AI concepts
Workflow-first design (processes before choosing technology)
A principle for designing AI deployments under which you first design the target business process and only then select the technology. The reverse of the dominant practice, in which the choice of AI model precedes the redesign of the process. Companies that apply workflow-first achieve twice the ROI from AI.
Primary source: McKinsey Global AI Survey, November 2025, BCG The Widening AI Value Gap, September 2025
The term gained popularity in 2025 after the McKinsey and BCG reports, which independently identified process redesign as the single strongest predictor of financial success in GenAI deployments. Both firms recommended reversing the order.
The dominant failure pattern
The standard sequence in Polish and global enterprises looks like this. The board decides that the company will deploy AI. The CTO gets a budget. The CTO launches an RFP and selects a technology vendor. The vendor recommends a use case where its tool has the best chance of showing value. The deployment starts. After six months it turns out that the process the AI was plugged into has no metrics, no owner and no clear definition of success. The pilot ends without ROI.
McKinsey 2025 calls this sequence technology-first design and links it to an 80 percent failure rate at the EBIT level.
Workflow-first in practice
The reverse sequence starts with a business question. Which process costs the company the most, in measurable units? What exactly do we want to change, which KPI should move, and over what horizon? What does the target process look like after the change, independent of the tool? What technical requirements does that process place on the AI model?
Only after answering these four questions does the technology RFP begin.
The data
The BCG report The Widening AI Value Gap (September 2025), a sample of 1,250 companies: 60 percent of organisations report no material value from AI despite continued investment. In the subgroup of companies that deployed AI after redesigning the process, the rate of immaterial value drops to 28 percent.
McKinsey adds: companies with a measurable AI impact on EBIT are 2x more likely to redesign end-to-end business processes before choosing a model. This is a correlation, but a repeatable one across three consecutive editions of the Global AI Survey.
The practical consequence
Workflow-first design lengthens a project by four to twelve weeks (the process redesign stage) and lowers the cost of selecting technology by 30 to 50 percent (because better requirements let you pick a smaller, cheaper vendor). The net cost is usually lower, despite the longer start.
Workflow-first is the foundation of the yesfor.ai method. The sequence we work in: processes, data, teams, and only then AI.