Product architecture · Gulf labour markets
Not in the translation layer. In the identity primitive, the authoring model, the business model and the nature of the signal.
I pitched a version of this idea to myself in 2024, and it does not survive contact with the workforce. Each of the four breaks forces a different architecture rather than a lighter version of the same one. The reason it is worth building now is that Saudi Arabia has quietly finished the hard infrastructure: identity, contract and wage payment are all solved, and not one of those rails carries a statement of what a worker can do.
Show someone the fact that most of the world's workers hold no professional record and, within about a minute, they will propose LinkedIn for frontline workers. I have heard it from investors, from friends, and from my own notebook in 2024.
The instinct is sound. There is an obvious gap, an obvious template, and an obvious adaptation path: lighter pages, more languages, cheaper data, a simpler onboarding flow.
The adaptation path is where it fails. What follows is a diagnosis of exactly where, because the failure is not diffuse. It is four specific breaks, each in a different layer of the product, and each forcing a structural change rather than a cosmetic one. I am writing them down partly because this is the most common idea in the category and partly because I spent money learning three of the four.
The door. LinkedIn's identity primitive is an email address, and email is an artefact of professional life: issued by employers and universities, checked on a computer, tied to a way of working. For the workers we serve it is an account opened once to obtain something else and rarely accessible afterwards. The replacement is a phone number with a one-time password, delivered through the messaging application the worker already uses daily. This sounds like an implementation detail and it is not: it determines onboarding completion, account recovery, and whether a record survives a worker changing a SIM card between countries.
The form. LinkedIn asks you to describe yourself, which presumes self-description is a practised skill and that the audience rewards it. Neither holds. The replacement is to take authoring away from the worker, establishing capability through structured assessment and institutional attestation instead, with the worker confirming and controlling rather than composing.
Who pays. LinkedIn is free to the member because the member is the inventory, sold to recruiters. That is a tolerable bargain for a software engineer, who loses little by being visible to people trying to hire her. It inverts badly here, where the equivalent is charging the least powerful party in a fee-heavy market for the privilege of being hireable. The replacement is a hard constraint: the worker never pays, at any tier, ever.
The signal. A LinkedIn endorsement costs the endorser nothing, and the market prices it accordingly. Household hiring cannot run on cheap talk, because someone is deciding who enters their home and cares for their children. The replacement is evidence with provenance attached to each claim, so that a practical assessment, an institutional stamp and an untested assertion are visibly different things rather than averaged into one number.
| Layer | The primitive | Why it fails here | What replaces it |
|---|---|---|---|
| Identity | Email address | An artefact of professional life, rarely accessible, does not survive a SIM change | Phone number with one-time password, in the messaging app already in daily use |
| Content | Self-authored profile | Describing yourself has never once produced a job for this worker | Structured assessment and institutional attestation; the worker confirms rather than composes |
| Revenue | Member as inventory | Taxes the least powerful party in a market already thick with fees | Employer and partner side carries the cost; the worker never pays |
| Signal | Peer endorsement | Free to give, therefore worthless when the decision is who enters your home | Evidence with visible provenance, never collapsed into a composite score |
The intuitive explanation for why frontline workers do not complete profiles is that they cannot read the form. I held that view. The evidence made me drop it, and the correction matters because it points at a different product.
Older ILO material on Filipino domestic workers deployed overseas found education levels well above the stereotype, with a majority in some destination markets holding tertiary education.1 That data is two decades old and I would not build a strategy on it. It was enough to stop me using literacy as the explanation for everything.
The better-evidenced barrier is structural. GSMA's State of Mobile Internet Connectivity 2025 reports that 3.1 billion people, 38 per cent of the world, live in areas with mobile broadband coverage and do not use mobile internet.2 Coverage is not the constraint. The barriers GSMA identifies are affordability, digital skills and relevance, and an entry-level handset costs 16 per cent of average monthly income in low and middle income countries, rising to 48 per cent for the poorest fifth.2
GSMA's Mobile Gender Gap Report 2026 puts women's mobile internet adoption in those countries at 65 per cent, with women 12 per cent less likely than men to be online and 810 million women unconnected.3 For a product whose Saudi beachhead is household work, an overwhelmingly female occupation, that is not background colour. It is a design constraint that sits on the primary user.
Relevance is the one that decides the architecture. The problem with the form is not that the worker cannot read it. It is that nobody has ever hired her because of something she typed, so the rational response to a profile form is to close it.
This is buildable now in a way it was not five years ago, and the reason has nothing to do with the technology. The state built the infrastructure.
Musaned is the Ministry of Human Resources and Social Development's official platform for domestic worker recruitment, contracts and visas, and it runs at real volume: 412,399 contracts processed in the first half of 2024 alone.4 From 1 January 2026, salary payment for domestic workers must run through approved digital channels, a requirement phased in since July 2024.5 Saudi Arabia also introduced a skill-based classification for work permits in mid-2025, although whether that framework reaches domestic worker visas issued through Musaned rather than the standard labour system is not something I have been able to confirm.6
Identity is solved. Contract is solved. Wage payment is now solved, with a date on it.
Not one of those rails carries a statement of what the worker can actually do. A household in Riyadh can verify who she is, register what she is owed and prove that she was paid, and has no way at all to establish, before she arrives, whether she can cook, or manage a child with a medical condition, or handle a household to the standard they need.
That is the gap, and it is a specific one. It is also why this is a capability layer rather than another recruitment platform: the recruitment infrastructure exists and works.
A missing signal does not produce an absence of cost. It produces a cost that lands somewhere less visible.
The household pays it as placement failure. Recruitment through Musaned carried fees ranging from roughly 5,900 riyals for an Ethiopian worker to 14,700 for a Filipina worker as at early 2024, and those fees are sunk whether or not the match holds.7 A placement that fails in month three costs the fee twice, plus the disruption, and the household had no instrument that would have told it anything in advance.
The worker pays it as mispricing. Arriving with no record of capability, she is priced at the floor for her nationality and negotiates upward, if at all, only after months of demonstrated work inside a single household that cannot transmit the result to anyone else.
The agency pays it as churn it cannot prevent, because it is selecting on the same absent information as everyone else.
What is a better signal worth to a household, per placement?
The fee figures are real and dated; the 10-point improvement is a chosen illustration, not a measured effect, and the true failure rate in this corridor is not published. The method matters more than the number: the budget for capability evidence already exists inside the fee that is being paid for everything except capability.
If you are assessing one of these products, as an investor, an employer or a regulator, the four breaks convert directly into a diagnostic. Each question has a right answer and a tell.
| Ask | What a designed answer sounds like | The tell |
|---|---|---|
| How does a worker sign in? | Phone number and a one-time code, inside an app she already opens daily | "Email, or Google sign-in" |
| Who writes the profile? | Assessment and attestation produce it; the worker confirms and controls disclosure | "She fills in her skills" |
| What does the worker pay? | Nothing, at any tier, permanently | "There's a free tier" |
| How does an employer tell a tested claim from an asserted one? | Provenance is attached to each claim and shown separately | "We show a match score" |
The fourth question is the one that separates most sharply, and it is where I would push back hardest on my own industry. A single match score is what every buyer asks for and it is the wrong thing to give them, because collapsing evidence of different kinds into one number destroys precisely the information that makes the record trustworthy. A stamp from an accredited partner and an untested self-assertion are not two values of the same variable.
One more design position, offered because it is the sort of thing that gets quietly dropped under delivery pressure: capability assessment should never happen inside the home of the employer who is evaluating the worker. That is not an assessment, it is an unpaid trial, and the worker is not in a position to refuse it.
Building for frontline workers is not LinkedIn with lighter pages. The identity primitive changes, the authoring model inverts, the business model flips to the other side of the market, and the signal has to carry evidence rather than opinion. Get those four wrong and you will ship a product that its intended users open once, and you will read that as an adoption problem and buy more marketing.
I want to close by declining a statistic, because it illustrates the standard I would like this category to hold itself to. There is a widely repeated claim that under 1 per cent of software venture funding goes to technology serving deskless workers. It would have been the most quotable line in this paper. It traces to a venture firm's own marketing material, published by an active investor in the category, with no disclosed dataset, sample or calculation anywhere in the chain. It is a thesis statement, not a finding, and I am not going to launder it into a citation.
The same discipline applies to the rest of the numbers here. The GSMA figures are measured and dated. The Musaned contract volume is official. The recruitment fees are real and two and a half years old, which I have said rather than hidden, and the regulatory position moves fast enough that anyone acting on this paper should check it again.
The opportunity does not need inflating. Three rails are built, the fourth is empty, and 412,399 households went through the system in six months without any way of knowing what the person arriving at their door could actually do.
Disclosure
The case above stands on its own evidence; nothing in it depends on what follows. I am the founder of UpSkillMe, and the four replacements described in section two are the architecture we have built: a phone-based worker door, assessment and attestation in place of self-authoring, a permanent commitment that the worker never pays, and evidence carrying visible provenance with no composite score. Section six is therefore a diagnostic that our own product is designed to pass, and the reader should treat it as an argument for a position rather than a neutral assessment of the category. The GSMA, ILO and Saudi ministry evidence is public and stands independently of us.
This paper closes the first month of the series. Papers 5 to 8 take up assessment design, what a skills passport must contain, and the case against a single score.
Written in British English. Saudi regulatory claims are stated as at September 2026 and are the most perishable material in this series. No figure for WhatsApp penetration among migrant workers appears here because no credible one exists; the claim about messaging behaviour in section two is an operational observation from the author's own matching work, not a measured statistic. The "under 1 per cent of software venture funding" figure is discussed in section seven and deliberately not used.