The 10 most effective ways to use AI

1. Introduction: AI's value comes from repetition, not experimentation

AI is often talked about as if it were one giant leap into the future. In reality, the biggest benefits come in a far more everyday way: when AI is plugged into recurring processes that eat up time, pile up on desks and tie up people's energy in tasks where a machine can do the first 70 percent.

So AI doesn't create value when someone tries ChatGPT once and is amazed. Value is created when AI becomes part of your workflows.

According to McKinsey's 2025 State of AI survey, organizations are still struggling to move from pilots to impact at scale, and realizing value from AI is tied above all to redesigning workflows, not just to rolling out tools.^[1]

AI isn't a magic wand. It's a lever. But a lever only works if you place it in the right spot.

2. ROI thinking: why isn't all AI use equally valuable?

In business, AI should be talked about less in the language of wonder and more in the language of returns. ROI, or return on investment, measures what you get back for what you put in. In an AI project, it asks: What benefit do we get compared with what the rollout costs?

The benefit can show up in four ways:

as time saved
as fewer errors
as higher revenue
as better decision-making

But not all benefits are equally easy to measure. In finance automation, savings can show up quickly: less manual invoice processing, fewer errors, faster turnaround. In sales, the potential may be much larger, but also less certain. With agent systems, the promise is enormous, but so is the risk.

Gartner estimated in 2025 that over 40 percent of agentic AI projects will be scrapped by the end of 2027, due to factors such as costs, unclear business value and “agent washing.” At the same time, Gartner predicts that by 2028, 15 percent of day-to-day business decisions will be made autonomously through agentic AI.^[2]

In other words: the bigger the promise, the higher the bar for leadership.

3. Four levels: reliable savings, growth, systemic intelligence and multipliers of understanding

It makes sense to divide AI use into four levels.

Level 1: Reliable savings and fast payback. AI handles the volume and a human checks the exceptions. Usually the best place to start.

Level 2: Growth levers and revenue streams. AI helps you sell more, serve better and scale communication. The potential is big, but the human role becomes more important.

Level 3: Systemic intelligence and multi-agent systems. AI moves from the user interface into back-end systems. The value can be large, but rollout requires designing processes, data and oversight.

Level 4: Multipliers of understanding. AI improves learning, analysis and decision-making across the whole organization. The impact doesn't always show up right away as a single line item, but it can raise the level of the entire organization.

Level 1 — Reliable savings and fast payback

1. Finance and accounting automation

Finance is one of the best starting points for AI, because the processes are repetitive, the data is structured or semi-structured, and errors have a clear cost.

AI can read receipts and purchase invoices and identify the supplier, amount, VAT rate, date and expense type. It can suggest account postings based on past data and flag anomalies: an odd amount, the wrong tax rate, an unusual supplier or a missing approval.

According to the 2025 Accounts Payable Automation Trends report, the use of AI in purchase invoice processes rose to 29 percent of respondents, up from 7 percent a year earlier, and 51 percent were considering adopting AI within the next 12 months.^[3]

Where does the ROI come from? Less manual processing time, fewer errors and corrections, faster reporting and better visibility into cash.

What do humans still need to do? The human role doesn't disappear — it shifts from routine to interpretation: borderline tax cases, new suppliers, unclear expenses, unusual purchases, and making sure internal controls work.

Good finance AI isn't an autonomous accountant. It's a tireless assistant that never gets bored of stacks of receipts.

2. Managing customer service volumes

Customer service is a classic ROI target for AI. The higher the message volume, the clearer the benefit. AI can classify incoming messages, identify urgent cases, summarize long conversations, suggest replies to customer service reps, answer routine questions on its own, add a conversation summary to the CRM and route sensitive cases to a human.

The key word is “route.” Not everything should be automated. An irritated customer, a complaint, a contract dispute or a major account may require a human.

Several benchmark and industry reports stress that AI has the most impact when the knowledge base is in good shape, the handoff between human and machine works, and the metrics track the quality of resolutions, not just response speed. Reports on evaluating customer service AI agents warn against measuring the wrong things at the wrong time, because that easily leads to stalled pilots and wasted potential.^[4]

Where does the ROI come from? Lower cost per ticket, faster response times, higher rep productivity, more consistent service quality.

The biggest risk is the wrong kind of automation. If AI answers incorrectly, too confidently or coldly, the customer doesn't think “the AI failed.” They think: the company doesn't care.

Start like this: an internal reply assistant, human-approved replies, a limited set of routine questions, clear escalation to a human, continuous quality monitoring.

3. Procurement and contract management — hidden potential

In many companies, procurement is an underrated gold mine for AI. Why? Because it doesn't just save working time. It can save money directly.

AI can go through contracts, supplier terms, pricing models and purchase history. It can automatically identify unusual payment terms, automatic price increases, termination windows, liability clauses, overlapping suppliers, fragmented spending, contract renewal dates, risky terms and opportunities to put contracts out to tender.

In 2025 procurement reports, spend analytics and contract management stand out as the main focus areas for AI. Procurement Magazine reported, citing the 2025 ProcureCon CPO report, that 90 percent of procurement leaders are either exploring or implementing AI agents to streamline procurement workflows.^[5]

AI doesn't get tired of reading 80-page contracts. It can surface anomalies that a busy human would miss.

Where does the ROI come from? Better prices, reduced contract risk, less supplier fragmentation, better payment terms, faster tendering, managed renewals.

In procurement, AI's value is often less flashy than in marketing. But it can show up much more clearly on the income statement.

Level 2 — Growth levers and revenue streams

4. Expanding existing customer accounts

Acquiring new customers is expensive. Retaining and growing existing customers is often more profitable.

AI can help identify signals in customer data that a salesperson or account team doesn't have time to notice early enough. It can analyze usage data, support requests, billing history, NPS feedback, customer meeting notes, contract renewal dates, feature usage, changes at the customer company and anomalies in purchasing behavior.

This can produce two valuable signals: churn risk and upsell opportunity.

For example, AI might notice that a customer actively uses the product in only one team, even though the company has five teams the same solution would suit. Or it might detect that usage has dropped, support requests have increased and the decision-maker has changed. The salesperson doesn't have to guess whom to contact — AI can suggest priorities.

Where does the ROI come from? Lower churn (customer attrition), more upselling, better account prioritization, less prep time for salespeople, better timing.

But there's one condition: the data has to be in good shape. If the CRM is full of old, incomplete and contradictory information, AI scales the confusion.

In sales, AI doesn't replace the relationship with the customer. It helps the salesperson come into the conversation better prepared.

5. Scaling marketing and communications

For many, marketing was their first contact with generative AI. And the reason is clear: AI is good at writing first drafts.

It can help with campaign ideas, audience-specific messaging, blog outlines, newsletters, social media posts, ad variations, SEO drafts, translations, personalization and A/B test design.

But marketing also has a major pitfall. AI can multiply the amount of content you produce. It doesn't automatically multiply how interesting it is.

Deloitte describes a paradox in its 2025 enterprise surveys on generative AI: investment keeps growing, but for many, ROI remains elusive. Leading organizations go after quick wins from well-chosen use cases, but long-term value comes from more strategic change, skills and adoption.^[6]

In marketing, this means: AI can write. But a human has to know what's worth saying.

Where does the ROI come from? Faster content production, better channel-specific adaptation, lower translation and localization costs, faster campaign testing, more effective personalization, better sales enablement materials.

The biggest risk is sameness. If everyone uses the same tools in the same way, the market fills up with mediocre text. Good use of AI in marketing doesn't start with a prompt — it starts with a point of view.

Level 3 — Systemic intelligence and multi-agent systems

6. Multi-agent systems and process automation

This is where we move into the next phase of AI. A single prompt is no longer enough. You build a workflow in which several specialized AI agents handle different parts of the process.

Example of a quoting process:

One agent reads a request for quote that arrived by email.
A second agent pulls prices and availability from the ERP.
A third checks the customer's contract terms in the CRM.
A fourth drafts the quote.
A fifth checks the risks.
The salesperson approves or edits the draft.

This is different from “I'll ask ChatGPT to help with a quote.” Here, AI is part of the engine room.

Where does the ROI come from? Shorter lead times, fewer manual handoffs, fewer errors, more standardized processes, people freed up to handle exceptions.

Why shouldn't you start with this? Because a multi-agent system multiplies both the benefits and the risks. If the process is poorly understood, the agents automate the mess.

According to Gartner, there's a lot of hype around agentic AI, and a large share of projects may fail due to unclear business value and costs.^[2]

A good sequence: document the process → identify recurring steps → identify decision points → define human approval limits → automate a limited part → measure → expand.

7. Knowledge management and putting your own data to work

In many companies, the biggest problem isn't a lack of information. It's finding it. Information lives on the intranet, in SharePoint, Slack, Teams, PDFs, project folders, the CRM, emails and people's heads.

This is where RAG, or Retrieval-Augmented Generation, makes a big difference. RAG means the AI retrieves information from the organization's own documents or databases and builds its answer on them. AI becomes the interface to the company's memory.

IBM's AI in Action report emphasizes that AI leaders build their roadmaps on four dimensions: strategy, tools, data management and applications. Data management in particular has to be built on accessibility and governance so that AI use cases can reliably deliver value.^[7]

Where does the ROI come from? Less time spent searching, faster onboarding, better decision-making, reuse of project lessons, better use of customer information, breaking down internal knowledge silos.

Imagine the question: “What did we learn from last year's three lost bids in Sweden?” If AI can retrieve the project notes, bid documents and sales comments, it can produce an answer that no single person might have time to put together.

The biggest risk: RAG isn't a magic fix. If the knowledge base is messy, outdated or contradictory, AI can give a well-phrased answer based on poor material. That's why the real work in a RAG project often lies in cleaning up documents, access rights and information ownership.

AI finds the information. But the organization first has to decide which information is reliable.

8. Coding, software architecture and testing

Software development is an area where AI is already part of everyday work. Developers use AI to write boilerplate code, generate tests, explain errors, understand legacy code, refactor, produce documentation, learn a new language or library, draft database queries and compare architecture options.

In Stack Overflow's 2025 developer survey, positive sentiment toward AI tools had declined from previous years, even though usage is widespread. That points to something important: AI is useful, but developers have also seen its limits.^[8]

In software development, AI's value isn't just about writing code faster. Often the bigger benefit is understanding. AI can explain a legacy system to a new developer or help find the spot where a change should be made.

Where does the ROI come from? Faster development, shorter onboarding, better test coverage, less documentation debt, faster bug fixes, legacy systems that are easier to understand.

The biggest risk: AI-generated code can look finished even when it isn't secure. According to news coverage of Veracode research, a significant share of AI-generated code contained security flaws.^[9]

The rule is clear: AI can write code. The developer is responsible for it.

Level 4 — Multipliers of understanding

9. Data analysis and BI

Business Intelligence, or BI, means using business data in decision-making. Traditionally, BI required reports, dashboards and analysts who translated the data into language management could use. AI is changing that.

Management can ask in plain language: “Why did our margin drop in Sweden last month?” AI can look for anomalies in the data, spot trends, compare segments and suggest follow-up questions.

This doesn't eliminate analysts. It raises the value of their work. Instead of routine reporting, analysts can focus on interpretation, modeling and assessing the impact of decisions.

Where does the ROI come from? Faster analysis, better questions, faster detection of anomalies, democratized data, higher-quality decision-making.

This is one of the most important “multipliers of understanding”. When data becomes easier to query, the whole organization learns to see cause and effect better.

The biggest risk: AI can find correlations that aren't causal relationships. It can also explain data in a way that's convincing but wrong. That's why, when AI is used for BI, it should show what data was used, what filters were applied, which findings are statistically significant, what is interpretation and what should be checked next.

Good AI analysis doesn't just answer. It teaches you to ask better questions.

10. Learning and skills development

AI is a tireless mentor. It can explain the same thing five different ways, build a learning path, ask practice questions, simulate a customer situation or act as a sparring partner.

AI can help with onboarding new employees, training salespeople, coaching managers, learning technical skills, improving language skills, simulating customer situations, understanding internal guidelines and continuous learning.

In its 2025 AI ROI reporting, Deloitte emphasizes the importance of AI fluency — ROI leaders see AI skills as a core competency, and 40 percent of them require AI training.^[6]

AI isn't just a tool you learn to use. It's also a way to learn other things faster.

Where does the ROI come from? Faster onboarding, better knowledge sharing, fewer training bottlenecks, more personalized learning paths, better customer and sales readiness, continuously updated skills.

The biggest risk: AI can teach the wrong things. That's why, when using AI for learning, you need to distinguish between general sparring, the organization's official information and topics that require an expert. AI can be a mentor. But the owner of the curriculum has to be a human.

The ROI formula: how to choose your first AI project

Here's a practical rule of thumb:

Potential savings × AI's realistic share of the work × probability of success − cost of rollout

Potential savings

How much time, money or how many errors are tied up in the process? For example: 1,000 invoices a month, 12 minutes of manual work per invoice, 200 hours a month, an hourly cost of €40 → a potential labor saving of €8,000 a month.

AI's realistic share of the work

What share can AI really handle? Not optimistically — realistically. If AI can handle 60 percent of the volume and pass the rest to a human, use the figure 0.6.

Probability of success

How likely is it that the project will succeed? That depends on the data, the process, the systems, the responsibilities and change management. It's better to start with a project that has a 90 percent probability of success, even if the benefit is smaller.

Cost of rollout

Include: software, integrations, training, security, data cleanup, project work, maintenance, quality control.

This formula keeps AI romanticism in check. It forces you to ask: what actually works?

Why do AI projects fail?

AI projects rarely fail because the model can't do anything. They fail because the organization doesn't know what it's trying to change.

1. The use case is too vague

“Let's use AI in sales” isn't a project. “Summarize every customer meeting into the CRM and surface follow-ups to the salesperson within 24 hours” is a project.

2. The process hasn't been mapped

If nobody knows how the work actually flows, it's hard to automate.

3. The data is bad

AI doesn't rescue bad data. It makes it look more credible.

4. The human role stays unclear

Who approves? Who checks? Who is responsible for mistakes?

5. The metrics are wrong

If you only measure the volume of content produced, you get a lot of content. If you measure resolved customer cases, time saved or sales growth, you get closer to real value.

6. Rollout stops at a training video

AI doesn't spread through an organization by sending people a link to a tool. You need examples, ground rules, support and leadership commitment.

According to McKinsey, high-performing AI organizations stand out in part because they have practices for human review of model outputs and build value across six dimensions: strategy, talent, operating model, technology, data, and adoption and scaling.^[1]

AI isn't an IT project. It's a redesign of work.

A 90-day roadmap for rolling out AI

Days 1–15: map your processes

List 20 recurring processes. Ask: what takes a lot of time, what repeats often, where do errors happen, where does information get lost, where does the customer wait, where do employees get frustrated?

Days 16–30: score the use cases

Score each use case on: savings potential, revenue potential, data availability, technical ease, risk level, probability of success, cost of rollout.

Choose one high-probability project. Not your biggest dream. Your best first win.

Days 31–45: build a pilot

Scope the pilot clearly: one team, one process, one product group, one customer service channel, one document type. Define the metrics before you start.

Days 46–60: test with humans in the loop

Don't let AI go straight into autonomous production. Test: errors, exceptions, user experience, escalation, security, documentation, cost per transaction.

Days 61–75: measure and improve

Compare before and after: handling time, error count, customer satisfaction, employee satisfaction, cost, lead time, quality control findings.

Days 76–90: decide on scaling

Three options: scale it, fix and retest it, or shut it down. Good AI leadership also means killing bad pilots quickly.

Conclusions: start with the boring stuff, then move to the strategic

AI's greatest value doesn't come from the flashiest demo. It comes from a process that runs every single day.

That's why an organization that thinks ROI first should start where the probability of success is high: invoices, tickets, contracts, CRM summaries, internal documents, reporting, code testing, onboarding.

Once those work, you can move on to more demanding areas: agent systems, dynamic pricing, predictive customer analytics, broader decision-support models, the organization's own AI memory.

The most important lesson is this:

Don't ask where AI sounds interesting. Ask where work repeats, information gets lost, customers wait or money leaks. That's where the ROI lives.

Summary: 10 use cases ranked by ROI

Finance and accounting automation — Fast payback, a clear process, measurable savings.
Managing customer service volumes — Big benefits at high message volumes, as long as escalation to a human works.
Procurement and contract management — Hidden potential: AI can save both time and purchasing spend.
Expanding existing customer accounts — AI finds the signals, humans build the trust.
Scaling marketing and communications — Speeds up production, but human insight determines quality.
Multi-agent systems and process automation — Big potential, but requires mature process management.
Knowledge management and putting your own data to work — The company's memory as a user interface.
Coding, software architecture and testing — Speeds up development, but responsibility for security stays with humans.
Data analysis and BI — Makes data queryable and decisions easier to understand.
Learning and skills development — Scales continuous learning across the whole organization.

One last rule of thumb

Choose your first AI project like this: Potential savings × AI's realistic share of the work × probability of success − cost of rollout.

Start with a routine task where the probability of success is 90 percent. Even a small benefit becomes big when it repeats every day.

AI's real ROI isn't in one big leap. It's in the power of compounding.