Artificial Intelligence, Data & Applied Automation

AI investment in Iran: underwrite the workflow, not the model demo

An investable AI company is not a model demonstration. It is a repeatable workflow with documented data rights, a measurable customer outcome, secure deployment and a commercial path that improves as the system is used. Investors should underwrite the operating system around the model: data, evaluation, human accountability, integration, compute, contracts and cash collection.

Inside this investment brief

• An investment thesis for applied AI that starts with a customer decision, not a model name.
• Opportunity patterns where workflow access, specialised data and accountable adoption can be assessed together.
• The route from pilot to production: evaluation, secure deployment, human escalation and operating economics.
• Diligence priorities across intellectual property, data rights, model suppliers, contracts, payments and end use.

The investment case starts with a decision that matters

Where applied AI can earn customer trust

• Enterprise workflow automation: narrow, repetitive decisions in finance, procurement, service, compliance or operations where a buyer, baseline and approval path are known.
• Persian-language document intelligence: contracts, invoices, technical files or service records supported by rights-cleared corpora, traceable answers and defined reviewer control.
• Industrial inspection and maintenance decision support: visual or sensor-led systems that are tested against field conditions, false-positive and false-negative rates, escalation rules and operational liability.
• Planning and optimisation: demand, inventory, routing, scheduling, energy or supply-chain decisions measured against the existing method and the actual behaviour of operators.
• Secure on-premise or private deployment: systems designed for access control, auditability, continuity and customer-specific data boundaries where these requirements shape the product itself.
• Applied AI services that become a product: implementation expertise converted into a repeatable deployment, support model and recurring revenue rather than open-ended bespoke consulting.

From pilot to production: the deployment gates

1. Define the use case: record the in-scope and out-of-scope workflow, the current baseline, the decision owner, the human authority and the customer outcome that matters.
2. Establish the data map: identify source, provenance, controller, permitted training or inference use, quality, retention, deletion, access roles and relevant contractual or sector conditions.
3. Build the evaluation plan: compare alternatives against a representative test set, task metrics, Persian or other language conditions where relevant, known failure modes and adversarial cases.
4. Design deployment and control: document integrations, data separation, access, logs, reliability, user escalation, fallback, rollback and incident ownership before broad release.
5. Run a real commercial pilot: agree the customer scope, acceptance criteria, support responsibility, operating data and conversion point to contracted revenue.
6. Prove scalable operations: monitor drift, cost per useful outcome, capacity, release discipline, supplier continuity and retraining requirements where they arise.

A model becomes an investment-grade asset only when data rights, evaluation, human oversight and a repeatable route to revenue move together.

Contracts, economics and control

• Customer value and revenue: the economic buyer, pricing basis, acceptance criteria, renewal trigger, service level and route to collection.
• Model-output claims: what the system is expected to do, what it cannot be relied upon to do, how evaluation is evidenced and where human review remains mandatory.
• Data and confidentiality: customer permissions, use restrictions, retention, deletion, security responsibilities and the boundary between client and provider environments.
• Compute and supplier dependencies: run cost, model and API terms, hardware or cloud reliance, capacity planning and a credible continuity scenario.
• Intellectual property: ownership and licences for code, model weights, prompts, evaluation assets, datasets, open-source components and third-party tools.
• Security and accountability: access logs, incident response, liability allocation, audit rights, change approval and board-level reporting on material model risk.

Data rights, model governance and human accountability

• Maintain a data register covering origin, purpose, quality, permitted use, retention, deletion and access; never assume that access implies a right to train, fine-tune or retrieve.
• Document the system as it is operated: model version, weights or API, prompts, retrieval sources, fine-tunes, evaluation set, known limits, owners and decision authority.
• Test before and after deployment: use-case-specific evaluation, appropriate red-teaming, prompt-injection resistance, data-leakage checks and a process for unreliable output.
• Name the human role: review, override, escalation, training, record keeping and responsibility for the operational consequences of the system.
• Protect the environment: identity and access management, secrets handling, environment separation, monitoring, backups and an incident playbook.
• Control change: versioning, approval gates, rollback, vendor or model substitution, drift review, retraining and evidence for material decisions.

Structure the transaction around rights, continuity and control

• Define the investment perimeter: operating company, code repositories, model artefacts, evaluation assets, data rights, customer contracts and critical suppliers.
• Verify chain of title: employee and contractor assignments, background and foreground IP, customer data, model and provider licences, open-source obligations and termination rights.
• Reserve governance matters: sensitive uses, model changes, new data uses, compute commitments, material security incidents and independent audit access.
• Test supplier continuity: cloud, hardware, model API, MLOps, annotation and the practical plan for a model switch or service outage.
• Structure capital, payments and exit mechanics only after case-specific review of currency, banking, tax, transfer, insurance, remedies and dispute arrangements.
• Screen counterparties, beneficial owners, end use, technology, hardware, end customers and payment routes against applicable sanctions, export controls, provider terms and banking requirements.

Diligence that changes the investment decision

• The real system architecture and dependency map, not only a pitch deck or a controlled demonstration.
• Evidence of production use: customer acceptance records, service commitments, support tickets, workflow adoption and measured outcomes.
• Data provenance, quality and rights, including the gap between test data and the data available in live operations.
• Evaluation results, recognised limits, failure cases, red-team findings, monitoring and the response path when performance changes.
• Bottom-up unit economics including inference, retrieval, annotation, integration, hosting, support, working capital and implementation effort.
• Customer, model, cloud, data, IP, employment and critical supplier contracts; together with cap table, signing authority, key-person exposure and audit rights.

Governance and transferability are core workstreams

FIPPA may be assessed as a project-specific investment framework where relevant, subject to the applicable law, investment permit, approvals and specialist advice. It does not replace the contracts, third-party licences, sector permissions or data rights required to train, deploy or operate an AI system. Any capital-inflow or technology-transfer arrangement must be structured and approved case by case. Hardware, accelerated compute, cloud, model APIs, encryption, software dependencies, payment rails, end users and beneficial ownership require review under applicable sanctions, export-control, provider and banking requirements. NIST AI RMF can serve as a voluntary diligence benchmark; it is not a substitute for law or a certification promise. This brief is not investment, legal, tax, regulatory, cybersecurity or sanctions advice.

Start with a decision-ready AI file

DEAL helps investors and operators prepare the evidence for a serious AI discussion before a prototype becomes an expensive commitment: the decision being improved, the rights behind the data, the performance measure, the point of human intervention, the feasible deployment route and the party that pays. Share a high-level outline for a confidential, structured review.