How AI changes business processes and business models

In short: AI changes a company when it reaches the processes and the way the company sells, not when another subscription is added. For an online store that means customer service wired to order data and a catalogue that AI shopping agents can read; for a B2B company, faster quotes, new services and outcome-based pricing.

Key takeaways

In 2025, 20% of EU enterprises with 10 or more employees used AI, up from 13.5% in 2024. Adoption is growing; value does not follow automatically.
In Italy the share doubled in a year, from 8.2% to 16.4%, but one in three companies using AI does not link it to any specific business area: use is still experimental.
The return depends on complementary investments (data, processes, skills) that cost money early and are hard to see.
In e-commerce, AI is becoming a sales channel: AI assistants can already complete purchases.
In B2B, pricing is shifting: some vendors charge per result instead of per licence.

For years, AI in business was a line in the IT budget: a chatbot, an email assistant, a CRM plugin. Useful tools that leave the rest untouched, from how an order is taken to how a price is set. According to Eurostat, the most common uses among European companies in 2025 were analysing written text (11.8%), generating images, video or audio (9.5%) and generating written or spoken language (8.8%): single tasks, handled one at a time.

For the basics on use cases, tools and governance, see the guide to generative AI for business. This article goes one step further: processes and business models. More articles are in the AI for business section.

What is the difference between using AI as a tool and changing your business model?

A tool makes one task faster, a process reorganises a chain of tasks, and a business model changes what you sell, to whom and how you get paid. Many companies stop at the first level.

A business model is the logic by which a company creates value for customers, delivers it and keeps part of it as revenue. A process is the sequence of activities, people and systems that produces a repeatable result, such as fulfilling an order or issuing a quote.

Level	What changes	E-commerce example	B2B example
Tool	A single task gets faster	Product descriptions drafted with AI and edited by hand	Draft sales emails or call summaries
Process	The workflow and who runs it	Support that reads order status and tracking and closes simple cases on its own	Quote built from price list, customer history and stock, approved by the sales rep
Business model	What you sell, through which channels, how you charge	Catalogue sold inside AI assistants as well as on the website	Technical support sold as a subscription or paid per result

Each level sits on the one below. An AI-based business model without solid processes underneath is a promise customers soon discover they cannot cash in.

Why doesn't buying AI software deliver results on its own?

Because the value depends on investments that never show up on the vendor's invoice: clean data, redesigned processes, trained people. Without them, AI speeds up work that was organised for another era.

Economists Brynjolfsson, Rock and Syverson call this the productivity J-curve: general purpose technologies, AI included, need intangible investments in new processes, business models and skills. Early on these investments absorb resources and measured productivity looks flat; the gains come once the organisation has adapted.

Klarna, the Swedish buy-now-pay-later company, shows both sides. In February 2024 it announced that its AI assistant had handled 2.3 million conversations in its first month, two thirds of customer service chats, the equivalent work of 700 full-time agents, with resolution time down from 11 minutes to under 2. In May 2025 its CEO acknowledged that the aggressive use of AI had lowered service quality and that Klarna was hiring human agents again, so customers can always reach a person. Redesigning a process also means deciding where people stay in the loop.

How many Italian companies use AI, and what holds them back?

In 2025, 16.4% of Italian companies with at least 10 employees used at least one AI technology: twice the previous year, but still below the EU average of 20%. The main barrier is not the technology but skills.

The figures come from the Istat report on enterprises and ICT 2025: the share was 5.0% in 2023 and 8.2% in 2024. The size gap is widening: AI is used by 53.1% of large companies and 15.7% of SMEs. The areas where AI is adopted most often are marketing and sales (33.1%), organisation of administrative processes (25.7%) and research, development or innovation (20.0%).

One figure shows how often companies stop at the first level, AI as a tool: 33.4% of companies using AI do not indicate any business purpose among those listed (up from 15.5% in 2024). Istat reads this as adoption that is widespread but poorly structured, still early and experimental. In practice, many companies have the tools but have not yet tied them to a process.

Among companies that do not use AI, 11.5% considered it without going ahead. These are the barriers they report, and how a real project can address them:

Barrier reported (Istat 2025)	Share	How a project addresses it
Lack of skills	58.6%	Role-based training and a cross-functional team, with an external partner in the first phase
Unclear legislation	47.3%	Classify each use case under the AI Act before starting: according to the European Commission, the vast majority of AI systems in use today fall into the minimal or no risk category
Data unavailable or of poor quality	45.2%	Start from a process with reliable data (orders, price lists) and fix the rest in parallel
Privacy and data protection	43.2%	Choose vendors with data processing agreements and decide which data never leaves the company
High costs	43.0%	A limited pilot, with cost per handled case compared with today's
Ethical concerns	25.7%	Written rules on human oversight and transparency towards customers

How does an online store change when AI reaches its processes and sales channels?

It changes in two places: customer service, which gets connected to order data, and sales, which start to flow through AI assistants as well as the website.

A generic chatbot saves little. One connected to the store platform, the ERP and the carrier can say where a parcel is, start a return according to policy, suggest an alternative to an out-of-stock item and hand complaints and high-value orders to a person. The method is the one behind business process automation: map the workflow first, then automate it.

Agentic commerce is selling through an AI assistant that searches, compares and buys on the user's behalf. The Agentic Commerce Protocol, an open standard developed by Stripe and OpenAI, lets a store make its checkout available to compatible agents, starting with ChatGPT, while remaining the merchant of record and deciding which products to sell. Stripe's documentation describes integrations of this kind, fed by a structured catalogue feed.

Product pages now have to work for an agent that reads them, too: complete attributes, current stock, prices, delivery times and clear return terms. It is the same work you need to show up in AI-powered search engines.

Which new business models does AI open up for B2B companies?

There are four practical routes: turn speed into a selling point, move from product to service, offer services that learn from customer data, and get paid for results.

Speed as part of the offer

In many B2B companies the bottleneck is the quote: price lists, customer discounts and stock live in different systems. A workflow that gathers this data and prepares a draft, leaving approval to the sales rep, shortens the sales cycle. If a competitor replies in three days and you reply in three hours, speed becomes part of the offer, especially in B2B e-commerce, where buyers expect real-time prices and availability.

From product to service

A machinery manufacturer owns manuals, data sheets and service history. An assistant working on those documents can provide first-line support around the clock, and what used to be a cost of the technical department can become a support subscription.

Services that learn from customer data

Every order, request or service call adds data that makes the next offer more precise: this is how AI turns data into decisions, not just text. In B2B e-commerce it means product ranges and reorder suggestions built on each customer's history; for a distributor, demand forecasts that drive stock and purchasing; for a manufacturer, maintenance planned on machine usage data instead of a fixed calendar. Retail examples are in the guide to AI in retail.

Once data becomes part of the product, the contract has to cover it. For connected products, industrial machinery included, the EU Data Act, applicable since 12 September 2025, gives users, business users included, the right to access the data generated by their use and to share it with third parties, such as an independent repair shop. A support service built on that data has to win on quality, not on exclusivity. When the data concerns people, GDPR rules on legal basis and information notices apply too.

Outcome-based pricing

Some software vendors already do it: Intercom charges $0.99 per outcome for its Fin agent, at most one per conversation, and nothing if Fin fails to resolve the issue or the customer asks for a human. The same logic works for B2B services: a fee tied to processed cases, closed tickets or qualified leads rather than hours worked, provided the result is measurable and agreed upfront. What this means for companies paying for software subscriptions is covered in the article on AI agents and SaaS.

How do you prepare people and the organisation for an AI-based business model?

With widespread skills, a team that brings different departments together and clear rules on what AI may do. Technology is the easy part: changing a business model takes experimentation and a fresh look at internal habits and incentives.

It is not just good practice. Article 4 of the AI Act, in application since 2 February 2025, requires organisations using AI systems to take measures to develop the AI literacy of their staff and of anyone using those systems on their behalf. According to the European Commission's Q&A, after the Digital Omnibus amendments that entered into force in mid-July 2026 the obligation remains but no specific level is mandated, supervision applies from 2 August 2026 and no certificate is needed: an internal record of training activities is enough.

A cross-functional team. The process owner (sales or support), the people who know the data and systems (IT, ERP, store platform) and someone who assesses contracts and privacy. With a single lead accountable for results.
Role-based training, not generic courses. The person writing product pages, the one approving quotes and the one handling cases escalated from AI need different things.
Room to experiment, within limits. One customer segment, a budget and a review date: an experiment that fails is information, not something to hide.
Consistent incentives. If support is measured only on tickets closed, AI will be used to close tickets, not to solve problems. Targets must be rewritten along with the process.
Written rules. Which data can go into which tools, who checks the output, when a person steps in. The full framework is in the guide to AI governance at scale.

The role of senior management is covered in the guide to AI leadership. When the new business model risks clashing with the existing one, a dedicated structure is also an option: the pros and cons are in the article on a separate AI company or an in-house team.

How much does it cost to bring AI into processes and the business model?

The licence is the most visible cost, but not the most important one. The real cost lies in integration, data and people, which is why it pays to think in cost per handled case rather than monthly fees.

According to Istat, high costs hold back 43.0% of the companies that considered AI without adopting it. Before deciding, list every item:

Per-user licences for assistants built into email, CRM or the store platform.
Model usage: APIs are billed by volume processed, per million input and output tokens, as shown for example in Anthropic's API pricing. Costs grow with volume.
Outcome-based fees, as with support agents billed per resolved case (see the Fin example in the B2B section).
Integration and automation: connecting the store, ERP, CRM and carrier, and keeping those connections working as systems change.
Data: completing catalogue and price lists, digitising manuals and service history.
People: training, the time key staff spend on the pilot, oversight of cases handed to a human.

The useful comparison is the total cost per quote, ticket or order before and after the pilot, measured on the same metrics. If you operate in Italy, part of the investment may qualify for digitalisation grants.

How to apply it in your company: first steps

Pick a customer-facing process that slows down sales or support today: quotes, order enquiries, product pages. Measure the baseline: time, volume, errors.
Clean up the data that feeds it. Price lists, catalogue, order history and technical documents must be current and reachable through an API.
Redesign the workflow before automating it. Decide what the machine does, what a person approves and when the customer talks to a human. Tools like n8n or Make connect the systems while AI models read and write: this is what my automation and AI solutions are about.
Run a 6-8 week pilot with metrics set in advance: response time, quote win rate, resolved tickets, customer satisfaction. For choosing the metrics, see how to measure the value of AI.
Only if the pilot holds up, think about the business model: a new channel, service or price, tested first on one customer segment.

A scenario: a B2B store selling supplies to restaurants, with around 1,500 products and customers who reorder every week. The first step is not a chatbot but a workflow that pulls data from the ERP and offers each customer their usual reorder with current availability. If it works, the next step is an automatic reorder subscription: a new business model built on a process that has already proved itself.

Mistakes to avoid

Measuring savings only. Cutting costs at the expense of service is paid for in lost customers: always track a quality metric too.
Automating a messy process. If nobody knows who approves a discount, AI will only make the confusion faster.
Hiding the human. Complaints, disputed returns and B2B negotiations must be able to reach a person.
Leaving the pilot unchecked. Without metrics, an owner and a review date, an AI project becomes a fixed cost nobody can evaluate.
Changing prices without data. An outcome-based offer without history on timing and results can leave you working at a loss.
Forgetting the rules. The EU AI Act transparency obligations have applied since August 2026: people using a chatbot must know they are talking to a machine.

Frequently asked questions (FAQ)

What does it mean that AI changes the business model?

It means AI changes what a company sells, through which channels or how it charges, beyond speeding up internal tasks. Examples include a catalogue sold through AI assistants or technical support billed per result.

Where should an online store start with AI?

With high-volume, customer-facing processes: order and return enquiries, product pages, purchase suggestions. Complete, up-to-date catalogue data comes first, because AI shopping agents will read the same data.

Which B2B uses of AI deliver measurable results?

Preparing quotes from price lists and customer history, first-line technical support based on manuals, and qualifying inbound leads. They share a result you can count: response time, deals closed, tickets resolved.

How long does it take to see a return on AI?

A pilot on a single process shows first numbers in 6-8 weeks. Changing the business model takes longer, because it needs new data, skills and processes: costs come first, benefits later.

Will AI replace customer service and sales staff?

It reduces manual work on repetitive tasks, but sensitive cases stay with people. Klarna, after handing two thirds of its chats to AI, started hiring human agents again in 2025 to keep service quality high.

What is agentic commerce?

It is selling through an AI assistant that searches, compares and completes the purchase on the user's behalf. Open protocols such as the Agentic Commerce Protocol let stores sell through these agents while remaining the merchant of record and keeping the customer relationship.

What are the risks of an AI-based business model?

Inefficiencies in the first months while processes and people adapt; investments that do not pay off without metrics and reviews; worse service if human contact is removed; obligations on personal data and under the AI Act. You reduce them by starting with a measured pilot on a single process, with an owner and a review date.

Sources

If you want to find out which process in your company could change the way you sell, start with a free diagnosis.