An AI consultant San Diego shop should hire for leftover, not for slide count. If you need a live workflow in the customer relationship management (CRM) system, practice management system (PMS), inbox, or dealer board you already pay for, book the consultant who will sit with Friday and write the connector. Nidrosoft is the San Diego studio of Cyriac Zeh, a design engineer who implements AI for small and mid-size businesses: audits, custom builds, phone and receptionist agents, training, and systems the team can own. A deck remains the right leftover when a board cannot vote until the story exists.
Key takeaways
- Compare leftovers. A strategy pack, a boutique scorecard, a product-shop app, a seat license, and a live process are different invoices.
- A build leftover is code, prompts, evaluations (evals), and a runbook in your accounts. A deck leftover is a narrative the board can vote on.
- Field names are the test.
RecallStatus,LastAttemptAt, andDoNotContactbeat "AI-powered engagement." - Fit is usually 30-250 people, or a smaller shop if one workflow is already clear. Workshops have been run in La Mesa.
- Other San Diego firms can win the deck row, the shop row, or the classroom row. Score them on their public positioning, then pick one leftover per invoice.
- Book a free audit if you want the map before anyone writes to production.
Contents
- Deck leftover versus build leftover
- What a consultant who builds actually does
- When you should still buy a deck
- The field-name test
- Local firms and honest rows
- San Diego work that needs a write path
- How Nidrosoft handles this
- Buying the audit without mixing invoices
- Frequently asked questions
- Leave with a process or leave with a pack
Deck leftover versus build leftover
A slide deck answers a board. A live workflow answers a Tuesday. Both are legitimate purchases. They fail when they share one statement of work.
A strategy unit interviews executives, writes frameworks, and leaves a pack the board can vote on. Typical speed is a quarter or more to first materials. You own recommendations and a governance model. That is the right leftover when nobody may write to production until the story exists.
A boutique consultancy diagnoses, shortlists vendors, and leaves scorecards. Typical speed is several weeks to a few months. You own the notes. That is the right leftover when your own lead engineer will watch Friday.
A product shop takes a spec and leaves an application. Typical speed is months set by scope. You own the repo if the contract says so. That is the right leftover when the objects and screens are already known.
Seat vendors leave licenses, a usage graph, and prompts if someone saved them. That is the right leftover when drafting and summary speed is the whole job.
A consultant who builds leaves a ranked map, then a process in your accounts. Typical speed is an audit in two weeks and a first system after a written plan. You own code, prompts, runbooks, and the system. That is the right leftover when the job is a repeating workflow inside tools you already pay for.
The difference shows up in the first calendar invite. Who is on it. Are they the people who will commit code. Can they accept a no from security. Will they sit with a service advisor for three hours and watch a broken field in the dealer system. Those questions decide speed more than a logo row.
If a proposal you received is 40 slides and no field names, you bought a deck. If a proposal lists RecallStatus, LastAttemptAt, OwnerUserId, and DoNotContact, you are closer to a build. AI consulting alternatives exists so you can keep those leftovers in separate rows. The county operating version of the build leftover is AI implementation for San Diego small businesses.
This page will not invent a Nidrosoft price, a confidential client, or a claim that every deck is waste. A Gaslamp hospitality group that cannot get capital committee time without a portfolio story should buy the pack. A Kearny Mesa lot that already knows the night job is private-party cars should buy the write path.
What a consultant who builds actually does
The first useful hours are with the person who already does the repeating job. That person is rarely the vice president who booked the call. It is the hygiene coordinator who works a recall report between check-ins, the inventory buyer who still opens Facebook Marketplace at 6 a.m., the listing coordinator who pastes remarks into the multiple listing service (MLS), or the office manager who rebuilds the same spreadsheet every Friday. ChatGPT paste history and Microsoft Copilot experiments are data, not embarrassment.
The job after watching is costing. Each recurring task gets hours, feasibility, and payback. Feasibility includes whether the CRM, PMS, dealer management system (DMS), or shared inbox has an application programming interface (API) or a usable export. It includes whether counsel will allow customer text in a third-party model. Payback includes time returned and errors avoided. A homepage chatbot loses this rank unless the hours and the access support it.
Scope is a paragraph, not a vision slide. It names what runs without a person, what a person still approves, and what you will measure after thirty days. Money movement, contracts, and customer sends stay under a named human. Nexuvo can reach a private seller overnight. A person still buys the car. Eviction Wizard can assemble a notice-to-writ path. A person still files. TokenTra can flag wasted model spend. A person still sets the budget. Omnira Dental can draft a confirmation. A coordinator still sends it.
Build happens in the company's tenant. GitHub, GitLab, or the host you already pay for. Linear, Jira, or the board the standup walks. Slack or Microsoft Teams for the thread that holds the unofficial path. The consultant asks for a time-boxed role with logging, not a personal admin account they will keep forever. Prompts live next to the code. Keys live in your vault. Nerlude exists because founders lose credentials in chat.
Own means Tuesday does not require the consultant's Slack. Someone internal can revoke the guest account and still close the week. If nobody internal will pair, the honest next step is a job posting instead of another month.
The four service names under that work are AI audits, AI builds, AI transformation, and corporate AI training. An audit is two weeks in the work and a ranked map you can take to another builder. A build is one live process. Transformation is the same method across more than one team, still starting in one department. Training is a sprint on your files with a playbook you keep. Workshops have already been run in La Mesa. None of those line items is a junior bench behind a partner who flies in for kickoff.
Phone and receptionist agents sit under builds when the phone is the workflow. Same-day software-as-a-service (SaaS) tools such as FrontDesk, DentalAssist, Rosie, Goodcall, Smith.ai, and Voksha can cover a simple script this week. Custom is the better fit when the agent must read the PMS, speak Spanish first for South Bay callers, or write a row the coordinator already trusts.
When you should still buy a deck
Buy a deck when the constraint is governance, not Tuesday. A board that cannot approve a production write until it has a portfolio narrative will stall every connector. Hiring an implementation seat into that stall wastes two weeks of Embed on people who are not allowed to show you the unofficial stack.
Buy a deck when three departments want five agents and nobody will name the first coordinator. A strategy pack can force a sequence. An implementation audit that starts in a steering committee with no floor sponsor will invent hours.
Buy a deck when counsel has not yet said what text may leave the building. A build that guesses that sentence is a liability. A pack that names the decision, the data classes, and the human gates is the right leftover. Come back when the sentence exists.
Buy a deck when you are comparing a company-wide operating model for a 1,000-person group. Past 250, this studio still works if you pick one department. A multi-market property firm or a hospital-owned dental service can buy an audit on one floor. A group that wants a single narrative for every market should hire the strategy unit first.
Do not buy a deck when you already know the night job. A Kearny Mesa dealer who can say "private-party cars after dinner" does not need a 40-slide opportunity landscape. A La Mesa practice that can say "hygiene recall list with dates" does not need a maturity model. A Chula Vista desk that can say "Spanish-first callers hang up on the English prompt" does not need a journey map. Those sentences are already Scope.
If someone is selling you both leftovers in one invoice, split the invoice. The AI consulting alternatives page exists for that split. Mixing them is how you pay for adjectives and still retype the row.
The field-name test
Ask the consultant to write the columns before they write the vision. A serious build conversation sounds like a schema review.
For hygiene recall: PatientId, RecallDueOn, LastAttemptAt, AttemptChannel, OwnerUserId, DoNotContact, RecallStatus. For inventory hunt: SourceUrl, AskPrice, Mileage, NeedsHumanBuy, CrmRowId. For notices: UnitId, NoticeType, ServedOn, ReadyToFile, CounselSignoff. For after-hours phone: JobType, Zip, UrgencyFlag, LanguagePref, AcceptNextMorning, TransferTarget. For token spend: Vendor, FeatureName, TokenCount, WastedFlag, BudgetOwner.
A proposal that cannot name four of those columns, or four of yours, is a deck wearing a build label. A proposal that names them and still wants a 40-slide kickoff is two invoices. Keep the columns. Buy the write path. Skip the theater.
Run a one-hour test on a recorded week. Pull ten real tasks. Ask the consultant which ones a person must still approve. If they say none, the write path is dishonest for money, filings, and clinical notes. If they say all ten, they are selling a chatbot that you will paste into. The useful answer is a split: these four run, these three wait for a human, these three stay manual.
Bring leftovers to the same hour. Copilot usage. A Rosie, FrontDesk, Goodcall, Smith.ai, DentalAssist, or Voksha trial. A prior deck. A product-shop repo. Model invoices if "we already use AI" is part of the brief. A consultant who ignores prior spend will duplicate it.
Write the leftover into the statement of work before you compare day rates you do not have yet. This page does not publish Nidrosoft prices. Code, prompts, evals, runbooks, and the system in your accounts is one leftover. A tenant you rent and cannot export is another. A deck is a third.
| Leftover | Typical artifact | When it is the right buy |
|---|---|---|
| Strategy pack | Board narrative, governance model | Nobody may write to production until the story exists |
| Boutique scorecard | Vendor shortlist, notes | Your own engineer will watch Friday |
| Product shop | Application against a known spec | Objects and screens are already named |
| Seat licenses | ChatGPT or Copilot usage | Drafting and summary speed is the whole job |
| SaaS receptionist | Live number this week | Simple script, simple calendar |
| Build | Live process in your accounts | Repeating workflow, named owner, access |
Local firms and honest rows
San Diego has public consulting and software seats. Praise what they actually publish. Do not invent a ranking.
Parsons AI publishes AI consulting and automation. Applied Intelligence publishes AI consulting. San Diego AI under James Schlauch is educator and consultant shaped. AI Upside Group is consulting shaped. SideGuy in Encinitas is custom software. Christian Johnston's "Your AI Guy" brand is his. Do not steal the phrase. Do not smear it.
Those firms can win. A classroom every Tuesday is a staffing model this studio does not publish. A spec you already wrote is a shop job. A board narrative is a strategy job. An implementation seat wins when the leftover is a ranked map and a write path.
Same-day SaaS receptionists remain a real row. FrontDesk, DentalAssist, Rosie, Goodcall, Smith.ai, and Voksha can put a voice on the line this week. That speed is a strength. Custom is the better fit when the agent must read the PMS, speak Spanish first for South Bay callers, transfer with context, or write a trusted row.
The partners page keeps those comparisons in one place. Use it when this page starts to feel like a studio brochure. If another consultant passes the field-name test and you already have a spec, hire them. Take our audit map with you if you bought one. We would rather lose the build than watch a shop guess.
San Diego work that needs a write path
The county's unpaid work has addresses. North County clinics along Coast Highway in Encinitas and Del Mar live or die on the confirmation text. A deck about "patient engagement" does not send the text. A queue with RecallDueOn does. South Bay shops in Chula Vista take bilingual callers who will hang up on an English-only prompt. A journey map does not keep Spanish through the confirmation. A transfer graph does.
East County trades in El Cajon and La Mesa close from the truck, then rebuild the estimate inbox at night. A maturity model does not return the estimate. A follow-up queue does. Oceanside and Carlsbad mix contractor days with tourism overflow. A vision slide does not know that Saturday morning holds exist and Sunday does not. Kearny Mesa lots run private-party inventory after dinner. A chatbot on the homepage does not create the CRM row. Nexuvo-shaped outreach does, and a person still buys the car.
La Jolla and UTC sit next to biotech and still have small companies that do not want a 40-slide operating model. Downtown Gaslamp hospitality runs reservations and no-shows. Military families move on orders and break recall lists. A system that cannot mark DoNotContact will annoy people who already asked to be left alone.
Workshops have been run in La Mesa. That is a published fact. It is a training leftover, not a build leftover. If the honest need is judgment on ChatGPT and Copilot, buy training and seats. If the honest need is a write to Open Dental or the dealer board, buy the audit and then the build.
Commute time is part of the calendar. A working session in La Mesa is a different drive than a screen share into an Oceanside yard. Remote audits are normal when systems are reachable. In-room days are useful when the unofficial stack lives on a clipboard. Say which you want when you book.
How Nidrosoft handles this
Nidrosoft handles a consultant search by selling the leftover in the title. Cyriac Zeh is the person in the standup. Formerly Anthropic, Microsoft on Bing and Edge, Intuit, and Gap Inc. 12+ years, 125 products shipped, 16 public. There is no junior bench behind a partner.
The commercial start is an audit. Week one is Embed. Week two is Diagnose and the first pass of Scope. You can stop after the packet. Many teams should. If you continue, Scope names the first outcome in plain sentences. Build writes into your accounts. Own names the inheritor before the first paid month.
We will not take a brief that is only "make us AI native" with no department and no repo. We will not take a brief that forbids the unofficial stack. We will not take a brief that requires twenty named consultants. We will not take a process that happens once a year and call it an engineering seat. We will not sell a homepage chatbot as the default first system. We will not publish a Nidrosoft price on this page. We will not claim a Health Insurance Portability and Accountability Act (HIPAA) certification the facts page does not carry.
If the honest next step is a strategy pack, we say that in the audit. If the honest next step is FrontDesk or Copilot seats, we say that. If the honest next step is SideGuy or an in-house hire, we say that. The last option is a valid audit.
How we work is the five words in calendar form. AI audits is the two-week packet. Book a free audit if you want the map before the write. Write cyriac@nidrosoft.com if a calendar slot does not fit.
Buying the audit without mixing invoices
Use the public calendar at https://calendar.app.google/GCtfuTx3Ms6K7d579 or write cyriac@nidrosoft.com. Name three things in the first note: the repeating job, the system of record, and the owner. If you already have a deck, attach it and write the boundary. If you already have a SaaS receptionist, write which calls it owns. If you already have a product-shop repo, write what it must not duplicate.
Bring field names if you have them. Bring leftovers. Expect friction in week one. Operators protect workarounds because the workaround is how Friday still closes.
At the end of two weeks you receive the ranked map, the risk notes, and a recommended next purchase. That next purchase might be seats and training. It might be a scoped build. It might be a longer transformation program after one department is live. It might be nothing from this studio.
Do not put "strategy, vendor selection, custom build, and change management" in one request for proposal (RFP). Split the leftovers. The county guide is AI implementation for San Diego small businesses. The firm-shape table is AI consulting alternatives. This page is the hiring rule for people who already know they need a write path.
Frequently asked questions
Is a consultant who builds still a consultant?
Both nouns fit badly and both get typed. Compare leftovers, not titles. Strategy firms leave decks. Boutiques leave scorecards. Product shops leave apps against a spec. This studio leaves a ranked map and, if you continue, a live process in your accounts. AI consulting alternatives is the table version.
When is a deck the better first purchase?
When a board cannot vote until the story exists, when counsel has not named what text may leave the building, or when nobody will name the first coordinator. Buy the pack. Come back when Friday has an owner.
Do I have to buy a build after the audit?
No. The audit is a complete purchase. Many teams should stop, buy seats, hire in-house, or take the packet to a shop. We would rather lose the build than watch a shop guess without a spec.
What size company should book a build-shaped consultant?
About 30-250 people is the usual fit. Smaller shops work when one workflow can pay for itself. Past 250, pick one department and a sponsor who owns the floor. A board-only story with no named coordinator is the wrong first room.
Can we start with a receptionist subscription?
Yes. FrontDesk, DentalAssist, Rosie, Goodcall, Smith.ai, and Voksha can start same day. Measure missed calls and booking rate for thirty days. Book a custom agent when the script must read your PMS, route Spanish-first South Bay callers, or write a trusted row.
How do we book without a price on this page?
Book a free audit or write cyriac@nidrosoft.com. Prices wait for stack and access. Any number on another firm's site is theirs. Bring the repeating job, the system of record, and the owner.
Leave with a process or leave with a pack
If you need a build, hire the consultant who will write field names, sit with the unofficial stack, and leave a process in your tenant. If you need a deck, hire the firm that sells decks and do not mix the invoices. The public proof on this side is 16 systems and three published results, not invented logos. Book a free audit if the leftover you want is a map, then a write path.