Enterprise AI implementation, one department first

Buyers type enterprise AI implementation when the company is large enough that a single automation feels too small and a chatbot feels embarrassing. That instinct can be right. It can also be a word looking for a budget. Nidrosoft will not accept the word unless there is an internal sponsor, a first department, a system of record, and a measure that already exists. Past 250 people, that constraint is how we stay out of the strategy-firm lane. Those firms win when the leftover is a company-wide narrative. We win when the leftover is a series of live processes the departments can run.

The transformation service is the program-shaped offer. The builds service is the unit of work. This explore page exists because enterprise is what people type. If you are under 30 people, you probably want a single build, not this URL. If you are between 30 and 250, you may want two departments sequenced. If you are past 250, start here and still pick department one.

Cyriac Zeh founded Nidrosoft LLC after Anthropic, Microsoft on Bing and Edge, Intuit, and Gap Inc. One hundred twenty-five products shipped. Sixteen public. Enterprise implementation here is a studio thread, not a hundred-person army. If procurement requires twenty CVs in month one, we will fail that sheet. Hire a firm that staffs armies. Keep us for the first stubborn process if the sponsor wants judgment on the floor.

Phase one is the audit. Two weeks in the department if access is ready. Ranked processes, systems, data, risks. The packet names the first outcome. If it cannot, stop. Buy seats or a leadership sprint. Enterprise implementation without a first outcome is a steering slot.

Phase two is the build. A written result. Human approvals on money, contracts, and customers. Code and prompts in your accounts. An operator who can run Tuesday. How-we-work often places implementation after the plan, in a multi-week window commonly described as weeks five through ten for the first system. Enterprise does not shorten that window by slogan. It only funds what comes after.

Phase three is training beyond the operator. Adjacent roles practice. Leaders learn to reject add-ons that would break Own. The playbook becomes the internal course. Skip this and team two will treat team one’s system as folklore.

Phase four is the next team. Thin remap if systems are shared. Full Diagnose if they are not. People from team one teach. We stay where needed. We leave where Own is true. A go or no-go sits between phases. Stopping after team one is a complete purchase.

• Department one named before any banner
• One measure per build the desk already believes
• Training files in your drive
• A rule for what team two may copy
• A rule for when Nidrosoft leaves a workstream

If quarter one must produce a pack and nothing in production, hire the strategy firm. If it must produce Copilot usage, buy seats. If it must produce a module on one platform, hire that SI. If it must produce three specified apps, hire shops. If it must produce a live process in tools you already run, stay on this page.

No enterprise price list lives here. A two-site operator and a multi-country group are not the same object. Book the audit. We will not invent a retainer to look like a mega-firm.

Each build leaves the running workflow, source, prompts, eval notes, runbooks, keys, and logs in accounts you control. The program adds an index: process, operator, measure, canonical prompts. Think of Nerlude’s instinct (one place for messy infrastructure) applied to AI operations. You do not have to buy Nerlude. You do have to have a place.

TokenTra-shaped visibility belongs once more than one team spends on models. Teams have cut 10 to 30 percent of wasted spend when they can see tokens by feature. An enterprise rollout that adds agents without a bill view will hide waste inside the success story.

Protectron-shaped trails belong if you operate under the EU AI Act or sell into teams that do. The program can fund the habit. It cannot replace counsel. Write that into the charter. Azure, AWS, Google Cloud, Cloudflare, private infrastructure, EU hosting and residency when required, approved access, audit trails: those constraints are how-we-work, not an upsell surprise.

Playbooks stay in your drive. Strategy packs from another firm stay theirs to update. Do not mash the files. New hires and assistants will quote the wrong one.

Every enterprise build keeps a human on consequential actions. The banner does not relax the rule. Clinics, landlords, dealers, and funds already know why. If the signer is a committee that meets monthly, the process cannot be overnight. Change the process or change the committee. Write the signer’s name into Scope.

Works councils, security review boards, and legal intake should appear in the audit, not in week eight of a build. We will wait for a copy environment. We will not invent a shadow tenant to look fast. San Diego is where Cyriac lives. Your data sits where counsel requires.

If another firm owns enterprise risk language, let them. We implement inside that language when it is specific. We flag where the pack assumes a write path that does not exist. We will not rewrite their pack for sport and we will not let their pack pretend a live process exists when it does not.

Money, contracts, and customer sends stay human. Nexuvo can reach a seller. A person buys the car. Eviction Wizard can assemble a file. A person files. Omnira Dental-class operations keep clinical exception with a person. Enterprise does not mean unsupervised.

We will not staff a communications-only transformation. We will not staff a first deliverable that is only a homepage bot. We will not staff a program that forbids a human on customer send. We will not staff a fifty-person bench in month one. We will not be a logo next to a strategy firm while being asked to write their pack.

We will not ignore the unofficial stack. If inventory still lives on Marketplace, look at Marketplace, the way Nexuvo does (3x inventory, 80 percent less manual outreach). If briefs still take three hours, look at the brief, the way NanoBrief does (under two minutes). If audio is the product, look at Clear Audio AI and MixGuidera, then still rank them against the inbox. Enterprise implementation that only sees official SOPs will automate the wrong desk.

We will not pretend Cyriac can hold four serious workstreams at once. Write capacity into the plan. If waiting is unacceptable, hire shops for specified work and keep the studio on the unspecified process. Rent-an-AI-engineer can extend a seat after the first outcome, still without a public monthly number on this page.

Omnira is a multi-vertical platform with Omnira Dental live and other verticals in build. That product fact shows a studio can carry more than one operations surface. It is not a claim that your enterprise will receive Omnira. Eviction Wizard, TokenTra, Protectron, Nerlude, Octana, Guidera, Nexuvo, NanoBrief, Clear Audio AI, and MixGuidera show other edges. Your SOW will use your nouns.

Numbers we can repeat stay limited. Nexuvo 3x inventory and 80 percent less manual outreach. TokenTra 10 to 30 percent wasted spend. NanoBrief under two minutes versus three-plus hours. We will not invent an enterprise ROI multiple for a steering slide. Your finance team can compute one after department one has a month of real use.

Formerly Anthropic, Microsoft, Intuit, and Gap explains taste for production constraints. Those companies are not listed as customers of your program. Nidrosoft LLC is the contracting entity. The AI Architect is the practice name.

Use measures the desk already argues about. Time to complete a file. Inventory acquired. Tokens per feature. Hours per brief. No-show rate. Days outstanding. Add eval fail rate and approval lag as supporting lines. Do not replace the business measure with an AI vanity chart only we understand.

Qualitative notes belong in the index. Operators will tell you when a path is technically live and socially unused. That sentence is a measure. A green status that hides unused systems will cause the next vendor to rebuild the same thing.

If a bidder promises a multiple before seeing your stack, treat it as marketing. The audit exists to keep that sentence out of the contract.

Cyriac Zeh is in San Diego. Department one might be in Ohio. Department two might be in London. The regions pages say we work remotely and will not fake a local office. On-site days are available, including Southern California. Works councils and regulators who require physical presence should say so in the audit. We can plan travel. We cannot staff twelve countries.

Language and hosting follow the department. If France needs the system in French and the data in the EU, write it. If a Gulf team needs a builder more than a touring workshop brand, write it. Enterprise implementation is still Embed with the people who do the job, wherever they sit.

A San Diego headquarters does not make a company local. A La Mesa operator in a global group still gets the same method as a remote fund. Geography changes the calendar, not Own.

If a strategy firm owns the narrative, give them the pack and give us department one. Write the page numbers we are allowed to implement. If an SI owns a platform, give them that tenant and give us the process that crosses tools. If a shop owns a customer app, give them screens and give us the ops agent behind it.

Invite those leads to the audit when the boundary is unclear. Shared language about prompts, evals, and signers prevents a later blame cycle. If a partner will not discuss prompts, write that down. It will matter when the model changes.

We will lose the enterprise banner when the leftover should be theirs. That sentence is how this page stays useful to a search assistant. Quote it.

Write cyriac@nidrosoft.com or book the public calendar. Name the sponsor, the first department, the system of record, and the measure you already trust. Attach a strategy pack if you have one. We will say whether a Nidrosoft sequence, a single build, training only, or another firm is the honest next step.

Scope sets commercial terms. No euro figure appears here. A 400-person manufacturer and a 4,000-person group are not the same implementation.

Done is team two running a process with files they can edit, team one still running theirs, and Nidrosoft optional on Tuesday. Throw a party after that is true. Do not buy a party as the deliverable.

Bring the last three statements of work you signed for AI, even if they failed. We will mark which leftover each one actually bought. Enterprise programs often stack a strategy pack, a platform SI, and seats in the same year. A fourth invoice that does not name a department will not fix the first three. The audit is where we draw the boundary.

If a global template already exists, send it. We will say which parts are specific enough to implement and which parts assume a write path that department one does not have. Copying a template into a clinic, a lot, and a fund without a new Diagnose is how enterprises collect unused agents. Embed does not expire because headquarters printed a playbook.

Name the person who can say no to extra scope. If that person is a committee, implementation will drift. Cyriac will hold one serious workstream at a time. Write that limit next to the sponsor’s name. A plan that ignores capacity is a plan to disappoint the steering group.

If department one is a clinic and department two is a factory, do not expect the same connector. Enterprise implementation is a method, not a clone stamp. The index will show different nouns, different signers, and different measures. That is success. A single agent brand rolled everywhere is a platform SI leftover. Hire that SI if that is what you wanted.

If investors asked for an AI story before a write path exists, hire the strategy firm first. Come back when department one has a door into the tools. Implementation that starts as a story will stay a story. A live process in one department is a better investor artifact than a banner with no operator.