Identify the business value of generative AI solutions

AB-731 · 116 questions

• A harbor-master office hears three vendor pitches: one labels incoming forms as berth request, complaint, or inspection; one forecasts next month’s slip occupancy from ten years of counts; one drafts plain-language welcome notes from the lease text. How should a leader label those pitches?
• A water-treatment plant already uses a fitted model that predicts weekend chlorine demand from temperature and flow. A consultant says they should “switch everything to generative AI” for that same gallon figure. What should the leader decide?
• A weekend farmers-market stall wants weekly produce blurbs that did not exist yesterday. A volunteer says their flyer that pastes the same three sentences is already “generative.” What should the leader recognize?
• A school-board archive compares a tool that tags meeting audio as budget, personnel, or facilities with a tool that writes a public summary from the transcript. How should the leader name those sides?
• A county-fair board wants a midway poster (image plus caption) from a one-line theme. A staffer says this is “computer vision, so it is not generative AI.” What should the leader conclude?
• A community swim-meet office says “we need AI” and assumes any chatbot is a custom-trained attendance predictor. What contrast should the leader draw?
• A library compares a script that always emails the same overdue-fee denial with a model that writes a tailored reply. Counsel asks why two runs of the model will not match word for word. What should the leader explain?
• A food pantry wants to forecast Saturday bag shortfalls and also draft donor-appeal emails. How should a leader treat those two needs?
• A transit-authority finance committee wants next-month fare-evasion rates to the nearest basis point. A vendor pitches a conversational model “because it understands transit.” What should the leader do?
• A parks department must turn seventy trail-condition briefs into visitor one-pagers every thaw season with two writers. Is a generative drafting solution a fit?
• A building-permit counter still walks every resident through the same paper packet. One vendor promises a writer that cuts the follow-up note from twelve minutes to three. Another would replace the walk-up packet with a scheduled digital intake and new clerk roles. How should the leader label those two offers?
• An elections office has ten days before early voting and must rewrite two hundred polling-place FAQs in ordinary language. One vendor offers a ready generative writer staff can use this week. Another offers a six-month commissioned custom model. What should the leader choose?
• A housing authority will use a generative writer for first-draft hardship-appeal packets that go to tenants. A vendor offers fully autonomous send with no clerk review. What selection constraint should the leader apply?
• After a storm, public works needs a count of clogged curb grates from street photos. A vendor pitches a chat writer “because it can talk about drains.” What should the leader do?
• A city clerk’s desk must answer residents with the official wording of adopted ordinances, not inventive restatements. A vendor pitches an open creative writer that “makes every reply original.” What should the leader prefer?
• A sanitation bureau wants a generator to summarize route-exception logs. Two contractors’ logs contradict each other, and no office owns a single source of truth. What should the leader do?
• A historic-preservation board will publish a downtown-grant pitch that newspapers will quote. Staff want a public pitch engine immediately. Fabricated dates or award claims would be costly. What should the leader do?
• A youth recreation center asks vendors to “bring AI” and then pick a brand. Staff have not said whether they need volunteer-shift summaries or a registration forecast. What should the leader require first?
• A senior-center office writes ordinary weekly activity blurbs. General language quality is already good enough for that newsletter. A vendor says they must specialize a model before any draft. What should the leader do?
• A planning department’s value is a house vocabulary of zoning findings that the board expects in every staff report. Ordinary first drafts are readable but miss those terms. When is specialization the right move?
• A volunteer fire-district secretary sends about eight equipment-request memos a quarter. A vendor quotes a fine-tune that needs months of labeled memos and a specialist. What should the leader decide?
• A snow-removal bureau wants answers from this winter’s plow-priority list, which is rewritten after every storm. A vendor says they must fine-tune a model on last year’s PDF. What should the leader prefer?
• A city recycling hotline fine-tunes a model on past pickup-delay notes and assumes it will never invent a missed street or a fee. What should the leader recognize?
• A ferry-terminal operations desk compares a cheap general model with a specialized boarding-announcement model they must retrain whenever tide windows and vessel assignments change. How should the leader score the specialized option?
• A cemetery office writes only interment-delay notices in one fixed format. A vendor pitches the largest general model so the office is “covered for everything.” What should the leader do?
• A botanical-garden gift-shop manager rewrites the typed instructions and tells the trustees they have “fine-tuned the model.” What distinction should the leader draw?
• A public-housing waitlist office sees the usage bill jump after staff paste entire applicant packets into every prompt. What should the leader name as the usage meter?
• A voter-registration desk asks the model for eighteen-hundred-word confirmation letters when an eighty-word note would finish the task. What should the leader treat as a cost control?
• A public-works salt-barn office pastes four winters of spreader logs into every morning briefing prompt. Most of the text never appears in the draft. What should the leader flag?
• A volunteer ambulance squad claims “AI saved us fifteen hours” with no pre-AI time study and no named task. What must the leader require before calling the result ROI?
• A municipal campground reservation window frees six clerk hours a week after a generative tool drafts inbound-stay notes. Headcount stays the same, and the extra time is not reassigned. Finance asks where the return is. What should the leader report?
• A municipal ice-rink box office can use a language model on four thousand seasonal-pass renewal notes or on one annual capital-plan memo. Where should the leader aim the usage spend?
• A county bookmobile office drafts two hundred weekly stop-change notes with a model, then a coordinator rewrites each note for ten minutes. The usage bill looks cheap. What should the leader include in the return calculation?
• A volunteer-run lighthouse museum writes five exhibit captions a month and already trusts the curator. A token-heavy pilot would cost more than that writing. What should the leader do?
• A city splash-pad desk now sends lightning-closure notices faster, but several name the wrong reopen hour and caregiver callbacks explode. Cycle time improved; rework and complaint cost rose. How should the leader judge the return?
• A soil-and-water conservation district compares a predictable per-person productivity subscription with a custom generation path billed on tokens. The board wants the cost-shape difference, not a product catalog. What contrast should the leader give?
• A municipal boat-ramp kiosk lets a model invent “Saturday launch hours” that were never on the posted calendar. The hours sound official. What should the leader call that failure?
• A city heating-assistance office wants a chat model to approve or deny winter-fuel applications from a conversational form. What should the leader decide?
• A civic-theatre playbill synopsis changes tone and facts each time staff rerun the same prompt. The board will not publish. What should the leader name as the gap?
• A flood-watch notice generated for a river town assumes every household has a basement, a driveway, and a car for evacuation. Apartment residents say the notice is not for them. What should the leader flag?
• A credit-union draft of small-business outreach letters sounds systematically warmer to certain ZIP codes. Marketing waste and fairness both suffer. How should a leader treat that bias?
• A 311 desk wants instant public replies. Legal wants a four-hour review window because a wrong street-closure time is a safety issue. What should the leader choose?
• A botanical-garden chatbot cites “last year’s USDA hardiness map” that it invented. Visitors replant on that advice. What should the leader require?
• A pediatric dental group’s generated “typical family” packets only show two-parent, English-speaking households. Other families feel unseen. What risk should the leader name?
• A hospice nurse treats a generated medication-timing note as if a pharmacist wrote it and skips the chart check. What reliability failure should the leader name?
• A municipal pool has six closure reasons with fixed legal wording. A generative rewrite keeps adding extra promises such as “we will reopen by noon.” What should the leader prefer?
• A seed-catalog publisher must produce twelve hundred SKU blurbs in three weeks with two writers. What value case should the leader see?
• A county clerk automates first-draft certified-mail cover letters; a human still signs. Where is the automation value?
• A rural electric cooperative writes the same outage-status pattern forty times in a storm. What should the leader mark as high-value automation?
• A municipal recycling facility’s bottleneck is a jammed sort line, not emails. A vendor pitches a generative inbox tool to raise throughput. What should the leader do?
• A ferry-terminal night window still needs English and Spanish sailing-change drafts after the day writers go home. A vendor says generative drafting will cover every shift and language. What should the leader treat as the real scalability claim?
• A volunteer-fire auxiliary already writes the same fifty donor thank-you notes after every pancake breakfast. Generative drafts would finish those notes before supper. The chief could use the saved evening to open a second equipment-check session for new recruits. How should the leader name those two outcomes?
• A rural electric-coop board receives eighty-page rate-hearing packets the Friday before every meeting. Two analysts cannot write a two-page public brief for each item by Monday. What business-value case is this?
• A leaf-collection desk adds an after-hours form that auto-drafts a reply to every missed-pickup complaint. Incomplete one-line submissions triple overnight, and reviewers spend the morning discarding empty drafts. What should the leader pair with the automation?
• Two branch desks at a county animal-control office use the same generative model for lost-pet notices. One prompt lists color, collar, last street, and a callback number. The other says “write a notice.” The first desk’s drafts need light edits; the second desk’s drafts are unusable. What should the leader conclude?
• A parking-authority office asks a model to “make the citation-appeal letter nicer.” Editors then spend the evening rewriting every draft because amounts, dates, and hearing rights are missing or wrong. Where did the vague prompt actually shift the cost?
• One municipal cemetery clerk writes a weekend-service prompt that consistently keeps the section, the start time, and the weather-delay line. Other cemeteries in the same city reinvent weaker prompts each Friday. How should the leader treat that strong prompt?
• Six weekend volunteers each write trail-card blurbs for the same municipal nature center, each with a personal prompt. The cards read like six different brands. What control should the leader apply?
• A community-garden board has never adopted a sliding-scale plot-fee policy. A staffer asks whether a carefully worded prompt can write the official policy so gardeners can be quoted a number this week. What should the leader say?
• A county emergency-management office is about to let staff draft flood-road closures with a language model. What must the leader require up front?
• A municipal street-tree desk wants a more expensive model because pruning notices come back vague. The current prompt never names the species, the block, or the reading level. What should the leader try first?
• A county animal-shelter desk gets clearer hold-notices when the instruction names the role (adoption clerk), the task (hold notice), and the limits (no medical advice, ninety words). What should a leader recognize?
• A town memorial-day parade office pastes two accepted program notes into the prompt so new blurbs match house style. The office has not paid for a fine-tune. What technique is that?
• A municipal leaf-collection office needs street, pickup day, and cart count for Saturday. Staff paste a paragraph and then retype it into a sheet. What should the leader require in the prompt?
• A county floodplain-map desk must list missing parcel documents and then write the resident note. Mixed one-shot prompts have been skipping the list. What should the leader require?
• A public-health immunization clinic keeps seeing drafts that invent Saturday hours and vaccine names the standing order never listed. What should the leader add to the instruction?
• A municipal splash-pad closure notice must be grade-six English for caregivers, not pump-engineer language. The prompt never names the reader. What should the leader require?
• A county bee-inspection office rewrites the same apiary-notice prompt three times, adding bloom week and banned promises. A trustee says the extra passes prove the model failed. What should the leader say?
• A municipal graffiti-removal desk must answer residents only from this year’s published surface-treatment FAQ, not from the open web. What requirement is the leader setting?
• A town ice-rink fee card changes on Labor Day. Last season’s grounded index would still quote the old prices. What must the leader require?
• A credit union wants staff answers grounded in product sheets, but tellers must not retrieve the commercial-credit playbook. What grounding requirement should the leader add?
• A pediatric dental group will publish generated after-visit notes only if each clinical fact points to the chart or an approved protocol. What go-live requirement is that?
• A rural electric cooperative assumes a general model already “knows” their private outage codes. The codes live only in an internal binder. What should the leader require?
• A regional blood-bank donor FAQ must not invent hours or eligibility. What should the leader make mandatory for that surface?
• A municipal archives chatbot is grounded in forty percent of the boxes; the rest are still on paper. Marketing wants to call it a complete archive. What should the leader do?
• A hospice wants medication notes grounded in the protocol binder. Clinical leadership still requires a nurse sign-off. How should the leader treat those two controls?
• A seed-catalog assistant finds the current SKU sheet first, then writes the blurb from that sheet rather than from memory. What pattern should the leader name?
• A municipal parking-authority fine schedule changes quarterly. A one-time fine-tune would be stale by the next board vote. What should the leader pick?
• A community-radio intern indexes show-promo fluff but not the engineering runbook, then wonders why technical answers fail. Where should the leader point?
• A folk festival wants both house tone and this year’s artist list. How should the leader split that problem?
• A town clerk’s desk answers dog-license questions from the published fee card. A vendor says the value of retrieval-augmented generation is “sounding more confident.” What primary value should the leader buy?
• A beach-tag office retrieves last summer’s tag rules instead of this year’s and answers a resident with total confidence. What should the leader conclude?
• A farmers-market coordinator needs a cheerful Saturday welcome poster with no hours, prices, or rules. A vendor insists on wiring retrieval first. What should the leader do?
• A resident asks whether the municipal pool is open Sunday. The assistant retrieves thirty annual reports and still misses the hours card. What must the leader require?
• A public-works safety desk wants short lessons from near-miss write-ups. The vendor indexes plow GPS tracks instead. What should the leader flag?
• A parking desk indexes both last year’s meter-rate card and this year’s with no precedence rule. Residents get opposite fines on the same block. What should the leader require?
• A recreation desk writes example notes only from teen summer-camp chats, then uses that voice for the senior-center newsletter. What should the leader flag?
• A rental-assistance desk photographs a wall of workshop sticky notes and calls that the grounded corpus for benefit decisions. What should the leader say?
• A building-inspection office has moisture-reading tables and inspector emails. A vendor says one generative model should both call wet or dry and draft the resident letter. How should the leader split the work?
• A historic-preservation board wants a tool that imitates past approve, continue, or deny calls. The only files are unlabeled hearing PDFs. What should the leader conclude?
• A rec-center desk still answers pavilion-rental questions from 2014 fee sheets. The council raised fees and changed the deposit in 2024. What should the leader require?
• A town clerk can ground a permit assistant on twelve years of dirty scanned packets — drafts, duplicates, and all — or on a smaller set of adopted ordinances and current forms. What should the leader choose?
• A housing-authority assistant is helping tenants until a staffer pastes sealed eviction files into the public prompt window. What should the leader conclude?
• A city attorney pastes an unreleased settlement dollar figure into a vendor chatbot to polish the memo. How should the leader treat that paste?
• Two vendors both draft accurate sidewalk-repair notices. One is a shared multi-tenant cloud; one is a city-controlled tenant. Quality scores match. What remains for the leader to decide?
• Housing inspectors have been pasting resident street addresses into personal consumer chatbots to draft courtesy letters. What should the leader require?
• A library director wants the catalog assistant on every branch desk, but counsel will not approve until access control and logging exist. How should the leader frame security?
• A volunteer fire-company clerk says the station laptop has a login password, so pasting incident notes into a public chatbot is already secure. What should the leader say?
• A 311 desk is drowning in ticket volume. A vendor pitches a friendly chatbot. The leader’s real need is next week’s load so the desk can staff. What should they buy?
• A records clerk must route incoming building-scan PDFs as permit, inspection, or complaint. A vendor offers a model that writes a lyric about each scan. What should the leader require?
• A county blood-donation van must estimate Saturday’s likely shortfall of O-negative units from labeled weekend history so the hospital can restock. A vendor says their chat writer can guess that number in conversation and the office can skip a forecast. What should the leader require?
• A street-lighting bureau wants two results from each cabinet photo: a flag when the panel is overheating, and a first-draft work order for the night crew. A vendor says one generative writer should detect the fault and write the order. What should the leader require?
• A municipal skate-park desk has eleven days of Saturday gate counts and wants a summer-demand model for session staffing. A vendor says those eleven days are enough to fit a predictor. What should the leader decide?
• A lakeside picnic grove already posts a fixed rule: no open fires on any red-flag day. A vendor wants to train a model to learn when fires are allowed so staff can stop reading the notice. What should the leader prefer?
• An animal-control kennel is eager to start a training job on intake photos. No one has named the decision the model should change or the metric that would prove the work is worth it. What should the leader require first?
• A mosquito-control bureau is told a vendor can install the trap-count model in one week and then leave. Staff ask whether that one-week setup is the whole lifecycle. What should the leader explain?
• A downtown curb-occupancy model was fitted on 2019 counts. The city then opened a new civic-center garage that changed where cars park, and the 2019 model is still in production. What should the leader require?
• A vacant-lot mowing model goes live this month. The vendor’s contract ends in four weeks, and no bureau employee is named as owner for data, validation, or monitoring. What should the leader require?
• A leaf-collection routing model was trained on last October only. The bureau wants to go live this fall without testing on a held-out season. What should the leader require?
• A compost-drop-off desk’s Saturday-overflow predictor now loses to a simple posted rule: if the bin was full Friday, send a truck. The model still bills every month. What should the leader do?
• A county animal-shelter website hosts a public lost-pet assistant. Anyone can type a question and anyone can upload a photo. A trustee calls it “just a chat box.” What should the leader require?
• A municipal compost-subscription desk keeps prompt and output logs that still show member names, cart numbers, and street addresses. A clerk says logs are scratch paper, not records. What should the leader require?
• A county weights-and-measures office posts a shareable link that generates inspection notices on official letterhead. The link does not ask anyone to sign in. What should the leader require before that generator can act in the office’s name?
• A municipal parking-permit counter uses one retrieval assistant. The front-desk identity can pull permit files and can also search employee personnel folders. What boundary should the leader set?
• A county mosquito-control district answers public spray-schedule questions in the same session that opens staff fuel worksheets for the truck fleet. What should the leader require?
• A city code-enforcement desk can generate a correction letter that looks official and can start a compliance clock, but the file shows no signed-in actor. What should the leader require?