Which AI consulting alternative should you buy?

Operators type the same sentence in different words. They have already sat through a vendor demo. Someone on the board asked for an AI plan. A department lead wants the inbox, the inventory hunt, or the intake form to stop eating the week. The search box then fills with strategy units, boutique advisors, offshore benches, and seat licenses. Those are different purchases. Mixing them in one RFP wastes a quarter.

Nidrosoft is a San Diego product and AI engineering studio. Cyriac Zeh runs it after design and engineering roles at Anthropic, Microsoft on Bing and Edge, Intuit, and Gap Inc. The studio has shipped 125 products and keeps 16 systems public. The method is Embed, Diagnose, Scope, Build, Own.

Public work is the proof we can name. Omnira Dental for practices. Octana for a video from one message. Eviction Wizard from notice to writ. Guidera on iOS. TokenTra, where teams cut 10 to 30 percent of wasted tokens. Protectron for EU AI Act trails. Nerlude for infrastructure. Nexuvo dealers acquire 3x more inventory with 80 percent less manual outreach. NanoBrief writes a brief in under two minutes instead of three or more hours. Clear Audio AI and MixGuidera for sound. Those are systems you can open.

This guide stays inside one job. If you already know you want a workshop rather than a build, use the implementation versus workshops page. If you already know you want a person in the room and need to compare who that person is, use the partners page. Here the job is choosing among consulting shapes.

Most buyers already lean toward one of five shapes. A strategy unit writes the story the board can defend. A boutique AI consultancy diagnoses and shortlists vendors. A product or development shop builds a specified app. ChatGPT Enterprise or Microsoft Copilot seats raise individual throughput on drafting and summary work. Nidrosoft embeds, ranks payback, and ships a process into the stack you already operate. Each row below is a fair read of what that shape usually leaves on the table.

Price is missing on purpose. Strategy units and boutiques bill by project or day. Shops bill by sprint. Seat vendors publish list prices that change by region and bundle. Nidrosoft scopes after an audit. If a page invents euros or a retainer to look complete, treat that page as marketing. Book time on the calendar if you want a number that matches your stack and your access.

Company size changes the row you should stare at. Under 30 people, one build has to pay for itself or the work should wait. Between 30 and 250, a named owner and a repeating workflow is the usual fit. Past 250, start in one department with an internal sponsor. A strategy firm can still be the right first seat for a board that needs a portfolio story before anyone is allowed to write to production.

A top-tier strategy consultancy wins when the artifact the company needs is a position the board can defend. That includes a multi-year transformation narrative, a risk and governance model that legal and the audit committee will sign, and a map of where AI sits across a portfolio of business units. Those firms hire for synthesis at executive altitude. They are built to hold a room of fifty people and leave a document everyone can point at.

They also win when the company is not ready to grant production access. If security will not approve a builder in the CRM, if works councils or legal still need a paper trail before a write path exists, or if two divisions are still arguing about who owns the data, a strategy unit can unblock that argument. Nidrosoft needs a decision maker and a path into the tools. Without those, an embed is theater.

Buy the strategy firm when the first deliverable must be a board pack. Buy it when regulators, investors, or a parent company asked for a plan and will not accept a working intake bot as proof. Buy it when you need org design, incentive changes, and a operating model on paper before anyone writes a connector. Those are real jobs. They are different from shipping Nexuvo-style outreach into a dealer CRM or wiring Omnira Dental into a practice schedule.

Nidrosoft can sit after that work, or beside it in one department, if the sponsor wants a live process under the narrative. We do not replace a firm that is being paid to reposition a twenty-country group. We do not write the enterprise risk taxonomy for a bank. If that is the mandate, hire the firm that does that every week. Keep our name for the later moment when a clinic, a lot, a fund ops stack, or a software team needs the system itself.

Boutique AI consultancies win when the company already has engineers, or already has a preferred vendor, and needs an outsider to force a ranked list. They are good at vendor bake-offs. They write the scorecard that keeps a noisy RFP honest. They will sit with a product owner for a few weeks, name three pilots, and leave. That is a complete purchase if your internal team will take the pilot to production.

They also win when the buyer wants advice without a long embed. Some operators are allergic to another team in Slack for a quarter. They want a short, senior conversation, a written recommendation, and their Fridays back. A boutique that refuses delivery is cleaner than a studio that keeps offering to stay. Hire that boutique. Read the report. Staff the next step with people you already employ.

The failure mode is a roadmap that nobody can execute. A handsome sequence of use cases with no owner, no access plan, and no one who has shipped a connector into the actual ERP will sit in a shared drive. Nidrosoft’s Diagnose step looks similar on a slide (a heat map of hours, feasibility, and payback) and then we stay for Scope and Build. If you do not want us to stay, do not hire us for the map. Hire the boutique and keep the work in-house.

Use a boutique when vendor selection is the job. Use Nidrosoft when the job is a live workflow in the tools you already run, with prompts and runbooks your people can change. Those two sentences should settle most bake-offs between us and a diagnosis-only shop.

A product studio or development shop wins when the thing you want is an application whose behavior is already agreed. A marketplace with known objects. A mobile client with a signed design. A feature on a product you already ship. Those teams are built for tickets, design QA, and a release train. If the model calls are a small part of a larger product, and a product manager can write acceptance tests, a shop is often faster and easier to manage than an AI studio that wants to re-diagnose the company.

They also win when you need a bench. Nidrosoft is a studio around Cyriac Zeh, not a fifty-person delivery floor. If you need parallel squads, a dedicated QA cell, and a project manager for a nine-month greenfield, hire a shop that already runs that machine. Ask them who writes the prompts, who owns evals, and what happens when the model changes. Put those answers in the contract.

Shops lose when the hard part is the messy process, not the screens. Inventory outreach that has to live on Facebook Marketplace, Craigslist, and the dealer CRM at once (the class of problem Nexuvo was built for) is a process problem. Token waste across providers is an observability problem. A notice-to-writ path is a legal operations problem. A shop that waits for a perfect spec will stall while the floor team invents another spreadsheet. That is the work we take.

Read the contract on ownership. Some shops keep libraries, agents, or prompt packs. Nidrosoft builds in your accounts so the system is yours from the first commit you accept. If a shop will do the same, and the spec is already true, they can be the better buy. We will say that on a call rather than stretch a build to fit our method.

Seat licenses win when the job is personal throughput. Lawyers drafting. Analysts summarizing. Engineers asking Copilot to finish a function. Marketers pushing a first pass. Roll the seats, write a short acceptable-use note, run a lunch-and-learn, and measure whether people actually open the tool. That program is cheaper and faster than any studio when the process you care about already lives in documents and tickets.

Seats also win as a first experiment. A 40-person firm that has never paid for a model should not start with a custom agent in the ERP. Give ten people ChatGPT or Copilot for a month. Watch what they paste. The paste list is a cheap Diagnose step. Nidrosoft will ask for that list in an audit. If you already know the list is “emails and meeting notes,” stay on seats until a workflow repeats enough to justify a build.

Seats lose when the work sits behind a login the model cannot see. Practice management software, dealer inventory, eviction files, token invoices, floor quality sheets. A person can copy a row into chat. That does not create a system. TokenTra exists because teams were already paying models and still could not see spend by feature. Protectron exists because a chat window does not classify an AI system for the EU AI Act. Omnira Dental exists because a practice cannot run chair-side operations out of a sidebar.

Nidrosoft is a poor fit if the only ask is “get everyone Copilot.” We will tell you to buy the licenses and, if you want help, book training so people practice on their own files. Corporate AI training is a service we sell. It leaves a playbook. It does not pretend a playbook is an integration.

Nidrosoft sells AI audits, AI builds, AI transformation, and corporate AI training. An audit sits with the people doing the work and costs recurring tasks. A build ships agents, automations, or a product into the ERP, CRM, inbox, or customer app. Transformation is the longer path from the first audit through builds, training, and rollout until your people run it. Training is a focused sprint with leadership or engineers on your own tools, with a playbook you keep.

Embed means we go into the company. Diagnose means each process is scored on hours eaten, feasibility, and payback. Scope means you buy a result: what runs itself, what a person still approves, what it saves. Build means the system goes live in your stack while someone in-house learns it. Own means there is nothing theatrical to hand over because it was yours from day one. We stay for what you ask and stay out of the rest.

The public archive shows the kind of system we mean: Omnira Dental, Octana, Eviction Wizard, Guidera, TokenTra, Protectron, Nerlude, Nexuvo, MixGuidera, Clear Audio AI, NanoBrief. Sixteen public systems, 125 shipped, more than twelve years of making software. We do not list confidential clients to pad a slide.

The person you meet is Cyriac Zeh. That matters against firms where a partner sells and a junior bench delivers. It also caps how many workstreams we take at once. If you need a factory, look at a shop.

Buyers ask the same six questions in every call. Who sits in the room. What artifact leaves. Who can change it after month three. How fast the first real process moves. Whether money and customers stay under a human. What it costs. The last question has no public Nidrosoft number on this site. The other five fit in a table.

Typical speed is a range. Strategy materials often take a quarter. Boutiques can finish a diagnosis in weeks if access is easy. Shops move at the pace of the spec. Seats land in days. Nidrosoft audits in the first two weeks and implements after a scoped plan, commonly in a stretch the how-we-work page lists as weeks five through ten for the first build. Access delays move the calendar.

If two rows look equally true, pick the artifact you need in ninety days. Board vote: strategy firm. Shortlist: boutique. Specified app: shop. Faster drafting: seats. A repeating process in your accounts: test Nidrosoft with an audit.

Ask every bidder to list the files they will leave. Strategy work leaves recommendations, sometimes a capability model, sometimes a training curriculum on paper. Boutique work leaves the scorecard and the notes. A shop leaves a repository if the statement of work says the IP is yours. Seat vendors leave an account. Nidrosoft leaves the system in your cloud or your tenant, the prompts, the connectors, the eval notes, and the runbook for the person who will operate it on Tuesday.

We build in your accounts when we can. Azure, AWS, Google Cloud, Cloudflare, or private infrastructure, with EU hosting and EU data residency when the work requires it. Approved access. Audit trails. That list is how-we-work, not a regional claim. If your counsel needs data in a specific region, say so in the audit. Do not assume a San Diego studio stores everything in California.

Ownership also means a human still signs money, contracts, and customer-facing actions. Nexuvo can hunt listings overnight. A person still buys the car. Eviction Wizard can assemble the path. A person still decides to file. TokenTra can flag waste. A person still sets the budget. If a vendor offers a fully autonomous write to customers or cash without an approval step, ask them who carries the error.

Training is part of Own. Someone in-house should be able to explain the system end to end. If that person leaves, the runbook and the repo should be enough for the next person. If a partner’s model depends on staying forever, you bought a dependency. We will stay when you ask. The goal of the method is that you do not need us for the Tuesday work.

• Code in your repository or tenant
• Prompts and eval notes you can edit
• Runbooks for the operator on the floor
• Connectors and account access you control
• A ranked map from the audit, even if you stop after Diagnose

Skip us for a board-only mandate with no path to a build. Skip us if the only budget is seat licenses and the success metric is adoption of chat. Skip us if you need a fifty-person delivery machine for a greenfield consumer product. Skip us if you want a legal opinion on the EU AI Act as a standalone purchase. Protectron is a product for teams who will do the work in software. It is not a law firm.

Skip us if no one with authority will join the audit. We can interview a team and still fail if the person who owns the CRM never appears. Skip us if security will not grant a builder any environment, including a copy. Skip us if the process happens once a year. Automation pays back on repetition. A beautiful annual report is a document problem.

Skip us if you want a guarantee that a model will never be wrong. We design for review, correction, and a human on the consequential action. Clinics, lots, funds, and landlords live with that constraint already. A vendor who promises perfect autonomy is selling a story we will not sign.

Smaller teams still fit when one build pays for itself. Larger organizations fit when one department goes first. We will not boil a 4,000-person company in one statement of work.

The first two weeks are Audit and discovery. We map how work moves and cost every recurring task. You leave with a ranked roadmap. Some buyers stop there and staff the rest themselves. That is a valid end. Some buyers continue into strategy and a scoped build. Some buy training so the leadership team can see the same map on their own tools.

Hello@nidrosoft.com and the public calendar are the doors. Cyriac Zeh is in San Diego. The work ships wherever the systems already run, including teams outside California. We do not invent a local office to win a search. If you need a partner who lives in your city, say so and we will tell you whether remote embed is enough.

Bring access notes, a list of tools, and one process that hurts. If you already have a strategy deck from a larger firm, bring that too. We will not rewrite it for sport. We will say which item on page 40 is a process we can ship, and which item still belongs to the firm that wrote the deck.

If the audit says seats are enough, buy seats. If it says a boutique scorecard is enough, hire the boutique. The page you are reading is only useful if that sentence is allowed to be true.

If the remaining question is who sits in the room and who types the code, open the partners guide. If the remaining question is whether a half-day room is enough, open implementation versus workshops. Services pages cover audits, builds, training, and transformation in more depth. Explore pages cover consulting, readiness, workshops, San Diego rooms, and enterprise implementation for people who arrived on those searches.