# Policy comparison: a better business-support system

These are **three design hypotheses**, not proven Portland investments. Illustrative costs below are transparent planning envelopes in 2026 dollars, **not actual city appropriations, bids, grants or benefit-cost estimates**. Staffing assumes loaded annual cost of $145,000 per position; other lines are explicit assumptions to be refined with city procurement and provider rates. Costs cannot be summed into existing Prosper spending. The programs are complements but should be compared with a simpler process and with doing nothing new. The goal is sustainable owner income, good jobs, useful services and public value, not maximum firm count.

| Concept and first-year envelope | Who benefits and who acts | What changes | Evaluation and key risks |
|---|---|---|---|
| **1. Improve existing delivery — ~$1.13m**: 4 case managers × $145k = $580k; supervisor $145k; referral/CRM and data $180k; language access $120k; independent evaluation $100k. | Existing and prospective city firms, especially businesses sent among bureaus; Prosper OSB + City permitting, Revenue and procurement. | One case ID, clear eligibility, named owner, response-time service standard, warm handoffs, closure reason and multilingual status. Simplify rules where possible. | Staggered rollout or randomized queue priority when ethical; compare resolution time, owner hours, repeat contacts and downstream outcomes. Risk: counting referrals as resolutions, data sharing barriers, higher intake without capacity. |
| **2. Shared human/AI services — ~$1.89m**: 5 experienced advisors × $145k = $725k; 3 specialists for bookkeeping/procurement/legal-triage × $145k = $435k; secure tooling/integration $300k; language/translation $150k; privacy/security & quality $130k; evaluation $150k. | Small teams, sole proprietors and eligible prospective owners; Prosper, contracted providers, libraries/SBDCs and city agencies. | Human-reviewed cash-flow templates, document preparation, procurement discovery, plain-language permit guidance, marketing and application drafts. A multilingual advisor checks outputs; users can choose a human-only path. | Randomize offer among eligible users or stepped rollout; measure task completion, error rate, owner time, cost, repeat usage, sales/profit/job quality at 6–12 months. Risk: wrong legal/tax advice, privacy, digital divide, uneven language accuracy, replacement of local service providers, and more applications chasing fixed grants. |
| **3. Remove structural barriers — ~$2.07m pilot operating envelope plus any separate capital**: 5 process/permit staff × $145k = $725k; process/IT redesign $550k; small-premises and demand analysis $300k; faster-payment implementation $180k; evaluation and controls $315k. | Applicants to open or expand; vendors awaiting public payment; City permitting, procurement, finance, zoning and partners. | Set permit cohort clocks, pre-application checklists, coordinated reviews, transparent reasons for delays, prompt payment and premises inventory. Any loan fund, guarantee or property investment requires a separate appropriation and accounting. | Compare eligible permit cohorts before/after and matched other jurisdictions, with seasonal and case-complexity controls; measure time-to-open, cancellations, capital tied up, rent and customer outcomes. Risk: shifting waits between bureaus, landlord capture, or expensive subsidies without new demand. |

These envelopes are comparable **first-year operating hypotheses** and include evaluation. They are not a recommendation to spend $5.09m immediately; stage gates should release funds only after a baseline, procurement review, privacy assessment and measurable pilot design. The first concept is the lowest-cost institutional starting point and may reduce need for the second. If owners' main problem is demand, rent or space, chatbot services alone will have limited effect. If regulations are irreducibly complex, navigation may still help but should not conceal avoidable complexity.

## Service design and measurement

Start with a **business journey**, not an agency directory: “Can I open?”, “What will it cost?”, “Where can I operate?”, “What must I file?”, “Who can buy from me?”, and “What happened to my request?” A user should receive a time-stamped answer with cited official rules, eligibility and a human escalation route. Do not turn AI output into a legal, tax or lending decision. Record versioned source citations, confidence and the eventual resolved case. Owners control use of their financial records; separate service delivery data from program evaluation and release only safe aggregates. Procure open exports to avoid dependence on a single vendor.

The service funnel must begin with an **eligible population denominator**, then awareness, inquiry, application, accepted case, completed task, resolved problem, owner time saved and later business/worker outcomes. Report each stage by business model, industry, city geography and owner group only where privacy and sample size permit. Report failure and abandonment, not just testimonials. Distinguish existing clients from new reach and unique firms from interactions; resolve the OSB district-count discrepancy first. Costs should be per **resolved problem** and per additional outcome, not per website visit or chat response. Surveys need response rates, comparison with nonrespondents and a qualitative note on who opted out.

For AI, test against both human services **and a simpler application or process**. Grant-writing assistance may improve prose or reduce owner time but cannot create funding; more polished applications could raise competition and advantage already-resourced applicants. Include simple-form and lottery/transparent-score alternatives. Track impacts on bookkeepers, translators, advisors and marketing workers, and whether software reduces labor hours or complements expertise. [Singapore's SME digital support](https://www.imda.gov.sg/resources/press-releases-factsheets-and-speeches/press-releases/2025/smes-go-digital-day) and [Seattle's consulting services](https://bottomline.seattle.gov/2026/01/15/helping-seattles-small-businesses-grow/) show possible service components, not proof of Portland effectiveness. The [NYC MyCity audit](https://comptroller.nyc.gov/reports/audit-report-on-the-new-york-city-office-of-technology-and-innovations-mycity-system/) documents why public AI guidance needs accuracy testing and an accountable human owner.

## Decision standard

Portland could credibly aspire to be exceptionally good at **starting and sustaining** diverse businesses if it can make time-to-open, first-year survival, median owner income after expenses, job quality, customer access and service resolution measurable and improve them relative to comparable places. A “#1 in the world” label is premature without a compatible international denominator or consensus definition. International cases are design references; do not convert incomparable national statistics into a rank. A high share of microfirms is not itself a trophy. A world-leading system lets firms grow beyond small-size thresholds and tests whether gains accrue to owners, workers and residents.