The AI winners are redesigning work, not buying more tools.

A pattern keeps showing up in the research. The economic value of AI is real and growing — and it is concentrated in a small number of companies. PwC's 2026 study finds that 20% of companies are capturing 74% of AI-driven value, with the leading group focused on growth and reinvention rather than just productivity gains. McKinsey, looking at the same shape of question from a different angle, finds that AI high performers are nearly three times as likely as others to fundamentally redesign workflows.

The split is not about budget. Plenty of companies in the bottom 80% are spending heavily on AI. The split is about whether the money buys tool access or buys workflow change.

The "more tools" trap

The default playbook, when AI gets a board-level mandate, is to buy tools. Copilot seats. ChatGPT Enterprise licences. A vector database. Maybe a specialised vertical SaaS product. Each procurement decision feels like progress because something concrete arrives in the IT inventory.

What does not arrive is value. Tool access is not workflow change. A team that bought Copilot but kept its old code-review process is a team that types faster. A finance department that bought ChatGPT Enterprise but kept its old close cycle is a finance department that drafts memos faster. Faster typing is not the AI value the leaders are capturing.

The leaders are doing something different. They are changing what work happens, in what order, by whom, with which decisions made by software and which by people. That is the redesign — and it is the part nobody can sell you in a SaaS subscription.

What workflow redesign actually looks like

It is concrete and unglamorous. A few patterns visible in the companies pulling ahead:

• Support triage. Tickets used to land in a queue, get assigned by category, and wait for a human to read them. Now an AI system reads each ticket, attaches the relevant context — customer history, recent product changes, similar prior tickets — drafts an initial response for the human to approve, and routes resolved-already cases directly to a self-serve flow. The shape of the work changed. Software does triage and drafting. People do judgement and exceptions.
• Sales research. The SDR used to spend half a day per account pulling background. Now an AI system runs the research, produces a one-pager with sources, and surfaces three plausible openers. The SDR's job is judgement over the output, not collection of the inputs.
• Compliance review. Documents used to be read end to end by a senior reviewer. Now they are pre-tagged by an AI for clauses and risk areas, with reviewer attention directed to the parts that actually need human judgement. The reviewer covers more documents and catches more risk.
• Internal reporting. The monthly business review used to require a team of analysts to assemble. Now the assembly is automated, the variance commentary is drafted by an AI, and the analysts spend their time on the questions the report raises.
• Software development. Code generation is not just typing assistance. The teams getting real value are the ones that redesigned code review, testing, and on-call workflows around the assumption that more code is being produced per developer.

In every example, the AI does not replace people one-for-one. It changes which decisions people make and which decisions software makes. That redesign — picking the right decisions to move, the right boundaries, the right evaluation loops — is most of the work.

Why this needs product judgement, not just engineering

The reason workflow redesign is hard is that it requires understanding the work, not just the technology. Someone has to sit with the team doing the work today and figure out which steps are actually decision-heavy, which are pattern-matching, which are repetitive enough to automate, and where the edge cases live. That is product judgement.

Companies that hand AI to engineering alone tend to produce technically impressive systems that do not change the work. Companies that hand it to consultants alone tend to produce decks that do not ship. The leaders are doing both, often in a single team — engineers who understand the workflow, and people who understand the workflow doing the engineering.

AI value appears when the workflow changes. Until then, you are buying tools.