· Sam Fielding · OpenAI · AI implementation · workflow automation · business AI

Companies Using OpenAI: 5 Verified Business Examples

Companies using OpenAI are applying it to internal knowledge, customer service, staff productivity and customer-facing products. Echelon AI Solutions reviewed five public examples from the companies' own reports and announcements. Their practical value comes from how each company narrowed the job, connected trusted information and kept people responsible for the final decision.

The Bottom Line

  • Morgan Stanley uses OpenAI to help financial advisers retrieve approved internal knowledge.
  • Klarna uses an OpenAI-powered assistant for a large share of customer service chats.
  • Moderna and BBVA gave staff controlled AI assistants for work across several functions.
  • Duolingo put GPT-4 into paid learning features while people still write and review the learning content.

Which Companies Are Using OpenAI?

The five examples below are based on company announcements, reports or filings. They show four distinct patterns: internal knowledge retrieval, customer service, employee assistance and AI inside a customer product.

CompanyOpenAI UseWhat The Workflow DoesWhat Stays Human
Morgan StanleyInternal adviser assistantSearches approved firm knowledge and returns an answer with source linksAdvisers review the information and advise the client
KlarnaCustomer service assistantHandles common service conversations and routes work through support operationsPeople handle difficult, sensitive and disputed cases
ModernaCompany-wide assistantsHelps staff retrieve, draft and analyse work across business functionsStaff remain accountable for scientific, legal and business decisions
BBVAEmployee productivity toolsSupports tasks such as summaries, translation, reports and codingEmployees review outputs and control the work
DuolingoLearning product featuresRuns conversational practice and explains answers inside Duolingo MaxLearning experts write scenarios, review outputs and respond to errors

These are examples, not a promise that the same workflow will produce the same result elsewhere. Each company had different data, controls, staff and operating conditions.

Morgan Stanley: Trusted Knowledge For Financial Advisers

Morgan Stanley announced in March 2023 that it was building an internal service with OpenAI’s GPT-4. The tool was designed to help financial advisers search the firm’s own research and intellectual capital. Its answers link back to the source documents so advisers can inspect the material before using it.

The important design choice is the boundary. The system searches approved internal content. It does not replace the adviser, decide what is suitable for a client or send advice on its own. Morgan Stanley described the adviser as central to the client relationship.

Source: Morgan Stanley, “Morgan Stanley Wealth Management Announces Key Milestone in Innovation Journey with OpenAI”.

Klarna: High-Volume Customer Service

Klarna launched its OpenAI-powered customer service assistant in February 2024. In its annual report for the year ended 31 December 2025, Klarna said the assistant handled 80% of customer service chats. The company also said its internal surveys showed no reduction in customer satisfaction.

That is a company-reported result, not an independent benchmark. It still shows the right shape for an automation candidate: a large queue of repeated conversations, clear account data and a defined path to customer support.

The assistant should not be read as proof that every customer issue belongs with software. Identity theft, disputes, hardship and unusual account problems carry more context and risk. Those cases need a person with the authority to investigate and act.

Source: Klarna Group plc, 2025 Annual Report filed with the US Securities and Exchange Commission.

Moderna: Assistants Across Business Functions

Moderna said it launched an internal assistant called mChat in 2023 using OpenAI’s API, then rolled out ChatGPT Enterprise across the company. Its public examples cover clinical development, legal work and corporate brand tasks.

This is an operating model, not one chatbot. Staff can build purpose-specific assistants around a defined job. The company also invested in training and internal adoption. That matters because access to a model does not tell an employee which information is approved, when to check a source or who owns the decision.

The repeatable lesson is to pair access with use-case rules. A legal research assistant, a clinical analysis assistant and a brand assistant need different source material, review steps and permissions.

Source: Moderna, “Moderna’s Collaboration with OpenAI: Transforming the Business of Medicine”.

BBVA: Controlled Employee Productivity

BBVA expanded its ChatGPT Enterprise agreement to 11,000 licences in May 2025 after an initial rollout of 3,300. The bank said licensed employees used the tool for translation, summaries, reports, coding, legal information and marketing work. It also reported more than 3,000 custom assistants created by staff.

Those figures come from BBVA’s own announcement. The useful part for another business is the rollout method. The bank combined licences with training, a user community and role-specific assistants. It did not treat general access as the full implementation.

This pattern suits work where a person already owns the output. The assistant can prepare a summary or first draft. The employee checks the source, applies judgement and remains responsible for what leaves the business.

Source: BBVA, “BBVA Expands Its Agreement with OpenAI to 11,000 ChatGPT Licences for the Bank’s Employees”.

Duolingo: AI Inside A Customer Product

Duolingo uses OpenAI’s GPT-4 in Duolingo Max. Its public product page describes conversational practice through Video Call and Roleplay. This is different from an internal assistant because the model output appears inside the customer experience.

Duolingo also describes the work around the model. Its team writes scenarios and initial messages, reviews AI output for accuracy and tone, and gives learners a way to report mistakes. The model runs the conversation, while learning specialists define and monitor the experience.

That human layer is important for any customer-facing AI. The business needs approved content, testing, feedback capture and a clear owner for corrections. A model connection alone does not provide those controls.

Source: Duolingo, “Introducing Duolingo Max, a Learning Experience Powered by GPT-4”.

What These Companies Automated First

The examples are different, but the first jobs share a few traits:

  1. The task already existed. Advisers searched internal research. Support teams answered common questions. Staff drafted reports. Learners practised conversations.
  2. The information could be bounded. The system used firm knowledge, account context, approved workspaces or designed learning scenarios.
  3. The output had an owner. An adviser, employee, support team or learning specialist remained responsible.
  4. The workflow could capture feedback. Source links, escalations, user reports and staff review created a way to find mistakes.

That is a better starting point than asking where a business can “use AI”. Start with one repeated job, the information it needs and the person who approves the result. Our guide to what to automate first gives a practical scoring method.

How A Safe OpenAI Workflow Works

A useful business workflow has more than a prompt. It needs six parts:

  1. Trigger: A question arrives, a support message is opened or a staff member starts a defined task.
  2. Approved context: The system retrieves the records, documents or fields allowed for that task.
  3. Model instruction: OpenAI receives the job, constraints and required output format.
  4. Validation: Rules check required fields, source references, confidence signals or prohibited actions.
  5. Human decision: A person reviews high-risk outputs, exceptions and anything that will affect a customer, employee or payment.
  6. Record: The business stores the input, output, source and decision so the workflow can be checked later.

The model is one component. Identity, permissions, source quality, exception handling and logging decide whether the workflow can be trusted. Those controls belong in the AI implementation plan, not in an employee’s personal prompt library.

What Should Stay Human?

Keep a person in control when the work involves:

  • Financial, legal, clinical or employment decisions
  • Sensitive customer circumstances or complaints
  • A payment, refund, account restriction or contractual commitment
  • Advice that depends on context outside the approved source material
  • A new or unusual case the workflow has not been tested against
  • Customer-facing content with brand, safety or accuracy risk

Human review needs a real action, not a rubber stamp. The reviewer should see the source material, know what the system changed and have a clear way to reject or correct the result. That is the operating model behind human-in-the-loop automation.

What An Established Business Should Copy

Do not copy the scale of these companies. Copy the sequence:

  1. Choose one repeated workflow with an identifiable owner.
  2. Map the current inputs, systems, decisions and exceptions.
  3. Limit the first version to approved data and one useful output.
  4. Put a person at the point where judgement or authority is required.
  5. Test against real cases, including bad inputs and uncommon requests.
  6. Measure accuracy, completion, escalation and rework before expanding.

This creates evidence inside your own business. If the workflow performs well, connect the next source or task. If it creates rework, fix the process before adding more users.

Frequently Asked Questions

Does Using ChatGPT Count As Using OpenAI?

Yes. ChatGPT is an OpenAI product. A company can also use OpenAI through an API, an enterprise workspace or software that embeds an OpenAI model. The implementation matters more than the access method. Business use needs approved data rules, permissions, review steps and an owner for the result.

If casual use is becoming a controlled business process, our ChatGPT consultant guide explains what the role covers and when outside help makes sense.

Are Companies Using OpenAI To Replace Staff?

The verified examples above do not support one simple answer. Klarna automated a large share of service chats, while Morgan Stanley kept advisers at the centre of advice and Duolingo kept specialists writing and reviewing learning content. The right boundary depends on risk, authority, customer impact and the quality of the available information.

What OpenAI Product Should A Business Start With?

Choose the product after defining the workflow. A controlled enterprise workspace may suit staff drafting and research. An API may suit a repeatable workflow inside existing software. A customer-facing feature needs stronger testing, monitoring and escalation. The decision should follow the job, data and control requirements.

Where Should A Smaller Business Start?

Start with one internal, reversible task. Good candidates include retrieving approved procedures, drafting a standard response for review or classifying a repeated request. Avoid beginning with autonomous payments, employment decisions or unsupervised customer advice. Prove the workflow on real cases before connecting more systems.

Build The Whole Workflow

The companies using OpenAI well have built a controlled path from information to action. Echelon AI Solutions maps that path, selects the right tools and builds the controls around it. See how our AI transformation work turns one proven workflow into a wider operating system.

Related Services

  • AI Engineering →

    Production AI systems built for a problem the business already knows it has, then handed over.

  • Xero Integration And Finance Automation →

    Bills drafted from supplier PDFs, statements cross-checked and every entity in one view, approved by your accounts team.

  • n8n Automation Agency →

    n8n workflows and AI agent steps built inside the systems a business runs, kept as code, tested before go-live and documented in plain English.

Sam Fielding
Sam Fielding
Managing Director, Echelon AI Solutions

Sam co-founded Echelon AI Solutions and leads transformation strategy, client engagements and growth. He has built and operated businesses across marketing and AI education, and has guided companies in retail, trades, hospitality and professional services through operational change. His focus is making AI earn its place through measurable business performance.