A hospital-grade website, an AI assistant, and a sovereign architecture, ready before day one.
A UAE hospital needed a modern bilingual website and an AI assistant its patients could trust. A fractional expert team, about one full-time person in combined effort, directed an AI delivery team to design, build, harden, and rehearse the whole estate before day one of deployment, at a fixed price less than a tenth of the modelled cost of a conventional build.
<10%of the modelled cost of a conventional team delivering the same scope, at a fixed price
Client
A healthcare customer in the UAE
Led by
Blackstone&
Engagement
Delivery, Fixed price
Sector
Healthcare
01 / Context
A hospital whose patients now start online
An established healthcare provider in the UAE had outgrown its web presence. Patients increasingly start with a search and a question, and the hospital needed a modern bilingual site and an AI front door that could answer safely, in English and Arabic, without putting trust at risk.
Healthcare raises the bar on every axis: regulated AI guidance, data protection law, information-security standards, and an audience for whom a wrong answer is never acceptable.
02 / The Challenge
Enterprise scope, fixed-price economics
The scope was the kind that conventionally takes a multi-role team several months: a full bilingual website, a patient-facing AI assistant with hospital-grade guardrails, hardened in-region cloud infrastructure, and the compliance evidence to stand behind all of it.
It had to be delivered at a fixed price the hospital could justify, and it had to leave the hospital able to run its own site, not dependent on an agency for every change.
03 / The Approach
A fractional expert core, an AI delivery team, and everything built ahead
We fielded a fractional team of four specialists: a senior architect, an AI product owner, a platform architect, and an AI data engineer, whose combined effort totals roughly one full-time person. Around them, an AI delivery team of developers, designers, testers, DevOps engineers, solution architects, and compliance and security specialists did the volume work inside a governed agentic loop that helped draw the delivery roadmap, write the software and the infrastructure code, and check it at every stage, with an independent AI reviewer gating each step before it counted as done.
We built ahead of the contract clock, at our own risk. The full site was designed, built, and put in front of the hospital for review first. The production estate followed as infrastructure-as-code on in-region cloud, deployed for live evidence windows, torn back down to a verified zero standing cost, and taken through a full dress rehearsal of go-live.
The discipline is what makes the model safe, not just cheaper. Each AI role runs in a harness we built internally: scoped to its job, fed structured plans rather than open-ended prompts, and required to produce evidence alongside its output. The agent that verifies a step is never the one that built it, deploy rights are withheld from the loop entirely, and decisions, spend, and anything public stop at a human. That tooling is what turns AI volume into delivery a hospital can accept.
The team, as fielded
04 / Outcomes & Impact
Ready to deploy, with the evidence to prove it
331 bilingual pages, 117 doctor profiles, and 26 department pages are built and client-reviewed, scoring 95 or above for performance and accessibility and 100 for best practices and SEO, with every critical accessibility issue remediated. The AI assistant answers in English and Arabic and fails closed: where it cannot ground an answer in approved content, it hands off to a person rather than guessing.
Beneath the site sits real infrastructure engineering, not just hosting. The production estate is defined entirely as code: a web application firewall, private in-region networking, workload protection, observability and alerting, autoscaling, and a rehearsed disaster-recovery position with defined recovery objectives. It is also a sovereign architecture: every piece of personal data is processed inside the UAE, on cloud infrastructure recognised under the national sovereign-cloud framework, and because the estate is code it is portable by construction. The hospital is never locked in.
The compliance work is done as architecture, not paperwork: regulated-AI, data-protection, and information-security obligations are mapped to the design in a traceability matrix, with hardening, disaster-recovery, and operational runbooks written. Everything is built and approved by the hospital, and the engagement now moves into deployment, at a fixed price less than a tenth of the modelled cost of a conventional build.
05 / Capability Transfer
Built from day one to be handed over
Self-operability is contracted, not aspirational. Six self-service content skills, a local authoring studio, and operator runbooks mean the hospital's own team will run the site after handover, with the training programme delivered as part of deployment. And because the infrastructure is code, the whole estate redeploys into the hospital's own cloud tenant rather than staying on ours.
By the numbers
0
Bilingual pages built and client-reviewed: 117 doctor profiles, 26 departments, full Arabic right-to-left support
≈1
Full-time person of combined fractional expert effort, directing an AI delivery team spanning development, design, test, DevOps, architecture, compliance and security
$0
Standing cloud cost while the production estate was built, validated live, and rehearsed, deployed only for evidence windows
How we made the call
The decisions behind the result
Anyone can list what was built. What matters to a buyer is the judgement: what we chose, what we rejected, and why.
Build the whole thing before day one, not a proposal about it.
We designed, built, and rehearsed the site, the AI assistant, and the production estate ahead of the engagement clock, at our own risk. The hospital approved working software instead of promises, and the engagement itself de-risks to deployment.
A fractional expert core directing an AI delivery team, not a bench of generalists.
The client pays for judgement: four senior specialists at a combined effort of about one person. The governed AI loop does the volume, and an independent AI reviewer gates every step, which repeatedly caught defects before they could ship.
Treat compliance as architecture, not paperwork.
Regulated-AI, data-protection, and security obligations were inputs to the design: in-region processing, private networking, an assistant that fails closed. The evidence trail builds itself as the estate is built, instead of being reconstructed at the end.
Keep the estate at zero cost until it earns its keep.
Infrastructure-as-code means deploy, prove, tear down, redeploy. The hospital pays standing cloud costs only from go-live, and the rehearsed estate carries no surprises into it.
Coming soon
Part two: deployment, go-live, and handover
This study is published mid-engagement. Part two covers deployment to production, go-live on the hospital's own domain, the training programme, and handover into the hospital's own cloud tenant.
* Published as a live, in-flight case study: everything described is built and client-approved at the time of writing; deployment and go-live follow in part two. Client identity withheld; details generalised to protect confidentiality.
See yourself in this story?
Technology is the science. Partnership is the art. Let’s write the next one together.