Practical AI Workflow Research
We test how AI actually performs inside real business workflows.
Less hype. More evidence.
Latest
Using the ten structured cases from Experiment #002, a human and an AI each assigned a priority, a next action, a reason and something to avoid. Both converged on the same operational pattern. AI was more conservative about missing information and stronger on operational prerequisites; the human was more commercially proactive and read customer psychology.
Ten synthetic packaging and printing inquiries were converted into a structured table by a human (about 30 minutes) and by an AI model. AI output was more systematic in extraction, completeness and flagging missing information. Human judgment remained stronger on commercial context. No time-saving figure was established.
We ran 10 synthetic customer inquiries through a qualification workflow — interpret the message, identify missing information, decide the next action — and compared how a human and an AI model handled them. The inquiries were incomplete more often than not, and the next action was rarely obvious.
What we test
FIELD turns real business workflows into experiments that reveal what AI can actually do. Each experiment has a question, a method, a human baseline, an AI baseline, results, failures and limitations — the same structure every time.
Observed → Inferred → Unknown
Find a real, recurring business workflow — not a demo scenario.
Investigate how the work is actually done and where it breaks.
Test AI against it, alongside a human baseline.
Document what worked, what failed, and what we still don't know.
Publish the evidence, then run better experiments.
Findings
AI Can Prepare the Next Action Without Being the Decision-Maker
In Experiment #003, AI consistently turned structured cases into a next-action framework — prerequisites, questions, what not to promise — while the human still supplied the commercial call.
Source: Experiment #003 · 12 Sept 2026
AI Caution Can Be Useful — Until It Slows the Conversation
In Experiment #003, the AI's conservatism protected against premature commitments, but in several cases the human chose to move the customer forward sooner.
Source: Experiment #003 · 12 Sept 2026
AI's Apparent Advantage May Be Standardization, Not Judgment
In Experiment #002, the AI's edge over the human was consistency and coverage — not a better read of the business situation.
Source: Experiment #002 · 29 Aug 2026
Missing Information Is Part of the Business Problem
In Experiment #001, most inquiries were incomplete, and the useful step was making the gaps visible before acting.
Source: Experiment #001 · 15 Aug 2026
Why FIELD
Evidence over hype
We don't tell you what AI should be able to do. We test what it actually does.
Real workflows
Customer inquiries, messy information, next-action decisions — the work businesses actually do.
Failure is documented
Every experiment has a failure-modes section. If AI wasn't useful, we say so.
Human judgment stays visible
Each test carries a human baseline so you can see where the two diverge.
Practical conclusions
Observed → Inferred → Unknown, every time. No universal laws from ten cases.
Research updates
Get new FIELD experiments and findings when we publish them.
