DN SME Agent Opportunity Finder v1.0
Prepared 22 October 2026 (Africa/Johannesburg). Batch 2, article 72.

SCOPE AND EVIDENCE
50 deliberately selected common business processes across six functions; not a representative survey. Ratings are authored editorial judgments for a hypothetical small business with existing cloud applications, authorised data, a stable process and a named reviewer. No company, product or autonomous agent was deployed or benchmarked for this dataset. There are zero measured deployments, zero measured payback periods and no measured accuracy or time-saving results. Blank payback values mean unmeasured, never zero months.

RUBRIC: each dimension is an integer from 1 to 5.
V (value potential): 1 incidental convenience; 2 occasional administrative help; 3 useful recurring support; 4 meaningful recurring operating workload; 5 potentially material workload or control benefit. Actual volume and realised value must be measured locally.
C (clarity of acceptance criteria): 1 subjective outcome with no simple check; 2 ambiguous criteria; 3 mixed objective and judgmental criteria; 4 mostly explicit checks; 5 clear schema, approved source or deterministic comparison.
E (integration effort, higher is harder): 1 one controlled input and draft output; 2 a maintained source plus an existing application; 3 coordinated records or multiple read integrations; 4 multiple changing systems and exceptions; 5 consequential cross-system authority or substantial custom work. These are relative judgments, not hours or implementation quotes.
R (error consequence, higher is more serious): 1 low-stakes reversible internal categorisation; 2 contained rework with review before effects; 3 potential customer, financial-record or policy impact; 4 significant financial, service or operational consequences; 5 severe or difficult-to-reverse money, rights, access, deletion or binding-commitment effects. The dimension rates the stated process, including its prescribed controls; it is not a measured probability of harm.

SCORING
Score = 35*(V-1)/4 + 25*(C-1)/4 + 20*(5-E)/4 + 20*(5-R)/4.
This maps the dimensions to 0–100. It is a weighted editorial priority score, not a percentage of readiness, accuracy, success or return. Weights are a declared design choice, not estimated coefficients. Small score gaps do not establish meaningful differences.

HARD GATES AND SORT ORDER
Protected processes concern sensitive approvals, money movement, binding commitments, destructive actions, consequential employment decisions, access rights or sensitive escalation. They are separated from initial pilot candidates regardless of score. A human owner and the business's existing formal controls are required. This framework does not approve autonomous execution or substitute for domain review.
Among non-protected processes, First pilot requires R<=2, E<=2 and C>=4; all other non-protected processes are Supervised. These thresholds are editorial choices. First pilot means a candidate for bounded evaluation, not permission to act without review.
Sort non-protected before protected, then score descending, value descending, error consequence ascending, integration effort ascending and stable process ID ascending. Rank is a unique display order; tied numerical scores remain tied in evidence, not statistically distinct.

TIME TO ROI
No payback ranking is supported. Measure baseline volume, handling time, review/rework, full running costs, realised cash value and setup cost. Simple payback = setup cost / positive monthly net realised benefit. If net benefit<=0, simple payback is not reached under those assumptions. Constant-benefit payback ignores ramp-up, time value of money and future volatility. Gross minutes released are capacity, not automatically cash savings. The existing DN AI Agent ROI Index provides a separate calculator.

VALIDATION
The code regenerates scores, sorting, JSON and CSV from the authored records. Arithmetic and structure checks do not validate the editorial ratings or real-world suitability. A separate assistant review is recorded separately. No named external expert review is claimed.

USAGE
Filter by business function and pilot class; retain the human control and validation criterion for every shortlisted process. Start with an existing deterministic rule or ordinary automation where sufficient. Use a bounded agent only where branching and tool selection add decision value. Keep internal systems and personal data out of the public feedback route.

CHANGE LOG
v1.0: initial 50-process editorial framework, 22 October 2026. Update when process controls or credible evidence changes; no automatic review cadence is promised.
