Nicolaj Leonhard-Christensen

Most companies roll out AI like any other software, and that’s why it rarely pays off.

Hi, I’m Nicolaj. This is where I think out loud about getting AI to pay off at work, and where you can see how to work with me.

Work with me Read my reflections

Why is it so hard to get value from AI?

AI tools often do work that was never anyone’s job, so nobody picks them up.

They tend to be picked to save money, like fewer agency hours or fewer freelancers. But a tool doesn’t make the work anyone’s job.

People say they don’t trust what the AI writes, but they rarely have an example.

In a rollout I was part of, six of the seven people who got the tools first never really started, and that was their reason. When the doubt comes without an example, the tool is rarely what’s broken.

Nobody has written down what good work looks like, so the AI can’t push back.

A weak draft and a strong one look the same to the tool, so the quality ends up depending on who is holding it.

Give the same AI tool to a beginner and an expert, and the expert wins every time.

AI lifts everyone a little, and the gap in your team stays where it was.

So I start with the work around the tool, and the AI goes in last.

  1. I start where your business makes money.

    Fifteen years in commercial work have taught me where AI is worth the effort.

  2. I write down how your best people do the work.

    Then the AI is held to it, whoever is using it.

  3. I stay until your team uses it.

    Someone owns it, a manager asks for it, and I work next to the team until it runs without me.

I have spent fifteen years getting the work right first, and this is some of it.

  • ABM programmes and a playbook for Hempel

    The win rate went from 38% to 59% in ten months, and the programme won silver at the Danish Digital Awards 2025.

    I led it as BirdsAtFive’s consultant. The case

    What the video shows
    1. Card: “In the Gulf, the spec decides the coating.”
    2. Card: “Engineering firms write that spec, long before sales gets involved.”
    3. Card: “Few of those firms knew Hempel. So Hempel rarely made it into the spec.”
    4. Card: “The fix: market to the 33 firms that write the specs.”
    5. The accounts: “We used research and AI to pick 33 key firms.” A count runs up to “33 key accounts”.
    6. The idea: “One idea for all of them: Specified to Last.” The campaign film plays.
    7. The content: “Papers on the questions engineers ask.” Three white papers. Then: “Case stories from projects in the region.” Two case stories.
    8. The reach: “Ads and emails aimed only at those 33 firms.” Four LinkedIn ads.
    9. The team: “Sales and marketing met every second week.” A sales lane and a marketing lane meet: “Every second week, for ten months”.
    10. The result: “Ten months later:” 33 of 33 target accounts engaged, project win rate from 38% to 59%, pipeline value up 43%, value of won projects up 19%.
    11. “And it won an award.” Silver at the Danish Digital Awards 2025, Best Digital Demand Generation (B2B).
    12. End card: “Specified to Last.” An ABM programme for Hempel by BirdsAtFive.

  • Birdie and the Hempel AI Hub

    Case stories written in-house at Hempel went up 300%.

    I owned the product at BirdsAtFive; developers wrote the code. The case

    What the video shows
    1. Card: “AI writes about your product in seconds.” An AI draft types itself: “Clearhull is the only silicone coating with 5 years between dockings.”
    2. Card: “And gets it wrong in seconds.” The claim is struck out.
    3. Card: “The right facts are in your reports and decks. AI never sees them.”
    4. Card: “So we built Birdie. Your facts go in once. Every AI tool reads them.”
    5. “Let’s take a look.” The Birdie app opens.
    6. “Your reports and decks go in.”
    7. “Birdie turns them into facts, each with its source.”
    8. “When two sources disagree, it shows you both.” A sales deck from March 2025 says “The only silicone coating with 5 years between dockings”. A market report from October 2026 says “No longer unique: Norvald launched SilkHull in September with the same 5 years”.
    9. “One click settles it.”
    10. “Now ask it to write.” A question is typed into the chat.
    11. “It writes with your facts. Each one has its source.”
    12. “And it leaves out the claim that’s no longer true.”
    13. End card: Birdie. “Your facts go in once. Every AI tool reads them.” Built at BirdsAtFive.

  • Topsoe’s first ABM pilot

    Five steps to test ABM before scaling it.

    I led it as BirdsAtFive’s consultant.

    What the video shows
    1. Card: “Topsoe wanted to try ABM for the first time.”
    2. Card: “ABM changes how marketing and sales work, so you test it before you scale it.”
    3. Card: “Here is how we ran their first ABM pilot.”
    4. The plan: “Five steps, from kick-off to scaling.”
    5. Step 1, anchoring: “Marketing and sales agreed what the pilot had to prove.”
    6. Step 2, capabilities: “We checked what their tech and teams could already do.”
    7. Step 3, accounts and team: “Sales and marketing chose the accounts together.” Then: “And the sales team learned how ABM works.”
    8. Step 4, develop and run: “Then one idea for the accounts: NOW.” The NOW ad: “When Marius goes ice-fishing, he counts on you. Seize the European winter diesel opportunity NOW.” Then: “Ads, a landing page and direct mail made for them.” Then: “A status meeting every second week kept it moving.”
    9. Step 5, scale: “It has just started. Step 5 decides what to scale.”
    10. End card: “Their first ABM, run as a pilot.” An ABM pilot for Topsoe by BirdsAtFive.

  • A sales CRM that asks for proof

    Its AI coach checks whether a warm deal really is warm.

    At BirdsAtFive. A colleague built the base; I designed how it sells.

    What the video shows
    1. Card: “Most CRMs are full of opinions.”
    2. Three labels appear, Warm, Committed and Going well: “All typed by us.”
    3. Card: “A CRM full of opinions is worth nothing.” The labels turn grey.
    4. Card: “So we built one that wants proof.”
    5. “Let’s take a look at a supposedly ‘warm’ deal.” The CRM opens (its screens are in Danish), and the coach is asked to challenge the deal.
    6. “The coach: ‘This one isn’t warm.’”
    7. “It counts the customer’s silence: 214 days.”
    8. “Then it preps the call.” The agreed next step: call the customer and ask whether the project still has a budget.
    9. “And who to call next, and why.”
    10. “Each signal comes with its source.”
    11. End card: “A sales CRM that argues back.” Built at BirdsAtFive.

  • Six years of e-commerce

    At Toys“R”Us and BR, Apopro and Vårdväskan. At Apopro, monthly revenue went from 5 to 20 million DKK.

The software in the videos shows made-up names and figures.

When you bring me in, it usually goes in four steps.

  1. I learn how you make money.

    I talk to the people who sell, the people who make and whoever decides, and look at where AI is used today and where it stops.

  2. We take one piece of work apart.

    Together we write down how it’s done well, step by step.

  3. AI goes into that work.

    With your own knowledge and your picture of good built in.

  4. I stay until it runs without me.

    That can be a few weeks, or longer as your part-time AI lead.

There are no fixed packages, and a talk or a workshop is a fine place to start.

Message me on LinkedIn

Reflections

Now and then I write a longer piece about what I see when AI meets real work.

  1. Write down what good work looks like before you build the AI toolComing next
  2. When people say they don’t trust the AI, ask them for an exampleLater
  3. Making AI mandatory gets people to log in, and the work stays the sameLater

Say hi

If you have a piece of work you’d like AI to take part in, or AI tools nobody uses yet, I’d like to hear about it. LinkedIn is the best place to reach me.

Message me on LinkedIn