The Trust Gap · Saudi Arabia · September 2026
The evidence, definitions and calculations behind the report. Each numbered footnote returns to the report’s reference list.
Source: Riyadh_Household_Hiring_Clean_Checked.xlsx, supplied by the data owner. On 12 September 2026 the owner confirmed that the 200 household journeys are actual research records and that the file’s generation notes were incorrect. The corrected copy preserves the household, placement and recovery records. This is separate from the public-discussion evidence in notes 13 and 18.
Setting: Riyadh; household journeys active during 1 September 2025–31 August 2026, including searches begun before that window. Follow-up cutoff: 11 September 2026. There are 90 childcare-led, 80 general-housekeeping and 30 non-clinical elder-support journeys. Initial routes: 80 overseas recruitment, 80 in-country transfer and 40 company service. The collection instrument and recruitment protocol were not supplied; these findings describe the supplied sample, not every household in Riyadh or the Gulf.
Time to retention measures search start to the start date of a placement subsequently retained for 90 days; it does not add those 90 follow-up days to the search. n = 112. The 2.116 mean counts every placement through the retained person, including that final placement; it is not a return rate. Retained but unsatisfactory means 22 of those 112 households, or 19.6%. Open, abandoned and under-90-day journeys are not classified as retained failures.
Active hiring hours include review, scheduling, interviews, travel, reference checks, onboarding and supervision through the cutoff (n = 196). Extra family work covers care, cleaning, cooking and other duties within the observation window, before and after hiring (n = 188). Neither measure is monetised. P10 and P90 use inclusive linear interpolation on nonmissing household values; totals are formed per household before calculating their percentiles. Zero is a known none; missing inputs are not turned into zero.
The journey’s 112 days, 95 active hours and 413 extra family hours describe the whole household journey. They are not amounts to add again at each stage. Extra family work is unpaid and is not assigned a monetary value in the survey financial total. Figure 1 separately values the hours assumed in its illustrative case.
Figure supplied by the report owner as household_costs_revised_validated.png. The supporting numerical model is 01_Household_Cost_Support_Model.xlsx: Inputs!B7:B35, Waterfall!A6:H13 and the Checks sheet. Its seven cost steps reconcile to this image; 02_Methodology_and_Use.md documents the scenario and 04_Evidence_Checklist.csv identifies the records needed for observed-cost claims. It models one failed placement and one replacement over 12 months. Source inputs are attributed in the figure to Riyadh_Household_Hiring_Clean_Checked.xlsx, “Cost scenarios”; the time, flight, living-cost and administration inputs are additional illustrative assumptions. These scenario values are separate from the household survey results below.
Initial recruitment: SAR 14,316 invoice + SAR 2,000 separate visa fee + 20 household hours × SAR 75 = SAR 17,816.
Initial training and supervision: 60 hours × SAR 75 = SAR 4,500. No separate course fee is assumed. Continuing gaps and rework: 20 hours × SAR 75 = SAR 1,500, with no additional cash assumed.
Failed placement and return: SAR 2,000 early-return flight + SAR 500 departure administration + 8 household hours × SAR 75 = SAR 3,100.
Replacement and cover: SAR 14,316 recruitment + SAR 1,000 administration/onboarding + SAR 21,120 childcare cover + 80 household hours × SAR 75 = SAR 42,436. Cover is 3 months × 22 days × 8 hours × SAR 40. These 528 purchased-care hours are separate from the household’s 188 own-work hours. The 80 household hours comprise 20 hours searching and 60 hours training.
Food and living: SAR 500 × 9 employed months = SAR 4,500, allowing a three-month employment gap. One SAR 10,000 refund is assumed received and deducted once.
Regular and trial wages for both placements and planned home-leave travel are excluded; the early-return flight and temporary childcare invoices are included. No lost earnings are added to the valued household hours. Every cost and hour is counted once. The total includes normal overheads: it is not the net cost caused by failure or a product-savings estimate.
Net non-wage cash SAR 49,752 + household time SAR 14,100 = SAR 63,852. Over 12 months: SAR 4,146 cash + SAR 1,175 time = SAR 5,321 per month. Monthly values are annual equivalents, not recurring bills.
The survey calculation uses the same 178 households with complete financial totals for every component. Mean financial cost is SAR 32,353.230337; median SAR 31,112.50; P10 SAR 10,122.50; P90 SAR 54,522. Component means sum to the mean total; component medians or percentiles must not be added. The displayed riyal amounts are rounded. Costs cover the observation window, including ordinary employment during each household’s recorded period. The total is not an estimate of loss caused by a bad hire or of savings an UpSkillMe product would create.
Initial hiring: Households DM, comprising package, government and uncovered setup payments. Restarts: DN, comprising repeat recruitment, transfer, exit, travel, administration and onboarding charges. Extra purchased help: DO before hire and DP after hire. Employment paid: DR, comprising ordinary wages, separately recorded trial pay, living costs, company service fees and overlap costs. Amounts owed: BH. Actual earnings lost: AW. Refunds received: BI, subtracted once. Each component is averaged only across the 178 complete J totals.
Services paid for but not delivered stay in the original payment; they are not added as a second loss. Refunds are deducted only once received. Unpaid or pending refund claims and deposits held are tracked separately and are not deducted from the bill. Wages owed are included once. No extra failed-hire allowance is added over recorded restart costs, and no value is assigned to unpaid time.
Budget comparison uses each household’s original budget over the matching recorded period. Median net cash minus budget is −SAR 2,720 (n = 174). P10 is −SAR 18,000; mean −SAR 3,973; P90 SAR 8,003, rounded to whole riyals. Lower cash spending may coexist with unpaid family work. Ordinary spending avoided is already reflected in actual spending and must not be subtracted again. The report publishes aggregate results only.
On 12 September 2026 the data owner clarified the screening workload as 100–150 CVs reviewed and 10–15 interviews per hiring round. A replacement begins another round. These ranges supersede the earlier approximate whole-journey averages. They are owner-confirmed ranges, not calculated means, confidence intervals or P10–P90 values. The research covers 200 household journeys; a count of hiring rounds and corrected round-level observations were not supplied.
The owner also clarified the outcomes: 25% of hires do not last the trial period, and approximately 20% of retained hires leave households dissatisfied. These definitions supersede the earlier complete-failure and after-replacement labels. A trial duration and exact trial-failure case count were not supplied. The percentages have different denominators and must not be added together. The household rows have not been changed.
The workbook independently records 22 unsatisfactory outcomes among 112 retained placements (19.6%). Its mean of 2.116 placements through retention includes the final person who stayed; it is 2.1 total placements, not 2.1 replacements or three hires. The repeat-placement mean is an observed result for that retained subset, not a calculation from the failure and dissatisfaction percentages. No national failure count or cumulative CV/interview total is inferred.
The canonical August 2026 workbook catalogues 77 Reddit threads with 4,104 recorded comments. Twenty threads, carrying 1,857 comments, are marked FULL READ; the remaining 57 are snippet or unread coverage. It maps 59 Facebook groups, more than 200 sampled posts represented in themes, and 12 public Facebook evidence records. Group membership counts are not unique people. The workbook contains 338 coded evidence records; 187 independent Reddit/public-Facebook events are used in the issue index. A record is not a distinct person or a documented individual case.
This is exploratory desk research, not a representative population survey. The main chart now shows direct event counts for all fourteen issue categories, rather than composite scores or population percentages. Events can carry several issue codes, so the counts cannot be summed to count unique people. The workbook’s composite rankings also incorporate reading depth, engagement, breadth and severity; they differ from simple counts. Low-frequency safety issues remain important, and low posting volume does not establish low harm. Gulf accounts retain their Saudi or UAE geography.
A low count does not mean a small problem. Safety, withheld documents and fear of retaliation need attention even when they generate few public posts.
Return to the reportAmal, Noor, Rina and DEMO-E01–E08 are fictional demonstration records, not live workers or measured UpSkillMe outcomes. Coverage requires a result that supports the requested task and a method accepted by that brief. Self-declared preferences, knowledge checks, reviewed evidence and observed tasks are different claims. Missing or unsuitable evidence does not establish task coverage. Every real check must retain the person, task, method, source, date, scope and outcome. A task check is not a universal worker rating or a guarantee of overall suitability. The CV introduces the person; the Card opens relevant evidence under the worker’s sharing controls.
Return to the reportThe W1–W3 / H1–H3 graph is a separate constructed example. Graph-E records are not additional evidence about Amal, Noor or Rina. The algorithm finds the maximum compatible assignment, allowing one worker and one household per pair. Revealing W3–H3 changes two possible simultaneous assignments to three without adding a worker or a job. This demonstrates the effect of missing information in the example; it is not a forecast of market-wide hiring gains.
Return to the reportA02–A06 are worked examples with reader-selected inputs. They are not quotations, forecasts, wage promises, product prices or measured platform results. A02 uses SAR for price and loss, hours for time, SAR per hour for time value and probabilities from 0 to 1. Its output is an absolute reduction in failure probability, in percentage points. A03 uses abstract value units, a suitability prior, sensitivity and specificity; it compares the best action before a check with the expected best action after its result, then deducts the check cost. A04 uses illustrative cost units; a full reassessment replaces a refresh on use five when selected. A05 counts avoided work and improved use of ability once, then separates that created value from its distribution. A06 separates operating contribution, acquisition payback and reuse. Outputs retain their example labels; invalid inputs and unbalanced distributions remain visible.
MethodV = ΔT × v + Δp × L − P
MethodEVSI = expected best action after result − best action now − cost
MethodC₀ + (n − 1) × Cᵣ, with reassessment replacing Cᵣ
MethodAdd avoided work and better use of existing ability once. This is the pool available to distribute.
MethodAdd each recipient and cost line, then compare the total with created value. A transfer is not new value.
MethodSeparate contribution from acquisition payback and reuse leverage. The service must cover its operating costs before network effects count.
The report uses accounts recorded in the supplied canonical research workbook. Their exact section/item locators are in notes 25 and 27. Arabic versions are translations; the salary-promise excerpt was already translated from Tagalog in the workbook, with bracketed clarification. Quotes are individual accounts, not prevalence estimates. The report removed general pay-premium multipliers: the compensation tables mix observed lows, model targets and scope/arrangement differences and do not establish that an UpSkillMe check causes a pay increase.
Return to the reportThis edition does not publish measured UpSkillMe hiring, pay or retention outcomes. Future results need a stated market, period, denominator, inclusion rules and comparison baseline. Matched, started, hired and active workers must be reported separately. Below are the measurement definitions retained for future editions; they are not claims of achieved results.
The figure notes below preserve the scope, inputs and calculations behind each chart. Recruitment-fee ceilings are a component of cost, not a complete failed-hire bill. The 20% replacement-rate input is a model assumption, not an observed Saudi failure rate. Conceptual evidence/protection comparisons do not replace the applicable Musaned terms. The former fifteen-minute interview illustration was not a measured national average; two-year visa validity is not evidence of a typical hiring commitment.
15 Worked example. The names W1–W3 and H1–H3 are fictional. Opening a record does not add a person or a job.
3 12 Formula: placements × 0.20 × {14,000 or 18,000}, rounded to the nearest 100. Same engine as the agencies calculator.
9 11 14 15 Worked example. Units are not Saudi riyals. Akerlof and Pallais explain why the information can have value; Raiffa’s EVSI is the accounting identity used here.
11 15 Reuse total = 150 + (n − 1) × 20. Pallais (2014) shows that a public work record can change later hiring; this figure only prices reuse, it does not claim a Saudi wage effect.
See the household calculation and component definitions in note 29.
Source: the supplied canonical workbook, Market Evidence → Highest-Severity Primary Pain Points, item 11 (the Riyadh CV/interview account) and item 5 (trained infant care). Primary Evidence → High-Value Verbatims, item 26 (salary promise, translated from Tagalog), item 12 (raise versus replacement fees), and item 5 (Dubai worker seeking a new job). The quotation preserves the account’s “around 200” and “4–5”; it is not a hiring average or conversion benchmark.
Quotes are reproduced from the workbook’s recorded extracts, not newly authenticated interviews. Where the workbook gives a platform or thread title but no stable post link, that is the available locator. Identifying details in the worker post are omitted with an ellipsis. The explanation under each quote is the report’s interpretation; it is not additional speech by the quoted person. No quotation is evidence that a product has already reduced hiring time, fees or risk.
Return to the reportSource: Competition and Solutions → Current Product Coverage and Shared System Failures, August 2026. The table groups discovery/referrals, recruitment administration, temporary services and portable work evidence. Its final column states what the household and worker need from a hire, not an assertion that every named provider lacks the relevant feature. The research workbook includes time-sensitive provider observations; this report does not turn historical feature notes into a current exhaustive market ranking.
UpSkillMe’s evidence review identifies the task, reviewer and date. Recruitment, employment administration, refunds and dispute resolution have their own responsible providers and formal processes. A skill record does not replace a licence, visa, Musaned or Tadbeer.
Return to the reportSource: Primary Evidence → High-Value Verbatims, item 23, attributed in the workbook to a UAE household account. It reports AED 2,200 salary + AED 500 food + AED 1,500 visa/medical amortised monthly = AED 4,200. This is one household’s stated calculation, not a provider tariff, Saudi salary estimate or official permit fee. The original account does not establish whether all other costs are included.
The accompanying cost categories and role distinctions follow Compensation Strategy → All-In Employer Cost Rules and Current Discrete Salary Bands. The live compensation workbook is the operational calculator; the August master is a strategy snapshot. Cash wage, service price, government/medical costs, benefits and replacement cover are separate categories. This report does not infer agency margin from an employer invoice or present planning targets as measured wages.
Return to the reportSource titles and publication details are retained in their original language.
MHRSD’s 2025 Musaned results, published 3 May 2026, report more than 950,000 contracts and more than 1.1 million CVs. Contracts and CVs are separate reported measures: the CV figure is not a live count of currently available candidates, and contracts are not distinct workers or successful placements. Salary transfers and employment-service transfers are not added to either figure.
The additional 20% outside Musaned is the report owner’s market estimate supplied on 12 September 2026. Applied to the 950,000 headline base, it gives approximately 190,000 additional hires (950,000 × 20%). “20% above” means an addition to reported volume, not 20% of the combined market. This is not an MHRSD statistic or a separately measured national total.
The 25% trial-failure figure comes from the owner-confirmed household research described in note 30. It is not applied to national Musaned contracts to estimate a national failure count.
Source: the canonical research master, “DEEP ARCHIVE | BASE_03 Facebook Groups (59)”, group IDs 1–59. These records are dated July 2026; the archive banner was updated in August. The sum of the 59 detailed membership records is 998,788, displayed as approximately 1 million group memberships. Memberships are not deduplicated people: one person may join several groups.
Daily activity uses the tracker’s “Posts/Day (Search Card)” column. Adding the printed numbers once per group gives 1,743.3, taking 1 for the 1–2/day range and retaining the printed values of approximate and “+” indicators. The report displays 1,700+ posts per day as a rounded description of the listed activity, not a new daily census. The inputs mix search-card indicators and estimates. Cross-posts, repeat adverts and unique authors or vacancies are not deduplicated.
Country subtotals: KSA 21 groups / 196,400 memberships / 294 daily indicators; UAE 28 / 710,455 / 1,357.3; mixed-market groups 10 / 91,933 / 92. The tracker banner’s “~1.56M Members” does not reconcile to the detailed rows, so the report uses the recalculated total. The separate eight-group post-theme snapshot is not added to these 59 groups.
Facebook is evidence of search activity, not proof that a resulting hire took place outside Musaned. The same household may search socially and contract through a formal platform. These group figures therefore do not establish the 20% outside-system estimate.