Business problem · AI direction
The first AI decision is what not to build.
A useful AI strategy is a portfolio of business decisions, not a list of tools. We gather real workflow problems, compare their impact and user value, test technical and data feasibility, surface risk, and select the smallest pilot that can disprove a critical assumption. Some ideas should accelerate; others need research, process work or a clear “not now.”
Diagnose this problem01 · Signals
Signals that the organization needs discovery before development
- Each team has a different AI idea but no shared way to compare them.
- Vendors and models dominate discussions while users and outcomes stay vague.
- A prototype looks impressive but has no owner, workflow or adoption path.
- Security and legal concerns appear late and stop otherwise promising work.
- Leadership wants movement, but nobody can define what a successful first investment proves.
02 · Diagnosis
How we turn ideas into a decision
Every candidate is evaluated through the same evidence-based lens so enthusiasm does not replace prioritization.
What measurable business outcome would change if this problem improved?
Who experiences the pain and would adopt a different workflow?
What capability is uncertain, and can it be tested cheaply?
Is representative, permitted data available for evaluation?
What failure could harm a customer, employee, decision or obligation?
03 · Evidence
A prioritization score that supports a decision
- 01
Business impact
- 02
User desirability
- 03
Technical feasibility
- 04
Data readiness
- 05
Risk and reversibility
04 · Direction
Four honest outcomes from discovery
Discovery is valuable when it can recommend waiting or stopping.
Accelerate
The pain, owner, data and feasible pilot are clear enough to build and measure now.
Research
Value appears real, but a technical, data or policy assumption needs a short test first.
Prepare
The opportunity depends on process, ownership or data work that should happen before an AI pilot.
Stop
The expected value does not justify the complexity, risk or operating burden.
Smallest credible test
Prove the hardest assumption—not the easiest demo.
A focused pilot should create a decision: proceed, change direction or stop with evidence.
- 1Define the current baseline and target outcome
- 2Name the riskiest assumption
- 3Select representative users and data
- 4Build only what tests that assumption
- 5Make a documented go, change or stop decision
FAQ
Should we begin by selecting an AI model?
No. Begin with a business decision or workflow problem. Model selection follows the quality, privacy, latency, scale and cost requirements.
How many use cases should we pilot at once?
Usually one or a very small number. Concentrating on a clear outcome and difficult assumption produces better evidence than many shallow demos.
What should an AI roadmap contain?
Prioritized problems, owners, outcomes, dependencies, data and risk work, pilot decisions and the operating capabilities required to sustain successful systems.