The first AI workflow should feel slightly obvious by the time leadership funds it.
Not obvious because it is trendy. Obvious because the evidence has narrowed the field: the work happens often, the baseline is painful, the owner is credible, the context can be assembled, the risk can be controlled, and the outcome is worth measuring.
An AI opportunity assessment is the discipline that creates that clarity. It is not a survey asking every department for ideas. It is a short investigation into which workflow deserves the next serious build slot.
The assessment should end with a decision
The output is not a ranked list of dreams. It is a memo that says build this workflow, narrow it first, fix the foundation, or stop.
Start with workflow evidence, not executive enthusiasm
Most opportunity assessments start too wide and stay too vague. They collect dozens of ideas: AI for sales, AI for finance, AI for onboarding, AI for reporting, AI for support. Each idea has a champion. Each idea can be made to sound valuable.
The assessment has to pull those ideas down into recent operating evidence.
For each candidate, inspect three to five real examples of the work. Reconstruct what happened:
- what triggered the work
- who owned it
- what records, documents, messages, and policies mattered
- where the delay or rework appeared
- which exceptions required judgment
- what system changed at the end
- which metric would prove improvement
The best candidate often emerges during that reconstruction. It is the workflow where people can show the pain, not just describe it.
What the research should change about the assessment
McKinsey’s 2025 State of AI survey shows why adoption is the wrong screening criterion. Regular AI use is already broad at 88% of organizations, but only 39% report EBIT impact and about two-thirds are not scaling enterprise-wide. The high performers are disproportionately the companies redesigning workflows, assigning senior ownership, defining human validation points, and tracking KPIs. An opportunity assessment should therefore ask which workflow can change operating behavior, not which department has the most AI interest.
NIST’s AI Risk Management Framework makes opportunity selection a risk decision too. Its govern-map-measure-manage structure is a useful forcing function before funding: what risk does the workflow introduce, how will it be measured, and who manages it after launch? If the candidate touches sensitive data, customer commitments, employee decisions, or system-of-record updates, the assessment should include risk mapping before the scope is funded.
Metacto’s Opportunity Mapping is the practical version of this assessment: a 2-3 week review that produces a ranked opportunity map, systems review, context and risk assessment, value case, target workflow, and first-build recommendation. Metacto’s Operational AI model shows why the assessment cannot stop at value: the workflow also needs context, agents, controls, and continuous operations.
The opportunity memo
Use the memo below as the assessment artifact. It is intentionally short enough for a COO, CFO, CTO, or functional leader to read before a funding meeting.
AI opportunity memo
A good memo makes the next decision easier. It should not require leaders to infer the workflow, risk, baseline, or owner from a pile of workshop notes.
Memo section: Workflow promise
- What it must answer
- What named workflow changes, for whom, and what operating result should improve?
- Weak answer warning
- The idea is described as a tool, department, or broad capability instead of a workflow.
Memo section: Evidence sample
- What it must answer
- What did three to five recent cases reveal about trigger, context, handoffs, exceptions, and write-back?
- Weak answer warning
- The assessment relies on interviews and opinions without looking at real work.
Memo section: Baseline
- What it must answer
- What are current volume, cycle time, review burden, error rate, rework, cost, or revenue impact?
- Weak answer warning
- The business case starts with assumed productivity gains rather than measured friction.
Memo section: Context readiness
- What it must answer
- Which systems and records are required, which are authoritative, and where are conflicts likely?
- Weak answer warning
- The workflow requires tribal knowledge or manual lookup that has not been designed into the system.
Memo section: Control model
- What it must answer
- What can AI read, draft, recommend, update, escalate, or never touch?
- Weak answer warning
- The opportunity sounds valuable only if the agent is given unsafe authority.
Memo section: Owner and cadence
- What it must answer
- Who owns adoption, review quality, technical reliability, and metric review after launch?
- Weak answer warning
- The sponsor wants AI value but no operating owner agrees to run the changed workflow.
Memo section: Decision
- What it must answer
- Should leadership build, narrow, fix the foundation, or stop?
- Weak answer warning
- The memo asks for discovery budget without naming the decision the discovery should unlock.
The five filters for the first workflow
The best first workflow is not necessarily the highest-value idea in the company. It is the best value-learning-risk combination for the next build.
Use five filters.
1. Repeated volume. The workflow happens often enough to justify a system and to create useful feedback. Rare executive work may be valuable, but it is usually a poor first automation target because the feedback loop is slow.
2. Measurable pain. The baseline includes delay, rework, manual lookup, review burden, error risk, missed revenue, margin leakage, or support load. If the pain cannot be measured at all, the project will struggle to prove improvement.
3. Available context. The workflow depends on records and rules the system can access with appropriate permissions. If the context lives only in a few people’s heads, the first investment may need to be knowledge capture rather than automation.
4. Controllable action. The workflow can start with AI preparing, drafting, classifying, or recommending while a human approves sensitive actions. If the only valuable version requires full autonomy on day one, the first release is too risky.
5. Real ownership. A named process owner is willing to change the workflow, review the metrics, handle exceptions, and decide whether to expand. Without that owner, the assessment should not recommend a build.
Baseline before the business case
The baseline is not a finance appendix. It is how the team learns whether the workflow is real.
Baseline before launch
These measures keep the ROI conversation honest. They separate hard savings, recovered capacity, quality movement, and confidence before the project asks for more budget.
Current cost
Volume multiplied by prep, review, rework, and follow-up time.
Quality delta
Correction, rejection, override, defect, or rework movement.
Capacity recovered
Time returned to the team without assuming headcount disappears.
Confidence level
How much evidence leadership has before expanding the build.
The baseline should include both operating metrics and business metrics. If the candidate is invoice exception triage, measure exception count, review minutes, rework, error rate, late approvals, and vendor friction. If the candidate is renewal prep, measure preparation time, risk coverage, follow-up quality, renewal outcomes, and CSM review load.
The point is not to build a perfect ROI model in week one. The point is to identify which metric the first release can credibly move.
A worked assessment example
Assume a mid-market B2B services company is comparing three candidates:
- renewal prep for customer success
- invoice exception triage for finance
- executive meeting briefs for operations
Renewal prep has strong leadership interest. CSMs spend hours gathering CRM notes, tickets, usage data, support escalations, and open commitments. The metric is preparation time and renewal-risk coverage. Context is available but spread across systems. Review is clear: CSM approves before any customer action.
Invoice exception triage has a stronger cost baseline. Finance can show volume, exception percentage, analyst effort, and delay. The risk is higher because the workflow touches AP and vendor decisions. The first release can stay safe if AI prepares evidence and recommendations without approving payment.
Executive meeting briefs are popular but fuzzier. The team wants summaries, but the trigger is broad, source ownership is unclear, and the success metric is subjective. It may be useful later, but it is not the best first workflow unless leadership can name the decision quality or prep-time baseline.
The assessment might recommend invoice triage first if the CFO owns the metric and source systems are accessible. It might recommend renewal prep first if customer success has stronger ownership and the company wants a lower-risk first workflow. It should likely park executive meeting briefs until the operating question becomes sharper.
The decision language
Every assessment should end in one of four decisions.
Build when value, owner, context, control, and measurement are all strong enough for a narrow production release.
Narrow when the opportunity is real but the first version crosses too many systems, teams, actions, or risks.
Fix the foundation when the workflow depends on missing data, unclear source ownership, weak identity rules, or nonexistent review paths.
Stop when the idea is mostly novelty, politics, or generic productivity with no accountable workflow.
This decision language protects the team from the worst outcome: every idea survives as a “maybe.” A good opportunity assessment removes bad ideas from the active queue.
The assessment standard
Metacto treats opportunity assessment as the first production decision, not a strategy exercise. The assessment should leave the company with a first workflow worth funding and a clear reason why the other candidates are later, narrower, or not worth building.
That clarity is valuable even before the build starts. It gives the COO a workflow owner, the CFO a baseline, the CTO a context and control question, and the functional leader a concrete operating promise.
When the first workflow is selected this way, the build starts with less theater and more truth.