The right starting point depends on the job, the ownership you need, and who will keep the system healthy after launch.
01 · The governing principle
Start with the smallest useful solution.
More technology does not automatically create more value. Begin with the least complex system that can reliably improve one business result, then earn the right to add more.
Our rule
Solve the business problem with the lightest system that can be trusted in real work.
01AutomateKnown steps
02AssistVariable work
03ProductizeMultiple users
04PrivatizeControl required
Each step adds capability—but also cost, testing, maintenance, and operational responsibility. Private infrastructure is a requirement to justify, not a status symbol.
02 · Six implementation levels
Buy a defined outcome, not “AI consulting.”
These offers cover the path from one decision to a complete private platform. A larger number means more scope and responsibility—not necessarily the right place to begin.
01Start
One clear question
AI Decision Session
Get an experienced answer, a concrete recommendation, and a realistic next step before committing to a larger project.
$500fixed · delivered in 2 days
Best forA founder or leader deciding what to build, buy, change, or stop.
You receive
Short pre-call questionnaire
60-minute working session
Recommended tools and delivery approach
Approximate budget, risks, and next step
One-page decision memo within 48 hours
Scope boundary: Advice and a written recommendation; implementation is separate.
Add logs, quality checks, alerts, access controls, backups, and approval paths appropriate to the risk.
LangfuseOpenTelemetrySentryGrafana
Technology names and logos are trademarks of their respective owners. They identify technologies discussed in this guide and do not imply partnership, endorsement, certification, sponsorship, or affiliation. The final stack is chosen after requirements and license terms are reviewed.
06 · After launch
Decide who keeps it working.
Every production system needs an owner. Choose a clean handover, ongoing care, or an embedded partner before launch—not after the first credential expires.
01
Build and hand over
Included
Receive the project files, configuration, documentation, training, and a clean operating handover.
02
AI Care Plan
From $1,000/month
Monitoring, updates, incident support, and a small allowance for maintenance and minor improvements.
03
Embedded AI Partner
$8k–$15k/month
Ongoing strategy, implementation, experimentation, and improvement with a three-month minimum.
07 · Total cost
The build fee is not the whole operating cost.
A credible proposal separates professional services from provider bills and ongoing responsibility. You should know which costs are fixed, usage-based, and optional before work begins.
01
Definition and build
Discovery, design, implementation, testing, documentation, training, and handover—the professional work covered by our project fee.
02
Model usage
Charges from a managed AI provider, or the GPU time required to run an open-weight model. Cost follows volume, model, and response size.
03
Hosting and data
Application hosting, databases, file storage, backups, network traffic, and any dedicated compute the system needs.
04
Connected software
Licenses for automation platforms, monitoring, communication tools, or business systems that remain third-party products.
05
Ongoing operations
Monitoring failures, renewing credentials, applying updates, restoring backups, and responding when an external service changes.
06
Continuous improvement
Reviewing real usage, correcting weak outputs, adding capabilities, and keeping the system aligned with how the business changes.
Our commercial rule
Our fee, third-party subscriptions, model usage, hosting, hardware, licenses, and ongoing support are shown separately. Where practical, provider accounts are opened in your company’s name and paid directly by you.
08 · Private AI and model adaptation
Self-host because you need to—not because you can.
Private infrastructure can be the right answer. It also transfers more uptime, security, update, and capacity responsibility to your organization.
Self-hosting is worth assessing when
Data or contract terms require a controlled environment.
A dedicated network boundary or hardware is mandatory.
Usage is large and predictable enough to evaluate dedicated compute.
A technical owner can operate, secure, monitor, and recover the system.
Managed services are usually better when
Speed to value matters more than infrastructure control.
Usage is early, low, variable, or difficult to forecast.
You want strong model quality without running GPU operations.
Your team does not have an infrastructure owner after handover.
Fine-tuning decision
Fine-tuning is step four, not step one.
It can improve a narrow, repeated behavior once the gap is measurable. It does not repair unclear processes, missing knowledge, weak examples, or absent quality checks.
01Define the task and quality measure
02Test prompting, workflow, and trusted knowledge
03Add tools, examples, and repeatable evaluations
04Fine-tune only if a stable gap remains
Important: Private hosting and fine-tuning are independent decisions. You can privately host a standard open-weight model, or fine-tune a model that is served by a managed provider.
09 · Realistic examples
Four project shapes, with honest ranges.
These examples are planning anchors, not quotes. Final price depends on scope, data readiness, integrations, security, deployment, and the level of operational handover required.
01$2.5k–$5k
Lead routing automation
Qualify inbound leads, update the CRM, notify the right owner, and draft the first follow-up.
Recommended offer
Workflow Automation Sprint
Where it runs
Managed services
Typical timeline
1–2 weeks
02$10k–$20k
Company knowledge assistant
Answer employee questions from trusted sources and route uncertain or sensitive questions to a person.
Recommended offer
AI Assistant Sprint
Where it runs
Managed or client-owned cloud
Typical timeline
Up to 30 days
03$25k–$75k+
Customer-facing AI product
Launch a branded application with users, billing, data, an AI workflow, and an administration layer.
Recommended offer
Custom AI Product
Where it runs
Client-owned cloud
Typical timeline
4–8+ weeks
04$40k–$120k+
Private document intelligence platform
Process sensitive documents with dedicated infrastructure, controlled access, monitoring, and a private model option.
Recommended offer
Private AI Platform
Where it runs
Private VPS, GPU, or cloud environment
Typical timeline
6–12+ weeks
10 · Buyer checklist
Find your likely starting point in six questions.
Use the first statement that clearly describes your situation. You do not need to understand models, hosting, or software architecture before the first conversation.
01
Do you have one strategic question but no build brief?
Start with an AI Decision Session.
02
Do you want to use AI, but do not know where value is hiding?
Start with an AI Opportunity Audit.
03
Can you describe one repetitive workflow from start to finish?
Choose a Workflow Automation Sprint.
04
Does one focused job require language, context, or judgment?
Choose an AI Assistant Sprint.
05
Will the solution need multiple users, workflows, or permissions?
Plan a Custom AI Product.
06
Must the infrastructure or model remain under your control?
Assess a Private AI Platform—with an operations owner identified.
Your next step
Bring one business problem. Leave with a clearer build decision.
In a free 30-minute call, we will understand the situation, identify the likely implementation level, and tell you what we would do next.
