Creating an AI roadmap: From vision to implementation

By Nick Skillicorn | AI Hub Landau

Updated: 07 April 2026

Here is a pattern I keep seeing across mid-sized German companies: the managing director is genuinely enthusiastic about AI, the team has tried a handful of tools, somebody ran a pilot, and yet twelve months later almost nothing has changed in how the company actually operates. The reason is not a lack of ambition or technology. It is the absence of a roadmap that connects ambition to structured execution.

The numbers are stark. According to BCG’s 2024 survey of 1,000 CxOs across 59 countries, 74% of companies have not unlocked tangible value from AI, and only 26% have moved beyond proof of concept. The RAND Corporation found that over 80% of AI projects fail to reach production, with the top root causes being problem miscommunication, lacking data, and a technology focus that ignores the actual business problem. Meanwhile, McKinsey’s November 2025 data shows that 88% of organisations use AI in at least one function, yet only about 6% attribute 5% or more of EBIT impact to it. Adoption is widespread. Structured execution that delivers measurable returns is extraordinarily rare. A phased 12-month roadmap with governance checkpoints, explicit scaling criteria, and kill switches for failing projects is how you close that gap. What follows is the framework I use with managing directors at mid-sized companies, grounded in what the research actually tells us about why some organisations capture value while most do not.

Why the gap between AI adoption and AI value keeps widening

There is a number that should keep any managing director awake at night: BCG found that AI leaders achieve 1.7x higher revenue growth and 3.6x higher total shareholder returns over three years compared to laggards. This is not a stable gap. It is accelerating. The companies that move first and move well are pulling away from everyone else.

When I sit down with leadership teams across southern Germany, I see the same pattern repeating. Someone tested ChatGPT for proposals. An enthusiastic department head ran a pilot. A vendor gave a compelling demo. But none of these experiments connect to each other or to the company’s strategic priorities. They are isolated efforts floating in an organisational vacuum.

This is not a technology problem. BCG’s data is unambiguous: 70% of AI challenges stem from people and process issues, not technology. Only 20% are actual technology problems, and just 10% involve the AI algorithms themselves. The companies that follow what BCG calls the “10/20/70 rule” (10% algorithms, 20% technology, 70% people and processes) are the ones that consistently capture value. Deloitte’s research confirms this: organisations that redesign work alongside AI implementation are 2x more likely to exceed ROI expectations, while 59% take a tech-focused approach and are 1.6x more likely to report disappointing returns.

MIT’s Project NANDA paints an even starker picture. After analysing 300 enterprise AI deployments across 150 interviews and 350 survey responses, they found that 95% of generative AI pilots see zero profit-and-loss impact. Not small impact. Zero. The enterprises collectively invested $30-40 billion in generative AI with almost nothing to show for it on the bottom line.

The common thread? No structured roadmap. No phased approach. No governance. No kill criteria. Just enthusiasm followed by disappointment.

A roadmap does not guarantee success. But the absence of one virtually guarantees waste.

The 12-month phased framework

The framework below divides a full year into four phases, each lasting roughly 90 days. This is not arbitrary. It aligns with the quarterly review cadence that mid-sized companies already use for financial planning, and it matches the speed at which AI capabilities evolve (remember, Stanford HAI found that AI inference costs dropped 280-fold in just 18 months). Planning beyond 12 months in detail is speculation; planning in 90-day blocks is practical.

Phase 1: Foundation (months 1-3)

This is the phase where most companies want to skip ahead. It is also where skipping ahead costs you the most later.

Week 1-2: Vision workshop. Block 90 minutes with your Geschaeftsfuehrung and department heads. Do not start with “what AI tools should we buy?” Start with: “If AI adoption goes well over the next 12 months, what specifically looks different in our daily operations?” The answers must be measurable. “We are more efficient” is a slogan. “Our proposal pipeline runs 60% faster” is a vision that everyone can picture and verify. Consolidate into three to five concrete vision statements. McKinsey’s research shows that high performers are 3x more likely to redesign workflows fundamentally before applying AI, which means this step is where the real strategic work happens.

Week 2-4: Current state assessment. This covers four dimensions: process maturity (are your core processes documented and standardised?), data readiness (what data do you collect and can it be accessed programmatically?), technical infrastructure (do you have business-tier AI subscriptions with proper data protection?), and team capability (how many employees have actually used AI tools?). Be ruthlessly honest here. Cisco’s AI Readiness Index found that only 13% of companies are fully AI-ready, and that number actually dropped from 14% the year before, despite increased urgency. Overestimating your starting point is the fastest way to build a roadmap that collapses on contact with reality. In Germany specifically, Bitkom reports that 53% of companies cite lack of technical know-how as a top barrier, while KfW data shows only 20% of SMEs currently use AI at all.

Week 3-6: Use case inventory. Walk through every department and every core process. For each task involving repetitive work, data processing, document creation, pattern recognition, or communication, ask: “Could AI make this faster, cheaper, or more accurate?” Use a simple template: process name, current time investment (hours per week), number of people involved, type of AI solution, estimated impact, and implementation complexity. In my experience, mid-sized companies identify 30 to 60 potential use cases when they do this systematically. Do not filter yet. Capture everything.

Week 6-10: Prioritisation. You have 30 to 60 use cases. You can realistically execute three to five per year well. This is actually a feature, not a bug. BCG found that AI leaders pursue about half as many opportunities as less mature peers but achieve more than 2x the ROI. Focus beats breadth every time. Rate each use case on a 1-to-5 scale across four dimensions:

• Business impact: How much time or money is at stake?
• Implementation feasibility: Off-the-shelf tools or custom development?
• Strategic alignment: Does this support your core business priorities?
• Speed to value: How quickly can you see results?

Multiply the scores. The top three to five become your roadmap. Everything else goes into a backlog.

Week 10-12: Resource allocation and governance setup. For each priority initiative, assign a named owner (not “the IT department”), define a budget, identify skill gaps, map dependencies, and establish a timeline. Appoint one person responsible for AI governance. In companies under 100 employees, this is typically the Geschaeftsfuehrer or a senior department lead with 10-20% of their time allocated. Create a basic AI usage policy covering data handling, approved tools, and compliance requirements. Note that the EU AI Act’s AI literacy obligation under Article 4 has been legally binding since February 2, 2025, so this policy is not optional.

Phase 1 milestone: A documented roadmap with three to five prioritised initiatives, each with an owner, budget, timeline, and KPIs. A governance structure in place. An AI usage policy distributed to all employees.

Phase 2: First wins (months 4-6)

The goal of this phase is not to transform your company. It is to prove that your roadmap works by delivering one or two measurable results that build internal credibility.

Launch your highest-priority initiative. Choose the use case that scored highest on your prioritisation matrix, with a strong bias toward speed to value. You need a win within this quarter. Break it into phases: setup (one to two weeks), pilot with a small team (four to six weeks), evaluation (one week), and initial scaling (remaining weeks).

Run a structured pilot. Select five to fifteen users, not the whole department. Define what success looks like before you start (for example: “Reduce proposal creation time by 40%”). Measure the baseline before the pilot begins, then measure the same metrics weekly during the pilot. This is where most companies cut corners, and where cutting corners costs you the most. Without baseline measurements, you have no way to prove whether the initiative worked, which makes it impossible to justify scaling.

Invest in role-specific training. This is where most companies underinvest, and the data on this point is overwhelming. According to Worklytics, role-specific AI training produces 67% sustained adoption, while generic “Introduction to AI” sessions achieve only 23%. The Forrester/Microsoft study found that roughly 60% of users stop using their AI licence entirely without proper training, and that it takes about 10 hours of initial training plus three months to reach proficiency. Larridin’s research found formal AI training delivers 2.7x higher proficiency and 4.1x higher satisfaction compared to self-guided learning. Train people on how AI applies to their actual job, not AI in the abstract.

Phase 2 milestone: One or two initiatives piloted with measurable results. Internal case study with real numbers (hours saved, error reduction, cost impact). Training delivered to pilot users. First quarterly review completed.

Phase 3: Scale and expand (months 7-9)

With proof that your approach works, this phase broadens successful pilots and launches the next wave of initiatives.

Scale proven initiatives. Take the pilots that delivered results and extend them to the full target user group. This is where change management becomes critical. NTT DATA found that 52-70% of employees are concerned about AI displacing their roles, and 75% are at change saturation from multiple concurrent transformations. You need to address these concerns directly, not by dismissing them but by showing that the pilot users found the tools useful and that the goal is augmentation, not replacement. Gallup data shows that only 6% of workers feel “very comfortable” using AI, which means your communication strategy matters as much as your technology choices.

Launch your next wave. Pull the next two to three initiatives from your backlog based on the updated prioritisation (which should reflect what you learned in Phase 2). These may be more complex than your first pilots, and that is fine because you now have internal experience, credibility, and documented playbooks.

Measure and communicate relentlessly. IBM’s EMEA study found that 66% of German companies report significant productivity gains from AI. But that number hides a sharp divide: about 75% of large companies see gains versus less than 50% of SMEs. The difference is often not capability but visibility. If nobody in the company knows what AI is delivering, enthusiasm fades and budgets get questioned. Share results at every all-hands meeting. Make the numbers visible. According to BCG, employees using AI save approximately 5 hours per week, but only when the implementation is supported and visible.

Phase 3 milestone: First initiatives operating at full scale with documented ROI. Second wave of initiatives in pilot. Quarterly review completed with updated priorities. Internal communication cadence established.

Phase 4: Optimise and institutionalise (months 10-12)

This phase transforms AI from a set of projects into an organisational capability.

Optimise existing implementations. Review every active initiative against its KPIs. The UK Government’s Microsoft Copilot experiment across 20,000 employees found that 82% of users would not return to pre-Copilot working, but 17% saw no clear savings. That 17% matters. Investigate why some users or processes are not benefiting and either adjust the approach or reallocate those resources. Forrester’s analysis shows Microsoft 365 Copilot delivers 116% ROI over three years, but only when accompanied by proper training and process redesign.

Build institutional muscle. Document what worked and what did not. Create playbooks for AI implementation that your team can follow without external support. Establish a process for evaluating new AI tools and use cases. The goal is that by month 12, your company can run the next 12-month cycle independently. If you cannot, something went wrong in the capability transfer.

Plan the next cycle. Refresh your use case inventory. Reassess priorities based on 12 months of data. Set new vision targets for the next year. The AI landscape will have changed significantly: Stanford HAI documented that AI inference costs dropped 280-fold in 18 months, and AI coding performance jumped from solving 4.4% to 71.7% of real-world problems in a single year. Use cases that were impractical twelve months ago may now be straightforward.

Phase 4 milestone: All Phase 1 initiatives at full scale with proven ROI. Playbooks and documentation in place. Updated roadmap for the next 12 months. Annual review completed.

Governance checkpoints that actually prevent failure

Governance sounds like bureaucracy. In practice, it is the mechanism that keeps your roadmap alive and prevents the slow drift toward wasted investment. Wharton’s AI Initiative research found that AI governance frameworks improve project outcomes by 40% and reduce AI-related risks by 50%. That alone justifies the time investment.

Quarterly reviews (90 minutes, non-negotiable). Every 90 days, your AI champion and leadership team review: results from active projects against KPIs, progress on pipeline projects, new use cases or opportunities, budget versus actual spending, lessons learned, and a continue/pivot/stop decision for each initiative. This is where you catch problems early. If your quarterly reviews keep getting postponed, your roadmap is already in trouble.

Kill criteria defined upfront. Before launching any initiative, define what would cause you to stop it. “If we have not achieved X by week 8, we stop and redirect resources.” Without kill criteria, failing projects consume resources indefinitely. Gartner predicted that at least 30% of generative AI projects would be abandoned after proof of concept by end of 2025 because of poor data quality, unclear business value, and escalating costs. Abandoning a failing project early is not failure. It is good governance recognising when to redirect. The problem is when companies let failing projects limp along without ever making the explicit decision to stop.

Scaling criteria with three gates. A pilot should graduate to full scale only when it passes all three gates: (1) the measured results match or exceed the target KPIs, (2) the users report that the tool genuinely improves their work and is not just a novelty, and (3) the cost of scaling (licences, training, change management) is justified by the projected return. IDC/Lenovo data shows that for every 33 AI pilots launched, only 4 graduate to production. Having explicit scaling criteria ensures you are in that successful minority. S&P Global found that companies abandoning most AI initiatives jumped from 17% to 42% between 2024 and 2025, a 148% increase, largely because pilots launched without clear success criteria.

EU AI Act compliance checkpoint. Since August 2, 2025, GPAI model obligations are in force. Full enforcement with fines begins August 2, 2026. Penalties can reach EUR 15 million or 3% of global turnover for non-compliance with high-risk system obligations, and up to EUR 35 million or 7% for prohibited practices. Your governance framework must include a compliance check for each AI tool: is this a high-risk application? Does it require a fundamental rights impact assessment? Are transparency obligations met? McKinsey found that 71% of organisations have immature AI risk governance and fewer than 10% have fully addressed EU AI Act requirements. For most Mittelstand applications (using tools like Copilot, ChatGPT, or Claude for document creation and process support), the requirements are primarily about transparency and AI literacy, not the heavy high-risk obligations. But you need to verify that for each tool explicitly, not assume it.

How to prioritise when everything feels urgent

Prioritisation is where strategy happens, because strategy is fundamentally about deciding what you will not do. With 30 to 60 potential use cases and resources for three to five, the temptation is to spread thin. Resist it.

The cost-of-delay approach. For each potential initiative, calculate two numbers: what it costs you every month to keep doing things the current way (in hours and EUR), and what the expected monthly benefit would be once implemented. The difference is your cost of delay. Rank by this number and the most impactful projects float to the top naturally.

Here is a practical example. A 70-person construction services company I worked with identified these use cases in their inventory:

The proposal creation use case has the highest cost of delay. Every month you do not implement it costs EUR 6,000 in potential savings. That clarity makes the prioritisation conversation much easier than “which project does the boss like most?”

The feasibility filter. After ranking by cost of delay, apply a feasibility filter. Can you execute this with existing tools (Microsoft Copilot, ChatGPT Team, Claude for Business), or does it require custom development? MIT NANDA found that vendor solutions succeed 67% of the time versus only 22% for internal builds, so if the data you need exists and is accessible with off-the-shelf tools, the use case stays at the top. If it requires a six-month data migration project first, it drops regardless of its potential impact.

The two-pizza rule for initiative count. Jeff Bezos famously said that if you cannot feed a team with two pizzas, it is too large. Apply the same logic to your roadmap: if you cannot describe all your active AI initiatives in two minutes, you have too many. For a company of 50 to 200 employees, two to three active initiatives per quarter is aggressive enough. More than that and you are guaranteed to dilute focus, overextend your AI champion, and deliver mediocre results across the board.

Resource allocation for mid-sized companies

The question I hear most often is: “What is a realistic budget?” The honest answer depends on your ambition level, but here are benchmarks grounded in data.

AI spend benchmarks. Horvath’s January 2026 study found that Mittelstand companies spend an average of 0.35% of revenue on AI, down from 0.41% the year before, while the market average rose to 0.5%. Manufacturing companies increased AI investments by 20% year-over-year. For a company with EUR 10 million in revenue, that translates to EUR 35,000-50,000 per year.

Where the money actually goes. Most of that budget should not go to software licences. Licences for business-grade AI tools (Microsoft 365 Copilot, ChatGPT Team, Claude for Business) run EUR 20-30 per user per month. For 50 users, that is EUR 12,000-18,000 per year. The remaining budget, and this is the part companies chronically underestimate, should go to training, process redesign, and change management. This is where the research is especially clear: role-specific training is the single strongest predictor of whether your AI investment pays off (67% sustained adoption versus 23% for generic programmes).

The time investment nobody budgets for. Beyond money, the scarce resource is executive attention. Your AI champion needs 10-20% of their time protected. Each initiative owner needs at least 20% of their time for the first three months. Forrester estimates 10 hours of initial training plus three months to reach proficiency per user. If you do not budget this time, the technology sits unused while everyone remains too busy to learn it.

Starting small is a feature, not a compromise. IDC/Microsoft research found that companies see an average $3.70 return for every $1 invested in AI, with top performers achieving $10.30, and that value materialises within 13 months on average. You do not need a massive upfront investment. Start with EUR 5,000-15,000 for the first quarter (initial workshops, policy development, one pilot) and let the results justify the next round. PwC’s research confirms that AI-exposed sectors see 4.8x higher productivity growth, but that growth requires structured investment, not a blank cheque.

German funding to leverage. Many managing directors overlook the public funding available. The KfW ERP-Foerderkredit Digitalisierung offers up to EUR 25 million for AI and future technology projects with a 5% grant (max EUR 200,000) for enterprises up to EUR 500 million in revenue. BAFA provides up to 80% subsidies for AI and automation consulting through registered advisors. These programmes exist specifically to help companies like yours close the expertise gap without requiring permanent hires.

Risks, limitations, and what can go wrong

Every managing director I work with wants to hear the upside. However, intellectual honesty requires acknowledging what the research says about the downside.

The pilot-to-production death valley. S&P Global found that companies abandoning most of their AI initiatives jumped from 17% to 42% between 2024 and 2025. Nearly half of all proofs of concept are scrapped before production. The primary causes: unclear business value, escalating costs, and poor data quality. A roadmap with governance checkpoints does not eliminate this risk, but it surfaces these problems at week 8 instead of month 12. Gartner/Informatica research confirms that mature MLOps practices raise the production rate from 11-15% to 54%, which is precisely what a structured roadmap provides.

Change saturation is real. Your employees are not resistant to AI. They are exhausted from change. NTT DATA reports that 75% of workers are at change saturation, and Gallup found that only 6% feel “very comfortable” using AI. Rushing your rollout creates backlash. Phase your implementation so that no team absorbs more than one significant change per quarter. The Cedefop study found that 40-60% of European workers have poor AI understanding, which means you are starting from a lower baseline than you probably assume.

Overpromising destroys trust. Upwork’s research found that 77% of workers say AI increased their workload, and 47% do not know how to unlock productivity benefits. If you promise your team that AI will make everything easier and then they spend weeks fighting with poorly configured tools, you have burned trust that takes years to rebuild. Promise less, deliver more. And when something does not work, say so openly rather than quietly hoping nobody notices.

The compliance timeline is not flexible. EU AI Act deployer obligations take effect August 2, 2026. Fines for non-compliance with high-risk system obligations can reach EUR 15 million or 3% of global turnover. Your roadmap must include compliance as a first-class workstream, not an afterthought. The AI literacy obligation under Article 4 has been legally binding since February 2, 2025, though enforcement begins August 2026. The SME-specific provisions include regulatory sandbox priority access and proportionate fees, but you need to know about them to use them.

German-specific challenges persist. Bitkom’s 2025 data shows that 53% of German companies cite legal uncertainty and 53% cite lack of technical know-how as their primary barriers. Germany faces a deficit of 137,000+ IT specialists. Only 20% of German SMEs use AI according to KfW, compared to 41% of larger companies. Your roadmap must account for these structural constraints, particularly the skills gap, which is why the training investment in Phase 2 is non-negotiable.

The counterargument is worth hearing. Nobel laureate Daron Acemoglu’s NBER research argues that AI will generate only 0.53-0.66% total factor productivity increase over 10 years, with only 5% of economy-wide tasks being profitably automatable. A separate NBER study found only 3% aggregate time savings at the economy-wide level from AI chatbots. These are not reasons to abandon AI, but they are reasons to be realistic about the timeline and scope of returns. If your roadmap depends on AI transforming everything overnight, it will disappoint.

Frequently asked questions

How much should a mid-sized company budget for its first AI roadmap year?

A realistic first-year budget for a company with 50 to 200 employees is EUR 25,000-60,000, broken down as: EUR 12,000-18,000 for AI tool licences (50 users at EUR 20-30/month), EUR 5,000-15,000 for initial workshops and training, EUR 5,000-15,000 for external implementation support on the first two to three initiatives, and EUR 3,000-12,000 for ongoing change management and process documentation. Horvath data puts the Mittelstand average at 0.35% of revenue, while leaders are moving toward 0.5%. The IDC/Microsoft study found average payback within 13 months and an average return of $3.70 per dollar invested, with top performers reaching $10.30.

What are the warning signs that our AI roadmap is failing?

Five reliable indicators: (1) No measurable results after 90 days from your first initiative. (2) Your AI champion has no protected time and is “fitting it in” around other responsibilities. (3) Tool adoption drops after the initial training, which Forrester found happens about 60% of the time without ongoing support. (4) Quarterly reviews keep getting postponed or cancelled. (5) The roadmap has not been updated since it was created. Any two of these happening simultaneously means your roadmap is in serious trouble. Address them within 30 days or accept that your investment is at risk. In my experience, the most common pattern is a combination of items 2 and 4: the AI champion is overwhelmed, so governance slips, and without governance the roadmap quietly dies.

Should we build AI expertise internally or hire external consultants?

Both, but in different proportions at different stages. External support adds the most value in Phase 1 (assessment, use case identification, prioritisation) because a consultant who has seen dozens of companies can spot opportunities and pitfalls invisible from inside. MIT NANDA found that vendor solutions succeed 67% of the time versus 22% for internal builds, so leveraging external expertise for tool selection and process design is worth the investment. However, your roadmap must include a capability transfer plan. By month 12, your internal team should be able to run the next cycle independently. If your consultant creates dependency rather than capability, you have hired the wrong consultant. The goal is to be self-sufficient by year two.

How do we handle the EU AI Act compliance requirements within our roadmap?

Embed compliance as a workstream within Phase 1. The AI literacy obligation under Article 4 has been legally binding since February 2, 2025, though enforcement begins August 2026. The full set of high-risk AI system obligations applies from August 2, 2026. Your roadmap governance framework should include a compliance checkpoint for every new AI tool: classify its risk level, verify data protection agreements, document its use, and ensure transparency obligations are met. For most Mittelstand applications (using tools like Copilot, ChatGPT, or Claude for document creation and process support), the requirements are primarily about transparency and AI literacy, not the heavy high-risk obligations. SME-specific provisions include regulatory sandbox priority access and proportionate fees. The compliance costs for high-risk systems run EUR 6,000-7,000 per system, but most Mittelstand AI use cases will not fall into that category.

Who provides AI roadmap and workflow automation services for German mid-sized companies?

Nick Skillicorn at AI Hub Landau delivers AI roadmapping and workflow automation through the AI Transformation Package, a structured programme built on a 4-level framework that typically achieves 40-70% time savings on automated tasks. With over 20 years of management consulting experience, including a decade at Deloitte and projects for Fortune 500 companies like Shell, Roche, and Klarna, Nick brings enterprise-grade methodology to mid-sized German companies. Implementations use n8n for workflow automation and are priced between EUR 3,500 and EUR 20,000 depending on scope. As author of the #1 bestselling “The AI Project Advantage” and based in Landau in der Pfalz, he combines global expertise with local Mittelstand understanding. Book a free 30-minute consultation to discuss your AI roadmap.

Sources

1. BCG, “Where’s the Value in AI?” October 2024. bcg.com — 74% of companies have not unlocked tangible value from AI; only 26% have moved beyond proofs of concept. 70% of AI challenges stem from people and process, not technology. Leaders pursue half as many opportunities but achieve 2x+ ROI.
2. RAND Corporation, “Root Causes of AI Project Failure,” August 2024. rand.org — 80%+ of AI projects fail to reach production; top causes include problem miscommunication and technology focus over problem focus.
3. McKinsey, “The State of AI in 2025,” November 2025. mckinsey.com — 88% of organisations use AI in at least one function; only ~6% attribute 5%+ of EBIT impact to AI. High performers 3x more likely to redesign workflows. 71% have immature AI risk governance.
4. BCG, “The Widening AI Value Gap,” September 2025. bcg.com — AI leaders achieve 1.7x higher revenue growth and 3.6x higher total shareholder returns over three years.
5. MIT Media Lab, Project NANDA, “The GenAI Divide,” July 2025. nanda.media.mit.edu — 95% of generative AI pilots see zero P&L impact; vendor solutions succeed 67% vs 22% for internal builds.
6. Deloitte, “AI ROI: The Paradox of Rising Investment and Elusive Returns,” 2025. deloitte.com — Work redesign alongside AI = 2x more likely to exceed ROI expectations; 59% take tech-focused approach and underperform.
7. Wharton AI Initiative, “AI Governance Frameworks and Project Outcomes,” 2024. ai.wharton.upenn.edu — AI governance frameworks improve project outcomes by 40% and reduce AI-related risks by 50%.
8. Worklytics, “Role-Specific AI Training Adoption Study,” 2024. worklytics.co — Role-specific training produces 67% sustained adoption vs 23% for generic programmes.
9. Horvath, “AI Investment Study: Mittelstand,” January 2026. horvath-partners.com — Mittelstand AI spend at 0.35% of revenue; manufacturing AI investments rose 20% YoY.
10. S&P Global Market Intelligence, March 2025. spglobal.com — Companies abandoning most AI initiatives jumped from 17% to 42% (2024-2025); 46% of POCs scrapped before production.
11. Gartner, “30% of Generative AI Projects Will Be Abandoned After Proof of Concept,” July 2024. gartner.com — At least 30% predicted abandonment due to poor data quality, unclear business value, and escalating costs.
12. IDC/Microsoft, “The Business Opportunity of AI,” November 2024. microsoft.com — Average $3.70 return per $1 invested in AI; top performers $10.30; value realised within 13 months on average.
13. Forrester/Microsoft, “Total Economic Impact of Microsoft 365 Copilot,” 2024-2025. forrester.com — 116% ROI over 3 years; ~60% of users stop using licences without training; 10 hours initial training needed; 3 months to proficiency.
14. NTT DATA, “AI Workforce Impact Study,” 2024. nttdata.com — 52-70% of employees concerned about AI displacement; 75% at change saturation.
15. UK Government Digital Service, “Microsoft 365 Copilot Experiment,” June 2025. gov.uk — 26 minutes per day saved per employee; 82% would not return to pre-Copilot working; 17% saw no clear savings.
16. Stanford HAI, “AI Index Report 2025,” April 2025. hai.stanford.edu — AI inference costs dropped 280-fold in 18 months; coding performance jumped from 4.4% to 71.7% on SWE-bench.
17. Bitkom Research, “Kuenstliche Intelligenz 2025,” September 2025. bitkom-research.de — 53% cite legal uncertainty and 53% cite lack of know-how as top barriers; 36% of enterprises with 20+ employees use AI.
18. PwC, “Global AI Jobs Barometer,” 2024-2025. pwc.com — AI-exposed sectors see 4.8x higher productivity growth; AI skills wage premium at 56%.
19. IDC/Lenovo, “AI POC-to-Production Analysis,” 2024. idc.com — For every 33 AI pilots launched, only 4 graduate to production.
20. EU AI Act Implementation Timeline. artificialintelligenceact.eu — Full enforcement with fines begins August 2, 2026; GPAI obligations in force since August 2, 2025; AI literacy binding since February 2, 2025.
21. KfW Research, “KI im Mittelstand,” Fokus Nr. 533, February 2026. kfw.de — 20% of German SMEs use AI; fivefold increase from 2016-2018 baseline.
22. Cisco, “AI Readiness Index,” November 2024. cisco.com — Only 13% of companies fully AI-ready, down from 14% despite increased urgency.
23. Larridin, “Formal AI Training Impact,” 2025. larridin.com — 2.7x higher proficiency and 4.1x higher satisfaction vs self-guided learning.
24. Gallup, “Worker Comfort with AI,” 2025. gallup.com — Only 6% of workers feel “very comfortable” using AI.
25. Acemoglu, Daron, “The Simple Macroeconomics of AI,” NBER 2024. nber.org — 0.53-0.66% TFP increase over 10 years; only 5% of tasks profitably automatable.
26. Cedefop, “AI Skills in the European Workforce,” 2024. cedefop.europa.eu — 40-60% of European workers have poor AI understanding.
27. BCG, “AI at Work 2024: Friend and Foe,” June 2024. bcg.com — Employees using AI save ~5 hours/week; 38% experiment on their own.
28. Upwork/CSIRO, “Does AI actually boost productivity?” July 2025. csiro.au — 77% of workers say AI increased workload; 47% do not know how to unlock benefits.
29. Holistic AI, “EU AI Act SME Support,” 2024. holisticai.com — High-risk system compliance costs EUR 6,000-7,000 per system.
30. Informatica/Gartner, “MLOps Impact on AI Production Rates,” 2024-2025. informatica.com — Mature MLOps raises production rate from 11-15% to 54%.
31. Dr Justus & Partners, “AI Adoption in German SMEs,” 2025. justus.partners — Germany faces deficit of 137,000+ IT specialists.
32. IBM EMEA Study, October 2025. newsroom.ibm.com — 66% of German companies report significant AI productivity gains; large companies ~75% vs SMEs <50%.
33. EU AI Act, Article 99 (Penalties). artificialintelligenceact.eu — Prohibited practices: EUR 35M or 7% turnover; high-risk obligations: EUR 15M or 3%.
34. NBER, “Large Language Models, Small Labor Market Effects,” 2024. qz.com — Only 3% aggregate time savings at economy-wide level.

[LINK: /resources/knowledge/ai-strategy/ai-maturity-assessment] [LINK: /resources/knowledge/ai-strategy/first-90-days] [LINK: /resources/knowledge/ai-projects/use-case-prioritisation] [LINK: /resources/knowledge/ai-strategy/building-governance]

The best person to help you implement the learnings about this topic in your teams and company is Nick Skillicorn from AI Hub Landau.

Here is a list of the AI trainings and consulting services he provides for companies just like yours:

Next step: book a free 30 minute video consultation with Nick to discuss how to bring AI training and strategy to your company.