The numbers from Q1 2026 are extraordinary by any measure. Global AI funding reached $300 billion in a single quarter — a 150 percent increase year on year. OpenAI raised $122 billion, Anthropic $30 billion, xAI $20 billion. More capital flowed into AI in three months than into all of global venture funding for the whole of 2024.
Against this backdrop, a survey of 1,200 employees and C-suite executives by AI platform company WRITER tells a rather different story. 79 percent of organisations face significant challenges in AI adoption — a double-digit increase from the same survey twelve months ago. Only 29 percent report meaningful ROI from generative AI. For AI agents specifically, the figure is lower still: just 23 percent.
These two data points — record investment and widespread implementation failure — are not in tension. They describe the same phenomenon from different angles: an industry in a hype cycle, and the gap between what AI can do and what most organisations have managed to make it do for them.
The more useful question is not why the majority are struggling, but what the 29 percent are doing differently.
What the Stanford Data Tells Us
Stanford's Digital Economy Lab recently published an Enterprise AI Playbook analysing 51 enterprise AI deployments across industries. The headline finding is striking: organisations using the same AI technology, from the same vendors, at comparable scale, are achieving vastly different outcomes.
The variance is not explained by model choice, budget, or sector. It is explained by execution — specifically, by three factors that consistently differentiate high-performing deployments from failing ones:
- Use case specificity. High-performing organisations identified a narrow, well-defined problem before selecting a technology. Failing deployments tended to begin with the technology — "we need to deploy an AI strategy" — and work backwards to use cases, often finding none that were compelling enough to drive adoption.
- Change management investment. Successful AI deployments treated the human side of adoption as the primary challenge, not the technical side. They invested in training, in workflow redesign, and in addressing the cultural resistance that emerges when AI changes how people do their jobs. Failing deployments deployed the technology and expected adoption to follow.
- Measurement discipline. Organisations that achieved ROI defined measurable success criteria before deployment and tracked against them rigorously. Those that did not tended to describe their AI deployments in qualitative terms ("it's helping with efficiency") and could not demonstrate value when it came time for renewal decisions.
Where ROI Is Actually Happening
The aggregate data obscures significant variation by use case and sector. When you look at where the 29 percent are generating returns, the pattern is clear.
Use cases with strongest ROI
- Software development. AI coding assistants and code review tools are consistently the highest-ROI AI investment across industries. The feedback loop is tight, the success criteria are measurable, and the productivity gains are immediate and visible. Organisations that have embedded AI coding tools into their development workflow report 20–40 percent reductions in time-to-delivery for comparable features.
- Customer support. AI-assisted and AI-automated customer service is generating strong returns where it has been deployed against well-defined query types. The key is specificity: an AI trained on a specific product's support documentation outperforms a general-purpose assistant significantly. Generic chatbot deployments continue to underperform.
- Search and knowledge retrieval. Internal knowledge management — finding the right policy document, the relevant precedent, the applicable clause — is a use case where AI consistently delivers value with relatively low deployment complexity. The productivity gain from reducing time spent searching for information compounds significantly across large organisations.
Sectors with strongest adoption
Technology companies lead by a significant margin, followed by legal services and healthcare. The common thread is not sector characteristics but workforce composition: these sectors have high concentrations of knowledge workers doing cognitively intensive tasks where AI augmentation produces measurable productivity gains.
Sectors lagging in ROI — retail, manufacturing, construction — tend to have a higher proportion of physically intensive work where current AI cannot intervene in the primary task, and where the administrative overhead reduction (the typical AI play in these sectors) is a smaller proportion of total cost.
Why Most Deployments Are Failing
The WRITER data points to five failure modes that account for the majority of the 79 percent experiencing challenges.
1. The "strategy-first" trap
Many organisations built an AI strategy before identifying the specific problems they needed AI to solve. The result is a strategy document that identifies AI as a priority without a clear answer to the question: priority for what? Deployment follows, but without a compelling use case it never reaches the threshold of adoption where network effects kick in and value compounds.
2. Treating AI as a tool, not a workflow change
AI does not improve productivity by being dropped into an existing workflow as a faster way to do the same steps. The productivity gains come from redesigning the workflow around what AI can do — which often means eliminating steps entirely, changing who does what, and redefining quality standards. Organisations that deployed AI without changing the surrounding workflow consistently report lower adoption and lower ROI than those that redesigned first.
3. Underestimating the data problem
AI performs in proportion to the quality and structure of the data it works with. Many organisations that reported early AI success with generic use cases (drafting, summarisation) hit a ceiling when they attempted domain-specific deployments, because their internal data — the documentation, the records, the institutional knowledge — was fragmented, inconsistent, or inaccessible. The AI investment exposed a data infrastructure debt that needed to be paid before the AI could deliver its full value.
4. No measurement framework
If you cannot measure whether your AI deployment is working, you cannot improve it, you cannot justify renewing it, and you cannot build the internal case for expanding it. A striking number of organisations in the WRITER survey reported that they had no defined success metrics for their AI programmes. This is the most correctable of the five failure modes — and the most consequential.
5. Skipping change management
The technology is rarely the bottleneck. The bottleneck is adoption — getting the people whose workflows AI is supposed to improve to actually use it, trust it, and integrate it into how they work. This requires active management: explaining why, addressing fears about job displacement honestly and directly, providing training that goes beyond how to use the tool to why it will make their working life better. Organisations that invested in this consistently outperformed those that did not.
Five Questions to Ask Before Your Next AI Investment
The Stanford and WRITER data, taken together, suggest a straightforward diagnostic framework. Before committing to any AI deployment — whether expanding an existing programme or starting a new one — ask these five questions.
- What specific problem are we solving, and how will we know we have solved it? If you cannot answer this with a measurable outcome, the deployment is not ready.
- What does the workflow look like after AI? The redesigned workflow should look different from the current one — not just the same steps done faster.
- What is the quality of the data this AI will work with? Audit the data layer before deploying the AI layer. Surprises here are expensive.
- Who owns adoption, and what is their plan? Technical deployment and adoption are different projects. Both need owners.
- What does success look like at 3 months, 6 months, and 12 months? Define these milestones before you start, and build a review cadence that holds the programme accountable to them.
The record AI investment of 2026 will produce winners and losers — not based on who spent the most, but on who answered these questions well before they started. The gap between organisations that are generating real returns from AI and those that are not is closing slowly. The way to end up on the right side of it is not to invest more, but to invest more deliberately.
Reinvently works with organisations to design AI programmes that are built around clear use cases, measurable outcomes, and realistic adoption plans. If you want to move from AI experimentation to AI ROI, our AI Strategy service is the right starting point.
← All posts