Skip to main content

Article

AI Adoption in Small Businesses

A practical way for small businesses to improve their operations by gradually incorporating AI.

By Matt Myers · I Develop Tech · October 2026 · 18 min read

Two-thirds of US small businesses now claim they use AI, nearly three times the share in 2023. However, far fewer businesses seem to actually be using it in their day-to-day operations. When small businesses were asked whether they had used AI in any business function in the past two weeks, only about 22% to 36% said yes, depending on their size.

Reported AI use compared with recent use

Share of US small businesses

US Chamber of Commerce, Empowering Small Business 2026 · US Census Bureau, Business Trends and Outlook Survey, two weeks ending Oct 4, 2026, businesses with fewer than 250 employees · separate surveys

In addition, most small businesses have been using AI in a limited capacity. In 2025, the most common uses of AI related to writing content (83%) and personal productivity (61%), while only 21% used AI to automate a process. About half of businesses were still experimenting with AI, and only 7% had fully integrated it into their company operations.

Only 7% of small businesses using AI have fully integrated it

Level of AI adoption among US small employer firms that use AI

49%
Experimenting
44%
Partially integrated
7%
Fully integrated
Federal Reserve, Small Business Credit Survey 2026 · employer firms with 1 to 499 employees that use AI, surveyed Sep to Nov 2025

Research suggests the businesses seeing the largest financial gains from AI are far more likely to have redesigned core workflows around it. Since nearly half of small businesses using AI are still experimenting, many have not realized those gains yet.

For a small business, this can be a daunting and unclear step. What does "redesigning a workflow around AI" really mean? And how does a small business with limited resources actually work towards this milestone?

In a small business training program, many participants cited the "barriers to deeper integration" related to "lack of technical expertise", difficulty "choosing tools", and "data privacy concerns". Nearly 75% of participants in this program said they would benefit from more training and help putting AI in place.

I've observed this same sentiment while working with small businesses over the years. As a result, I've written this article to demonstrate an iterative path a small business can reference, using practical examples.

Four Roles for AI

While every organization's use of AI may differ slightly, for the sake of discussion, we'll classify AI usage into the following four roles:

  1. Assistant: Using AI individually to help out with your tasks
  2. Builder: Using AI to build tools that systematically improve operations
  3. Teammate: Using AI to interpret information within a process for people to act on
  4. Operator: Using AI to make decisions without a person needing to intervene

It's important to note that these roles should be assessed at a task-level. For example, some critical tasks may never be suitable for the Operator role due to risk factors or non-determinism, but they might still benefit from one of the other roles.

The four roles at a glance

Done by AI
  1. Role 1AssistantDrafts work for a person to finish
    Person asks a question or requests a draftAI drafts or answersPerson reviews and uses the result
  2. Role 2BuilderHelps build tools that run without AI
    Person defines a process's rulesAI helps build the automationAutomation runs without AI
  3. Role 3TeammateInterprets and recommends
    New item arrives in a processAI interprets and recommendsPerson decides and acts
  4. Role 4OperatorInterprets and acts on its own
    New item arrives in a processAI decides and acts within limitsExceptions go to a person

This breakdown will help us navigate a few different ideas:

  • To use AI within the workplace you do not need to blindly trust its output
  • You can use AI in a way that mitigates risk and data privacy concerns
  • Different tasks within organizations may be suitable for different roles of AI
  • There is a natural progression path for maturing AI adoption within an organization

To start, let's discuss the most basic way to use AI: using it as an assistant for individual tasks.

AI as an Assistant

Most small businesses that use AI use it in this capacity. To put it simply: A person asks AI for a draft or an answer and decides what to do with the result.

Benefits and challenges of AI drafts

Benefits

  • Quick first drafts

    AI can summarize information and turn rough notes into drafts or answers very quickly.

  • Easy to start

    A person can start with a ChatGPT or Claude account without changing any business process.

  • Useful in any role

    Staff can use it for writing, research, and summaries, whatever their job.

Challenges

  • Inaccurate information

    AI may be trained on outdated information, over-simplify a nuanced topic, or hallucinate.

  • Every user becomes a reviewer

    The person using AI must verify its claims, and may lack the expertise to do so or begin to experience "automation complacency" over time.

  • Corrections stay with one person

    Unless a correction is shared, other team members may get the same incorrect result in the future.

While AI can speed up drafting and research, inaccurate output that gets past a reviewer can be costly. In 2025, Deloitte agreed to refund part of its fee for an Australian government report that cited research that didn't exist. Deloitte had disclosed using generative AI to help produce the report.

Centralized Knowledge

This is where the concept of "Centralized Knowledge" comes into the picture: a shared place for the information that both people and AI assistants work from. It generally holds two kinds of information:

  • Reference material: Facts about the business, such as policies, product details, and process documentation.
  • Instructions: Guidance for a specific task that tells AI how to interpret a request and what the result should look like, such as the format, the tone, or which documents to use.

Having a centralized repository of information can help by:

  • Enabling specific knowledge-holders to maintain info pertaining to their expertise
  • Serving as a source of truth for specific organization information
  • Allowing for iterative improvement of the AI assistants because inaccuracies can be corrected within a shared space

People and AI work from the same centralized knowledge

Example: a store changes its return window

Done by AI

Owner corrects it once

Store manager updates the return window 30 days60 days

Centralized knowledge

Reference material

  • Return policy Updated
  • Product care guide
  • Supplier list

Instructions

  • Customer replies: friendly, short
  • Answer only from store documents

AI reply for a sales associate

"…you can return it within 60 days with your receipt."

AI returns FAQ for a new hire

"How long do I have? 60 days from purchase."

Every later draft uses the updated policy, and each person still checks their draft before using it

Most importantly, new hires and less experienced staff can work from the knowledge of the business's most experienced people.

A related study measured the effect of using AI to spread top performers' knowledge. Among more than 5,000 customer support staff, an AI assistant trained on past support conversations, with extra weight on top performers' chats, raised productivity by 15% on average and by about 30% for less experienced staff.

Whatever tool a business uses, centralized knowledge generally has five elements:

  • A shared home: Everyone works from the same reference material and instructions, kept in a team ChatGPT or Claude project or in a common folder.
  • Clear ownership: The people who know each subject best maintain its material, and everyone else uses it.
  • Version history: A record of changes means a bad edit can be traced and corrected.
  • Answer rules: The instructions tell the AI to answer only from the business's own documents and to flag anything it is unsure about.
  • Business accounts: Employees use company accounts, so business information stays out of personal ones.

Centralized knowledge also gives employees an approved way to use AI. In a global study, 57% of employees said they hide their use of AI at work and present AI-generated work as their own. When a business provides the tools and reference material, employees may have less reason to hide their AI use.

Eligible Assistant Tasks

A task may be eligible for an AI assistant when:

  • A person starts each request, rather than a process triggering it.
  • That person reviews AI's draft or answer and decides what to do with it.

Examples

AI as an assistant

1 / 3

Writing jobs that tend to get postponed

  • Onboarding and training guides
  • Store policies and procedures
  • Descriptions for new products
  • Replies to difficult supplier or customer emails

AI turns rough notes into a first draft

Example: a store's return policy

Done by AI

Owner's rough notes

  • Returns within 30 days
  • Receipt needed
  • Sale items final

AI draws on centralized knowledge

  • Writing instructionsFriendly tone, short sentences
  • Store policiesGift receipts already get store credit

AI draft

"Items can be returned within 30 days with a receipt. Gift receipts get store credit. Sale items are final." Asks: "Do online orders follow the same 30 days?"

Owner

Answers the question, adds a line about online orders, and posts the policy

An edit that applies every time goes into the instructions

Refined over time, this knowledge becomes useful beyond the assistant role. A well-documented process can serve as requirements that instruct AI to build out an automation, which leads to discussing the builder role.

AI as a Builder

As a builder, AI helps create tools for deterministic workflows that might otherwise take too much time or technical skill for a small organization. Once built, these tools often don't need to call an AI model to run, so a business can benefit from AI without depending on it day to day.

Benefits and challenges of AI-built automations

Benefits

  • Consistent results

    The automation follows the same steps every time, so its results don't vary the way an AI model's answers can.

  • Predictable cost

    The automation runs without calling an AI model, so there's no token cost on each run. In my experience, unexpected usage costs are one of the most common surprises for businesses running AI in their processes.

  • Data privacy

    AI can help build the automation using test data, so sensitive data doesn't need to reach an AI model and the automation can run within your private network.

Challenges

  • Rules have to be defined

    Cases that staff handle from memory, or that go unhandled, have to be defined before the automation can follow them.

  • Technical upkeep

    Someone has to build, test, and update the automation, usually a technical employee or contractor.

  • New scenarios

    Situations the rules don't cover will come up after launch and need to be added.

To better understand the builder role, let's walk through an example scenario:

How many people picture automating with AI

Example: one online order

Done by AI
Order: 2 candles, gift wrap, rush shipping
AI decides how to fulfill it
Order is handled

This seems simple enough. However, there are some subtle concerns that can make this approach less desirable:

Concerns with AI handling every order

Example: the same online order

  • Results can varySimilar orders may be handled differently
  • Token cost on every orderCosts rise with order volume
  • Order data sent to the AI modelCustomer details leave the business's systems

This is where using AI as a builder may be relevant. Instead of using AI to interpret and make decisions on each order as it comes in, AI can be used to plan and build out a standard automated process that aligns with your fulfillment rules:

Using AI to build a standard automation

Example: the same online order

Done by AI

Once, before launch · built with test orders

AI helps build the automation
People test it and deploy it

Every order after that, on the business's own systems

Order: 2 candles, gift wrap, rush shipping
Automation applies the fulfillment rules
Same steps every time · no AI model involved
Order is handled

Orders that fall outside the rules are flagged for staff

Testing and Analysis

By using AI in this capacity, we can:

  • Analyze the rules and uncover gaps that the known rules don't cover
  • Validate the automation using test data before launch to increase confidence
  • Generate up-to-date documentation so staff can understand and inspect the workflows

As new scenarios are uncovered, staff can use AI to refine the automated process's rules to better handle these situations. Using AI as a builder in this way creates an iterative path for AI to improve operations with fewer of the risks of inconsistent answers, unpredictable costs, or data privacy concerns.

Eligible Builder Tasks

A process may be eligible for an AI-built automation when:

  • Its steps can be defined as rules that apply the same way every time.
  • Its inputs are structured data rather than free-form text.

Examples

AI as a builder

1 / 3

Reordering inventory

A local store needs to reorder inventory regularly. In my experience, the forecasting and purchasing features in retail systems often miss the more nuanced situations like special orders, backordered items, and differences in supplier cost and timing. This generally leads to staff spending a large amount of time either assembling or correcting these orders by hand. By using AI in a builder capacity where it writes most of the code, the store's technical employee or contractor can build a reordering automation that accounts for these factors.

After launch, staff can still override an order when a supplier is late or a seasonal item sells faster than expected, and those cases get added to the rules over time.

Suppose staff spend about four hours every week building purchase orders. With the ordering rules written down, the store's technical employee or contractor might spend around ten hours working with AI to build the automation. If a person still spends about 30 minutes a week reviewing the orders, the store recovers those ten hours in about three weeks and saves roughly three and a half hours every week after that. If a contractor builds it, compare their fee with the staff time saved each week.

AI builds a weekly purchase order automation

Example: a store's weekly purchase order

Done by AI

Building it · person and AI together

  1. 1 Store owner Describes how orders are built today "Order enough for six weeks, more before holidays"
  2. 2 AI Drafts the automation from those rules
  3. 3 AI Asks about cases the rules don't cover "What should happen when an item is backordered?"
  4. 4 Store owner Defines the missing rules "Skip it and reorder when the supplier restocks"
  5. 5 AI Updates the automation and runs it on past weeks
  6. 6 Store owner Compares its orders with what staff actually ordered

Repeat steps 3 to 6 until the orders match how staff would order

Every week · no AI model involved

Sales and stock data

Sold, last 4 weeks
48
On hand
6
Supplier lead time
10 days

Automation's draft order

Candles
40
Special order
1 blue vase
Mugs
Skipped, backordered

Staff adjust

Raise candles to 60 for a holiday weekend, then send. Overrides like this go back into the rules.

However, there are limits to this approach. Although AI-built automations can prove reliable, they only work for deterministic workflows where someone has defined the rules in advance. Many other business tasks may be ineligible for this approach because they require interpretation or are based around unstructured data. For these tasks, you may need to evolve your usage of AI and incorporate it more like a teammate.

AI as a Teammate

As a teammate, AI can interpret dynamic inputs and make recommendations as they arrive. In this capacity, AI's output generally aims to enrich a person's task with context or recommend actions to streamline it. A person can then use this output as a draft or discard it if it isn't helpful. The output should generally be internal, such as context added to an internal work item or a drafted reply (needing review) to a customer inquiry.

Benefits and challenges of AI interpretation

Benefits

  • Help where rules fall short

    AI can interpret free-form requests, so processes that can't be fully automated still get help.

  • Enriched context

    AI can gather related information from other systems and add it to the work item, so people spend less time searching.

  • Faster common cases

    For routine items, AI's recommendation may already be right, so staff can confirm it and save their attention for unusual cases.

Challenges

  • Results can vary

    The same input can lead to a different interpretation each time the process runs.

  • Costs grow with volume

    AI runs on every item, so token use rises with the amount of work.

  • Data reaches the AI model

    Each item's content is sent to an AI model unless the sensitive parts are removed first.

Used this way, AI can support staff by gathering context and recommending actions for each item. However, because teammate workflows rely on AI models, it's important to mitigate some of these challenges.

Keeping People in Control

The following practices serve as quality-control levers organizations can use to limit these challenges:

  • Strip sensitive data: Before the AI gets the data to interpret, sensitive information is stripped to ensure data privacy or compliance requirements
  • Interpretation and drafts: AI interprets, matches, summarizes, or drafts, and people decide what to do with the result.
  • Limited access: AI can read what it needs and add notes or drafts, but it can't send messages, close tickets, or edit records.
  • Approval for customer-facing output: A person approves anything that reaches a customer or changes a record before it takes effect. Internal context, such as ticket notes, can be corrected after the fact.
  • One run per item: Running AI once per ticket or order keeps token costs in line with the volume of work. Clear limits in the instructions, such as which sources to check and how long the output should be, also keep AI's work predictable and its token use in check.
  • Tracked corrections: The corrections people make to the AI's output show where AI's instructions need refinement.

Eligible Teammate Tasks

In my experience, a task may be eligible for an AI teammate when:

  • It is a recurring step within a process.
  • Its inputs require interpretation that fixed rules can't perform reliably.
  • A person handles each item and acts on AI's result.

Examples

AI as a teammate

1 / 3

Matching new bug tickets to earlier reports

At a tech company, a new bug ticket often starts with little more than a title and one line of description. Whoever picks it up may spend time working out whether the same problem has been reported before. AI could read the ticket as soon as it's created and look through the support inbox, past issues, and the backlog for earlier reports of the same problem, even when they use different words, and check the CRM for affected customers. It could then add the likely matches, the affected customers, and a short summary to the ticket.

Developers and product owners would still review the ticket and decide what to do, such as when to schedule the fix, but AI could cut the time they spend searching. Half of developers surveyed said they lose 10 or more hours a week to non-coding tasks, and finding information was the top time-waster they named.

AI gathers context for each new ticket

Example: a one-line bug report

Done by AI

Support rep logs a ticket

"Export button does nothing"

AI looks for related information

  • Support inbox3 similar emails this week, all from Safari users
  • Issue trackerA bug from March: export fails in Safari, marked fixed
  • CRMOne affected customer renews next month

AI adds its findings to the ticket

Likely cause
The March Safari bug is back
Affected
3 customers, 1 renewing soon
Summary
Export fails in Safari

Product owner

Moves the fix into this sprint because of the renewal

Developer

Starts from the March fix instead of from scratch

Notes on what was missing or unhelpful refine AI's instructions

In the teammate role a person is still responsible for approving each action. While this approval step lowers risk, it can also bottleneck certain high-volume tasks. Once a process is stable and AI's output no longer needs frequent refinement, the business can let AI take the action itself as an operator. An operator can be viewed as a more experienced teammate, since it builds on the recommendations generated in that role. However, letting AI act without approval raises the stakes, so it comes once AI has proven to be reliable within a specific task.

AI as an Operator

As an operator, AI makes decisions and acts on them without a person approving each decision. AI in this role is often called an agent. The agent acts on at least some routine decisions itself, and people review the cases it can't resolve within its limits.

Benefits and challenges of AI decisions

Benefits

  • No waiting on approvals

    Routine decisions are handled as work comes in, so high-volume tasks don't queue up for a person.

  • Staff focus on exceptions

    People spend their time on the cases the rules don't cover.

  • Rules improve over time

    Escalated cases show where the written rules need work.

Challenges

  • Mistakes go unnoticed

    With no one approving each decision, a mistake can reach a customer or another system before anyone notices.

  • Results still can vary

    Even stable rules can encounter unknowns and produce different outcomes than expected.

  • Costs still can be hard to predict

    Token use depends on how much work each decision takes.

Letting AI decide routine cases can remove the wait for approvals, but this is also the hardest role to get right. Few companies, even large ones, have given AI this much responsibility. At US companies with 1,000 or more employees, two-thirds of departments use AI agents, but only 17% of departments let an agent complete a task without a person approving each step.

Approved Actions and Escalation

When an agent acts without approvals, these safeguards keep its decisions within bounds:

  • Small, reversible decisions first: Flagging, routing, or holding an order is safer than issuing a refund or canceling the order.
  • Action limits: Define which actions are allowed and set caps, such as a maximum order value, so anything beyond them goes to a person.
  • Written rules: The agent follows documented rules that build on the centralized knowledge from the assistant role.
  • Escalation: When a case doesn't fit the rules, the agent hands it to a person.
  • Logging: The agent records every action, and someone regularly reviews a sample of routine decisions along with every escalated case.
  • An off switch: Staff can pause the agent and return to the manual process.

These safeguards are a starting point rather than a full implementation plan. Putting an agent into production involves more technical detail, which is where a technical employee or contractor can help.

In my experience, businesses do best with AI as an operator when they treat an agent the way they would treat a new employee, by gradually increasing its responsibility as it proves itself. The teammate role works like a trial period for this: while people still make each decision, they can see how often AI's recommendations need correcting. Once corrections are rare, and spot checks confirm reviewers are still catching mistakes, the agent can take on the cases that are easiest to reverse. From there, its scope of responsibility can grow as it proves successful.

Gradually increasing AI's responsibility

Example: share of cases AI handles without approval at each stage

  1. Stage 1 , AI as a teammate

    AI recommends, and people make every decision

    Without approval: 0%

    People correct about 1 in 4 recommendations

  2. Stage 2 , AI as a teammate

    People refine AI's instructions from their corrections

    Without approval: 0%

    Corrections drop to about 1 in 25

  3. Stage 3 , AI as an operator

    AI handles cases people define as low-risk

    Without approval: 40%

    Everything else goes to a person; staff review a sample

  4. Stage 4 , AI as an operator

    People add more cases to AI's defined scope

    Without approval: 85%

    Unknowns and listed exceptions go to a person

AI handles within its defined scope People decide Return to an earlier stage if problems appear

As the agent takes on more work, it's also a good practice to have a "manager" for AI operators who periodically reviews its output for feedback and improvement opportunities.

Customer-Facing Decisions

When an agent's decisions reach customers directly, a mistake is harder to contain and can affect how customers see the business. A support chatbot is a common example, because each answer it gives is a decision about what to tell a customer.

Customer-facing agents often have to address two primary risks. The first is legal liability. Tribunals and courts have started holding businesses responsible for what their chatbots tell customers. In 2024, a Canadian tribunal ruled against an airline on this point.

Jake Moffatt asked Air Canada's chatbot about bereavement fares after his grandmother died. The chatbot told him he could request the bereavement discount after his flight, but the airline's actual policy said otherwise. When Air Canada refused the discount, the airline argued that the chatbot was responsible for its own actions. The tribunal disagreed:

"It should be obvious to Air Canada that it is responsible for all the information on its website."

Moffatt v. Air Canada, 2024 BCCRT 149

In May 2026, a German higher regional court reached a similar conclusion about a company's website chatbot.

The second risk is service quality. Klarna said in 2024 that its AI assistant was doing the work of 700 full-time customer service staff. A year later, Klarna's CEO said cost savings had played too large a role in the company's move to AI customer service, and that service quality had suffered as a result. He described investing in human support as the way forward.

Both risks shrink when the chatbot answers only from approved content, hands uncertain questions to a person, and makes a person easy to reach.

Eligible Operator Tasks

A task may be eligible for an AI operator when:

  • Its results have proven reliable, and people rarely correct them.
  • Its decisions can stay within defined rules and limits.
  • Cases outside those limits can be handed to a person.

Examples

AI as an operator

1 / 3

Scheduled checks on orders and listings

An online retailer's technical employees can use AI to write down the rules the business uses to evaluate orders and listings, such as what counts as a late shipment, which order values need approval, and which problems need a person. An agent then reviews current orders and listings on a schedule. It applies the written rules and makes judgment calls that fixed rules handle poorly, such as whether a product description is incomplete or a customer's note signals a problem. Orders and listings that pass go through without individual review, and the agent sends the rest to staff.

Ideally, most orders and listings pass, and staff focus on the few that don't.

AI handles routine orders on a schedule

Example: one morning's scheduled check

Done by AI

Every morning

The agent runs its check using the team's written rules

Agent checks

  • Order systemShip-by dates for every open order
  • Carrier trackingOrder 5121 hasn't been picked up
  • Product listingsNew listings and their descriptions
  • Order 5120On trackContinues
  • Order 5121Will ship 3 days lateSent to staff
  • Order 5122On trackContinues
  • Listing: ceramic vaseDescription just says “TBD”Sent to staff

The team updates the rules from what it sees

Closing Thoughts

After years of wide availability, it seems that few small businesses have successfully integrated AI into their operations, even though the businesses seeing the largest financial gains are far more likely to have done so.

This appears to follow an existing trend where the most intricate technological evolutions are first adopted by larger companies that have more budget, staff, and in-house expertise to dedicate to them.

However, AI has the potential to be different. It can ease some of those limitations by filling in missing technical knowledge and iteratively solving problems at a reasonable cost. In my experience, one technical person working with AI can now build tools that once needed a larger team. This may make AI one of the more useful technologies small businesses have had access to in a long time.

The bottleneck has moved as a result. For most small businesses, the bottleneck is now the time it takes to define their workflows and willingness to evolve their operations. A business that writes down their operational workflows can now automate or improve them quickly with AI's help. That one-time effort can then save staff time every week.

In addition, documentation used to be de-prioritized and viewed as a tedious administrative step, whereas now it might just be the first tangible step you can take to improve your business:

  1. If you aren't using AI yet, start with it as an Assistant, and use it to document and refine your workflows along the way.
  2. Use that documentation to automate time-consuming tasks with AI as a Builder.
  3. Add AI as a Teammate to streamline tasks that need interpretation, with recommendations that follow your procedures.
  4. Once those recommendations prove reliable, use AI as an Operator to automate tasks that fixed rules can't.

Unlike many technologies, AI offers an incremental path where each step can add value on its own. Whether you aren't using AI yet or already use it regularly, I hope this article shows how you can gradually incorporate it into your business operations as the technology keeps advancing.

About the author

Matt Myers

I Develop Tech

Matt Myers has built software since 2013 and has led engineering teams since 2016, including CTO roles since 2020. Through I Develop Tech, Matt helps small businesses with system integrations, cloud infrastructure, and AI adoption.

Connect on LinkedIn

Sources

Figures were checked against each primary report on Oct 19, 2026.

  1. US Chamber of Commerce, Empowering Small Business, Oct 2026 (report)
  2. US Census Bureau, Business Trends and Outlook Survey, Oct 2026 (data, May 2026 summary)
  3. Federal Reserve Banks, Small Business Credit Survey: 2026 Report on Employer Firms, Mar 2026 (report)
  4. McKinsey, The State of AI in 2026: On the Road to ROI, Aug 2026 (report)
  5. Goldman Sachs 10,000 Small Businesses Voices survey of program participants, Mar 2026 (press release)
  6. Google Cloud, What Are AI Hallucinations? (article)
  7. Parasuraman and Manzey, Complacency and Bias in Human Use of Automation, Human Factors, 2010 (article)
  8. Deloitte Australia report refund, Oct 2025 (OECD AI incident record, The Guardian)
  9. Brynjolfsson, Li, and Raymond, Generative AI at Work, Quarterly Journal of Economics, 2025 (article, working paper)
  10. KPMG and University of Melbourne, Trust, Attitudes and Use of AI: A Global Study 2025 (report)
  11. Atlassian, State of Developer Experience 2025 (report summary)
  12. Wharton and GBK Collective, The Adaptive Advantage, Oct 2026 (report)
  13. Moffatt v. Air Canada, 2024 BCCRT 149 (decision)
  14. Higher Regional Court of Hamm, 4 UKl 3/25, May 2026 (press release, German)
  15. Klarna, AI assistant announcement, Feb 2024 (press release)
  16. Fortune, Klarna CEO remarks to Bloomberg, May 2025 (article)