Supermarket aisle with packaged food and a chilled dairy section, illustrating AI that pays off in food and FMCG business

AI that pays off in the business

Many companies have bought AI, but it does not pay off yet. I connect the technology with the business, so the effect of AI shows in value and not in savings. Below are two cases from each end of idea-to-market in food: marketing and NPD at Danish Crown Beef, and R&D at Arla Foods Ingredients. For each of them I describe how far the company has come with AI and what it has given the business.

Roadmap for upskilling AI competence & experience

Why AI Adoption matters

In consumer goods, about half of leaders call idea-to-market their most strategic process, yet only 11% have scaled AI there (BCG and The Consumer Goods Forum, 2026).

Below follows my model for generating return on AI in the companies where I assisted with AI-adoption and cyber security.

AI Development Stages

  • 1. Licences bought and tested

    AI is available and tried out. It saves individual employees time but rarely moves the business.

  • 2. AI in daily use

    Most employees use AI in their everyday work. Adoption limited to individual work flows and pilots.

  • 3. Front-runners and advanced use cases with business effect

    AI front-runners use agents and specialist analysis tools, and apply AI in the development projects themselves.

  • 4. AI changes what and how fast you launch

    AI development initiated on this step in business development, marketing and NPD, where time to market and time to earnings are reduced significantly.

AI in marketing and NPD at Danish Crown Beef

From 2025 I sat on the steering group of the Group AI adoption & transformation programme. We identified AI tools and business cases for sales, marketing and NPD. When I left in 2026, my department had taken them from licence to daily work in marketing and NPD.

10 AI use cases

  • Trend analysis in Claude Projects

  • Market analysis in M365 Copilot Researcher/Analyst

  • Synthetic consumer research & testing in NPD by Ipsos PersonaBot and NIQ BASES

  • Conjoint and correlation analysis in Copilot-assisted Excel

  • Adobe AI for segmentation, audience profiling and CX

  • AI eye tracking heatmaps in creative development

  • Presentations with Copilot-assisted PowerPoint

  • AI image generation in Midjourney

  • Video creation and editing in CapCut

  • Podcasts in NotebookLM

Effect on the business

Consumer research and ad testing normally take 2-3 months. Synthetic consumer research and creative tests therefore cut a significant amount of time in launch projects.

Time to market was reduced by months. The quality of new product concepts and campaigns went up. We documented that new initiatives started earning money sooner. Thereby increasing (AI) ROI in both marketing and NPD.

Team reviewing printed layouts and design materials on a table, illustrating creative development and concept testing in marketing and NPD

Arla Foods Ingredients

It’s not the tool that makes AI pay off. It’s the users.

In many companies, AI has been bought and tested, but it does not deliver to its full potential. The tools save individual employees time, but rarely create growth or speed up time to market (time to earnings).

AI in R&D at Arla Foods Ingredients is long past the pilot stage. Most employees have an extended AI licence and use AI in their everyday work, and a group of front-runners use agents and specialist analysis tools as expert users. The next step is to bring AI into the development projects themselves, connected to data from production, lab and pilot plant and chosen by business impact. The value goes far beyond development time. Together with management, IT and a reference group of managers, dairy technicians, developers and scientists, I mapped how R&D uses AI and what it gives the business. The mapping was done, in October and it will be followed up by November. Focus is more awareness around classification rules, and how hanging fruits to secure cyber security.

What it gives the business

Dairy technicians, developers and scientists assessed the value themselves. Because they are the ones doing the work.

  • Development time gets shorter. On a development project, the gain means faster time to market.

  • AI makes the work better and raises the level of innovation. Communication gets easier, and users get new ideas by asking better and more targeted questions

From the workshops with users, the future potential was quite tangible, namely a significant reduction in development time, even in very complex projects. The workshops also defined specific KPIs to measure AI ROI.

A business case that only counts hours often underestimates AI in R&D.

Where the next
value is

None of the colleagues who rarely use AI say they lack a licence. They point to know-how for their own role, and training aimed at their role. For me that was the clearest proof that the value sits with the users and not the tools themselves. An advanced user elevates the work many times more than a beginner, and the ability to give AI relevant input decides the level of innovation achieved. So the next gain lies in lifting beginners and intermediate users, not in more tools. Adoption is solid, and can be lifted further to go to the top of the scale.

Security is the precondition for scale

In R&D the data that matters most is often the most sensitive. That is also where AI can lift the work the most. When people are unsure what the tools do with data, the safe choice is to keep AI away from the core work, and then the value stays small.

I therefore measured value and risk in the same study, so management can see both on one page and decide where to open up, where to train and where to set limits. The awareness programme is measured before and after, so the effect can be documented.

Method

A questionnaire to all of R&D, mapping use by tool, task and data type. Value and risk were measured in the same study and brought together in a scorecard for AI adoption, behaviour and knowledge, and IT and technical controls. Risks were scored on likelihood and impact. Built on CFCS’ approach to behavioural change, OWASP Top 10 for LLM, SANS maturity model, ISO 27001/42001, NIS2 and EU AI Act article 4.

I started with thorough interviews in R&D and used the survey to confirm or refute what I assumed. Innovation was rated highest in the scorecard, followed by risk and cost. I ran the survey and the analysis in Forms. The questionnaire was developed with the help of Claude Projects with reference to benchmarks for best practice and regulation in AI use and cyber security. I handed the company the principles for responsible use and validation that I worked by.

Dairy technicians in hairnets working in a dairy pilot plant, illustrating AI in R&D for dairy ingredients

A catalogue of AI measures

  • Raise adoption & ROI

    • A list of AI potential. Ranked by business impact.

    • QBC. Action oriented input and proven questioning techniques that raise the quality of AI prompting and scope. Lifting innovation level.

    • Use case examples per data type

    • Marketing & innovation bots. Scientific workbenches for R&D and AI marketing workflows

    • Truly empower users to develop AI-skills

    • Actively seek disruptive business ideas

  • Make it safe

    • New license upskilling. It teaches what the provider actually does with your data.

    • A shared practice for verification and validation

    • An incident & cyber security card

    • Picking low-hanging fruit. Small tricks such as turning off training, feedback and deleting your own data in the tool to prevent storage.

  • Anchor it

    • A trained permanent local AI facilitator to help users navigate and improve business value

    • Managers as empowering ambassadors. Team members approach their nearest manager first when something goes wrong or to get answers.

My role as a AI-navigator
I deliver a catalogue of measures tailored to each company and function to increase adoption, earnings and ROI, and to harvest low-hanging fruit and strengthen cyber security. And I deliver adoption as a playing coach and project manager. The measures increase use and make it safe at the same time. None of them requires costly new technology, but they are paid back and can be implemented in a relatively short time.

Has your company bought and tested AI, but it does not deliver?

Give me 20 minutes of your time. I am looking for a company to work with this winter, where I put AI to work so it shows in cutting time to market and faster routes to earnings.

You will find my contact information below.

The competence behind the work

AI pays off when someone understands both the business and the risk. Through a one-year programme I am adding applied AI, cyber security and AI governance to 30 years of commercial and innovation leadership.

The UP-skilling programme

  • Applied AI, 2026: Adobe Experience League, HubSpot AI for Marketers and Make Academy, with 6 months of applied cases in market analysis and marketing workflows

  • Cyber Security Fast Track, 2026: full-time, covering NIS2, ISO 27001, risk management, cyber awareness and crisis communication. Happy42 and Aarhus University Katrinebjerg

  • MIT AI Executive Academy, Cambridge MA, January 2027 (admitted & in progress)

  • IAPP AI Governance Professional (AIGP), January 2027