SAR 28.14M. 19 non-resident vendors. Jul 2023 to Jun 2026.

Every figure below comes from an actual engine run on a fictional client population built to exercise every rule. Every number reconciles, and the build refuses to produce a report that does not.

1,501,895.84

total exposure: tax + fines to 30-Sep-2026

Basis: illustrative sample engagement (fictional data)

104,068.99

over-withheld: refund candidates, never netted

Basis: illustrative sample engagement (fictional data)

337,500.00

duplicate payment found along the way

Basis: illustrative sample engagement (fictional data)

87%

of exposure from one root cause: management fees to the parent with nil WHT

Basis: illustrative sample engagement (fictional data)

Findings

Eight findings, priced to the as-of date and ranked.

Each finding is written in 5C form: condition, criteria, cause, effect and recommendation. The findings add back to the lead schedule to the cent.

F1Nil WHT on group management fees, including one quarter paid outside AP4 lines · High

1,308,000.00

F3Software and trademark rights treated as services; one invoice paid twice3 lines · High

87,637.50

F4Permanent-establishment risk: 210 days on site1 lines · High

36,787.50

F5Wrong domestic rates (related-party recharge; no US treaty)2 lines · High

31,586.25

F2Treaty rate on an expired certificate1 lines · High

26,910.00

F6Mixed supply not split; gross-up not applied2 lines · Medium

6,748.34

F7Late remittance, wrong period, no remittance evidence4 lines · Medium

4,226.25

F8Evidence and data gaps4 lines · Medium

0.00

Total exposure, tax + fines to 30-Sep-20261,501,895.84

Basis: illustrative sample engagement (fictional data). Findings F2 to F8 sorted by exposure; F8 records evidence gaps that carry no figure until the documents arrive.

The flow

Ten steps, at one glance

Step	Name	AI?	Primary output	Gate after
S1	Intake & population build	T1 (optional)	As-received extract + hash log	G1
S2	Completeness & data quality		TB tie-out, DQ log	G2
S3	Scope & residency screening		In/out of scope per line
S4	Nature-of-payment classification	T2	Proposal → gate → reviewer disposition	G3 already passed
S5	Treaty, TRC & PE screening		Treaty eligibility, PE flag
S6	Rate determination & recalculation		Expected WHT, variance
S7	Filing & payment timeliness		Due dates, delay fines
S8	Return reconciliation		ERP ↔ Form Q ↔ engine by month
S9	Exceptions, reason codes, risk, evidence		Ranked register, document gaps	G4
S10	Review, sign-off, reporting	T3 (optional)	Findings report	G5 → G6 → G7

S1T1 Document AI on scanned contracts and TRCs (optional)

Intake & population build

Get the complete population of payments and credits to non-residents, in the 30-field template, with enough control information to prove later that nothing was lost.

Record a file hash and row count for each item received, and agree the row count and total with you in writing.
Store the as-received file untouched; corrections happen downstream, with a reason.
Where contracts arrive as scans, FlowCount Document AI populates contract description, service location and tax-borne-by-payer, and flags those fields as AI-extracted.
Eight recommended fields are added to the template (payment ref, days on site, tax borne by payer, DTA application ref, SAMA FX rate, beneficial owner, certificate date, RCM booked). Without the first four, PE, gross-up and treaty-procedure checks can only return INDETERMINATE.

The one data request (issued at G0)

PBC	Item	Format
PBC-01	AP payments and credits to non-resident vendors, all company codes, the review period, in the 30-field template	CSV / XLSX from ERP
PBC-02	Trial balance for each year in scope, at GL-account level	XLSX
PBC-03	Form Q returns as filed, with portal acknowledgements and SADAD receipts	PDF / portal export
PBC-04	Contracts / SoWs for the top 80% of spend by vendor, plus every related-party agreement	PDF
PBC-05	TRCs held, by vendor and year; DTA application references	PDF
PBC-06	Related-party register / group structure chart	XLSX / PDF
PBC-07	Intercompany netting statements and manual journals to non-resident vendors	XLSX

Gate · G1: every item received or waived in writing; row counts and control totals agreed; hashes logged.

S2Rules only

Completeness & data quality

Prove the population is whole before testing whether it is right.

Sum the population by GL account for each year and compare it to the trial balance postings to non-resident vendors. Every difference must be explained; an unexplained difference becomes a finding.
Run mandatory-field tests and the duplicate test. Both occurrences of a duplicate are flagged, so the result does not depend on row order.

Expected output: completeness

GL	Category	TB FY2025	AP population	Difference	Potential WHT	Status	Explanation
610100	MGMT	4,500,000.00	3,000,000.00	1,500,000.00	300,000.00	FAIL	JV-2025-0931 intercompany netting: Q3 2025 management fee, not paid through AP
610200	TECH	2,118,200.00	2,118,200.00	0.00	0.00	PASS
… 9 other GLs				0.00		PASS

Expected output: data-quality log

Payment ref	Test	Finding	Downstream effect
PY-25-00688 / PY-25-00702	Duplicate	Invoice INV-88213 paid twice, 337,500.00 each	WHT still tested (a payment occurred); overpayment recovery raised to AP
PY-26-00391	Mandatory field	Payment date missing	Rate tested; period and timeliness INDETERMINATE

Why this step earns its fee. The single largest finding in the sample, 300,000 of potential WHT, is not in the data the client handed over. It only surfaces because the population was tied back to the ledger.

Gate · G2: the population ties to the TB, or every difference is a logged finding.

S3Rules only

Scope & residency screening

Remove what WHT does not apply to, and record why, so nothing is silently excluded.

Out of scope	Code	Sample line
Resident vendor in the non-resident population	SCP-010 (exception: vendor master fix)	Riyadh Office Supplies
Supply of goods	SCP-030	Goods element of a mixed supply
Inbound freight into KSA, on or after 15 Sep 2023	SCP-020	One line clean (nothing withheld); one line 5% withheld: refund
Air ticket for a flight departing outside KSA, on or after 15 Sep 2023	SCP-021	5% withheld: refund

An out-of-scope line only becomes an exception if tax was withheld on it (a refund) or the vendor master is wrong.

S4T2 Propose

Nature-of-payment classification: the AI step

Decide the WHT category by substance. This is where most real exposure hides, because ERP tax codes follow the GL account and the GL account follows whoever raised the PO.

The AI proposes. It reads contract (weight 3) > invoice (2) > PO (1.5) > GL prior (1) and returns a category, a confidence, and the quoted evidence it relied on, restricted to the rulebook categories plus MIXED and UNCLASSIFIED.
The gate decides. A proposal is auto-accepted only if confidence ≥ 0.85, it agrees with the GL mapping, and it is not MIXED or UNCLASSIFIED.
A reviewer decides the rest, with a disposition, a name and a rationale. With no decision the line stays UNRESOLVED: no WHT figure is computed for it, and it is never guessed.

Expected output: one row of each type

Payment ref	AI proposed	Conf.	AI evidence (abridged)	Gate result	Reviewer disposition	Final
PY-25-00301	MGMT	1.00	contract "Management services"; invoice "management fee"	Auto-accepted		MGMT
PY-25-00688	ROYALTY	0.83	contract "right to access and use"; invoice "SaaS"; GL → TECH	Conflict with GL → review	Confirmed AI: right to use software is a royalty (firm position). Contested; conservative view taken and the alternative quantified.	ROYALTY
PY-26-00170	TECH	0.57	"intellectual property" → ROYALTY; "design services" → TECH	Low confidence → review	Overrode AI: outright assignment of IP is not a licence to use; consideration is for design services.	TECH
PY-26-00318	UNCLASSIFIED	0.00	no substantive text (contract blank)	AI abstained → review	Decided: schedule obtained, IT systems administration, a technical service.	TECH
PY-26-00144	MIXED	0.60	"spare parts" → GOODS; "installation supervision" → TECH	Mixed → review	Decided: goods 85% / services 15% per price schedule.	TECH (15% of base)
PY-26-00330	OTHER	0.83	"training" → OTHER; GL → TECH	Conflict → review	None yet: course outline requested	UNRESOLVED

Sample totals: 30 auto-accepted · 6 reviewer-decided (3 confirmed, 1 overridden, 2 decided where the AI abstained or flagged mixed) · 1 awaiting a document.

Gate · G3 must already be passed: golden set at target, thresholds set. At G5, a 10% random sample of auto-accepted lines is re-performed, and every evidence quote is verified to exist verbatim in the source.

S5Rules only

Treaty, TRC & PE screening

Apply a reduced rate only where it is supportable, and catch treaty claims that are not.

Treaty rate = the lower of the treaty article rate and the domestic rate (Art. 7, 8, 10, 11 or 12).
The treaty rate is used only if the TRC is valid for the payment date. Expired, missing or pending means the domestic rate.
No DTA, no relief (for example the United States).
Art. 7 relief is blocked where on-site days reach the PE threshold (default 183); the line is rated High and escalated to the Partner.

Expected output

Payment ref	Country	Category	Art.	TRC	Days	Domestic	Treaty	Applied	Result	Note
PY-25-00131	IN	ROYALTY	12	Valid		15%	10%	10.0%	PASS	Valid TRC, DTA application on file
PY-25-00615	CN	ROYALTY	12	Expired		15%	10%	10.0%	FAIL	TRC expired → domestic applies
PY-25-00520	US	TELECOM	7	Missing		5%		0.0%	FAIL	No comprehensive DTA with KSA
PY-25-00977	PK	TECH	7	Valid	210	5%	0%	0.0%	FAIL	Art. 7 relief unsafe: on-site presence ≥ PE threshold
PY-25-00120	GB	INTEREST	11	Valid		5%	0%	5.0%		Benefit available, not used → refund

S6Rules only

Rate determination & recalculation

Compute what should have been withheld, and compare it to what was withheld. Re-performable as live spreadsheet formulas.

Effective dating matters. The same related-party vendor providing technical services is correct at 15% in Aug 2023 (before the amendment) and over-withheld at 15% in Oct 2025, where the correct rate is 5%: a 34,400 refund.

The arithmetic

Amount SAR     = ROUND(Payment Amount × FX, 2)        FX = SAMA rate on payment date
WHT Base       = Amount SAR × Base Portion            portion < 1 only for mixed supplies
               ÷ (1 − rate)  if tax borne by payer    gross-up
Expected WHT   = ROUND(WHT Base × Applicable Rate, 2)
Variance       = Expected − Actual                    |variance| ≤ 1.00 → PASS

Expected output: selected lines

Payment ref	Category	Amount	Portion	Gross-up	Rate	WHT base	Expected	Actual	Variance	Code
PY-25-00301	MGMT	1,500,000.00	100%	N	20%	1,500,000.00	300,000.00	0.00	300,000.00	TRT-300
PY-25-00688	ROYALTY	337,500.00	100%	N	15%	337,500.00	50,625.00	16,875.00	33,750.00	DQ-800
PY-26-00090	RENT	172,000.00	100%	Y	5%	181,052.63	9,052.63	8,600.00	452.63	GRS-250
PY-26-00144	TECH	780,000.00	15%	N	5%	117,000.00	5,850.00	0.00	5,850.00	CLS-120
PY-25-00731	TECH	344,000.00	100%	N	5%	344,000.00	17,200.00	51,600.00	(34,400.00)	RTE-210
PY-23-00840	TECH	430,000.00	100%	N	15%	430,000.00	64,500.00	64,500.00	0.00	OK-000

Tie-out: the workpaper formula total for expected WHT, less the engine total, equals 0.00.

S7Rules only

Filing & payment timeliness; delay fines

The due date is the 10th of the month after payment, for both the Form Q return and the remittance. The delay fine is 1% of unpaid tax per 30 days or part, so one day late is a full period. Fines on under-withheld tax accrue to the as-of date fixed at G0.

The fine arithmetic

Due date               = DATE(YEAR(pay), MONTH(pay)+1, 10)
Late periods (paid)    = ROUNDUP(MAX(0, paid − due) / 30, 0)
Fine, late payment     = ROUND(Actual WHT × 1% × late periods, 2)
Periods to as-of       = ROUNDUP(MAX(0, as-of − due) / 30, 0)    only where under-withheld
Fine, unpaid variance  = ROUND(Under-withheld × 1% × periods, 2)
Exposure               = Under-withheld + both fines

Expected output

Payment ref	Paid	Due	WHT paid	Days late	Periods	Fine (late)	Under-withheld	Periods to 30-Sep-26	Fine (variance)	Exposure
PY-25-00350	15-May-25	10-Jun-25	02-Jul-25	22	1	4,000.00				4,000.00
PY-26-00222	08-Apr-26	10-May-26	11-May-26	1	1	75.75				75.75
PY-25-00301	30-Apr-25	10-May-25					300,000.00	17	51,000.00	351,000.00
PY-25-00615	15-Jun-25	10-Jul-25	09-Jul-25	0	0		23,400.00	15	3,510.00	26,910.00
PY-26-00391	(missing)		09-Jul-26			INDETERMINATE				0.00

Sample totals: fines for late remittance 4,226.25 · fines on unpaid variance 126,854.46.

S8Rules only

Return reconciliation: ERP ↔ Form Q ↔ engine

Three independent views of each month should agree. Where they do not, the difference is a finding, not an adjustment.

Expected output: months with differences

Period	ERP booked	Form Q filed	ERP − Form Q	Engine expected	Engine − Form Q	Reading
2025-04	0.00	0.00	0.00	315,050.00	315,050.00	Management fee not withheld; crane hire filed in the wrong month
2025-09	33,750.00	16,875.00	16,875.00	101,250.00	84,375.00	Form Q omits WHT booked on the duplicate payment; both SaaS lines under-withheld
2025-10	51,600.00	51,600.00	0.00	17,200.00	(34,400.00)	Pre-amendment related-party rate still in the ERP
2026-04	14,025.00	14,025.00	0.00	314,025.00	300,000.00	Q1 2026 management fee not withheld

Reconciling item: 3,030.00 of engine-expected WHT cannot be assigned to a month (no payment date). It is shown on the sheet, not hidden.

S9Rules only

Exceptions, reason codes, risk & evidence

Every exception gets a reason code (24 codes in 12 families), a risk rating from a documented model, an owner and a recommended action. Each evidence gap becomes a document request.

The risk model

Impact points from exposure:  ≥100k → 5 · ≥25k → 4 · ≥5k → 3 · ≥1k → 2 · >0 → 1
+1 related party with exposure   +1 treaty misapplied   +2 classification unresolved
+1 evidence missing (+2 if line ≥ 500k)                 +1 late payment
High ≥ 5 · Medium ≥ 2 · else Low            PE risk and duplicate payments force High

Expected output: exception register, top of the ranking

ID	Payment ref	Vendor	Code	Risk	Exposure incl. fines	Recommended action
WHT-C01	JV-2025-0931	(not in AP extract)	CMP-900	High	300,000.00	Obtain JV support; assess WHT on the netted amount; extend the population to every settlement channel
WHT-001	PY-25-00301	Parent Holdco Ltd	TRT-300	High	351,000.00	Pay with fines, or obtain TRC and file a DTA application within 5 years
WHT-002	PY-25-00544	Parent Holdco Ltd	TRT-300	High	342,000.00	As above
WHT-003	PY-26-00301	Parent Holdco Ltd	TRT-300	High	315,000.00	As above
WHT-004	PY-25-00688	SaaS Vendor Inc	DQ-800	High	37,800.00	Recover overpayment; reverse WHT on credit; fix AP duplicate control

25 rows in total.

Check matrix across the population

Check	PASS	FAIL	INDET.	N/A
C1 Scope	37	0	0	0
C2 Classification	36	0	1	0
C3 Treaty eligibility	5	6	0	26
C4 WHT amount	20	16	1	0
C5 Return period	25	7	1	4
C6 Filing timeliness	25	7	1	4
C7 Payment timeliness	22	9	2	4
C8 Evidence	36	0	1	0
C9 Data quality	34	3	0	0

Gate · G4: every AI proposal dispositioned or carrying an open request; tie-outs = 0.00; register complete with owners.

S10T3 Draft (optional narrative)

Review, sign-off & reporting

Manager review re-performs every High exception from source documents and a random 10% of auto-accepted lines. The Partner approves ratings and wording; an engagement quality review runs where triggered.

The sample meets four EQR triggers: exposure above 1M, a PE finding, a contested SaaS position, and the first use of the classifier.
The report shows the contested alternative: treating SaaS at 5% instead of 15% would mean 67,500 less tax.
Integrity check built into the report generator: findings must add back to the lead schedule to the cent (1,501,895.84) and opportunities to 104,068.99. If they do not, the report is not produced.
G7: issue, lock, archive within 60 days, retain 10 years. Rulebook lessons go to Tax Technical.

Gate · G5 → G6 → G7

What you receive

A decision-ready report. The audit-grade file behind it stays on record.

The full workpaper file (16 tabs, 1,724 live formulas in the sample, 0 errors) is what an EQR partner, a regulator or a ZATCA audit would inspect.

Deliverable	What it contains	Audience
Findings report (PDF / DOCX)	Executive summary, 8 findings ranked, a 30-day action list, scope and an explicit "How AI was used" statement, 5C findings, the refund opportunity, appendices	CFO, Head of Tax
Exception register with owners and status	Every exception, coded, rated and priced	Tax, AP, Treasury
Return correction schedule, by month	ERP ↔ Form Q ↔ engine, ready for amended returns	Tax
Refund claim schedule	Over-withholding by vendor and period, with claim deadlines	Tax
Information request (outstanding PBC)	What is still needed to conclude INDETERMINATE lines	Client finance
Control recommendations	Preventive and detective controls, owner and frequency	Financial controller

Built to catch

Ten pitfalls this method is designed to catch

Pitfall

Testing only the population the client gave you

Caught by: S2 trial-balance tie-out (CMP-900)

Pitfall

Trusting the GL / ERP tax code

Caught by: S4 substance-based classification; GL as a weak prior

Pitfall

A flat rate table across years

Caught by: Effective-dated rulebook (the 2023 amendment)

Pitfall

Treaty rate "because they're in a treaty country"

Caught by: S5: valid TRC for the payment date, no-DTA countries, PE block

Pitfall

Computing WHT on the net when the payer bears the tax

Caught by: Gross-up (GRS-250)

Pitfall

Mixed goods-and-services contracts

Caught by: MIXED abstention → apportionment (CLS-120)

Pitfall

"Paid on the 11th, near enough"

Caught by: 1% per 30 days or part: one day late costs a full period

Pitfall

Netting refunds against exposure

Caught by: Reported separately (principle P6)

Pitfall

A guess dressed up as a number

Caught by: UNRESOLVED lines carry no expected WHT

Pitfall

Document order changing the answer

Caught by: Shuffle test in the regression suite

Honest limits. Sample data is fictional. Domestic WHT rates and the 15-Sep-2023 amendment were checked against public sources on 29-Sep-2026; treaty rates and FX in the sample are illustrative and are verified against primary text by Tax Technical before any client use (gate G3). Contested positions are flagged, not hidden. This page describes a method; it is not tax advice. A real engagement is signed by a licensed professional at G6.

Four weeks. One data request. A priced answer.

Week 1 acceptance and data, week 2 complete and classify, week 3 quantify and review, week 4 report and decide.

Book a 4-week review Add refund recovery