How much does an AI agency cost?
AI agency pricing depends on the work actually purchased: diagnosis, discovery, data preparation, POC, development, integrations, testing, production deployment and maintenance. To compare proposals, separate these cost lines and the work that remains with the client.
The examples below come only from public information recorded in AIPartnerLens provider profiles. They describe different services and are not an average or general market range.
Last editorial review: July 18, 2026.
For procurement planning, compare these cost lines with the overall AI project budget and the AI agency selection checklist.
What an AI agency proposal may include
Every line should state its deliverable, assumptions and client responsibilities. A short phase may be useful on its own or prepare the next stage; it does not mean the solution is production-ready.
| Cost line | Purpose | What to verify |
|---|---|---|
| Diagnosis | Short interviews and review of the current situation to decide whether the topic merits deeper discovery. | Observed scope, people interviewed, diagnostic note and decision expected. |
| Audit | Structured analysis of processes, data, risks, tools and opportunities before choosing a solution. | Method, access required, deliverables, recommendations and independence from implementation. |
| Workshop | Working session to prioritize use cases, align teams or define a first scenario. | Preparation, participants, deliverable, duration and remaining work after the workshop. |
| Discovery | Turn the need into scope, architecture, data requirements, timeline, risks and success criteria. | Assumptions, backlog, responsibilities, estimate, acceptance criteria and production plan. |
| POC | Feasibility test on a limited scenario and dataset to support a decision. | Question tested, data, test set, thresholds, reusable deliverables and stop conditions. |
| MVP | First usable version for a limited group, with essential features and close monitoring. | Users, support, security, instrumentation, technical debt and steps required before scaling. |
| Automation | Workflow that triggers, transforms or passes information between several tools. | Number of scenarios, exceptions, connectors, volumes, error recovery and account ownership. |
| AI agent | System that prepares or executes several actions using tools and control rules. | Allowed autonomy, approvals, audit trail, model costs, tools available and error handling. |
| RAG | Retrieval and generation over a document base with source citations. | Documents, permissions, update frequency, evaluation, citations, hosting and answer quality. |
| Chatbot | Conversational interface for internal users, prospects or customers. | Channels, knowledge base, volumes, human escalation, analytics, security and editorial maintenance. |
| Business application | Custom interface that integrates AI features into a specific process. | Design, development, authentication, roles, integrations, testing, hosting and operations. |
| IT integration | Connection to company applications, APIs, data, identities and security rules. | Systems involved, environments, technical accounts, rate limits, testing and responsibilities. |
| Industrialization | Work required to make a solution reliable, observable and operable beyond a pilot. | Deployment, observability, backups, security, performance, cost, support and rollback. |
| Training | Enable users, administrators or technical teams to adopt and operate the solution. | Audience, objectives, materials, exercises, documentation, assessment and follow-up. |
| Maintenance | Correction, monitoring, updates and changes after production deployment. | SLA, scope, volumes, on-call coverage, fixes, enhancements, quality monitoring and recurring cost. |
Documented examples of agency services
Genee publishes several formats that cover different scopes: a Make scenario, automation program, build & run, MVP and business application. These figures describe only those public offers; they are not a general market range.
Make automation fixed package
Genee
From €3,500 excl. VAT
One complete Make scenario: discovery, production deployment, monitoring and training for one business owner.
Publicly stated delivery time: 7 business days.
Direct public source · reviewed July 18, 2026
Comparability: Comparable only with a focused Make scenario, not with a multi-tool program, AI agent or business application.
Make automation program
Genee
€8,000 to €25,000 excl. VAT
Process audit, 10 to 30 scenarios, custom HTTP connectors, documentation and handover.
Publicly stated duration: 6 to 12 weeks.
Direct public source · reviewed July 18, 2026
Comparability: The number of scenarios, integrations and their criticality must be similar before this amount can be compared with another proposal.
Make build & run
Genee
From €1,800/month
One to three new scenarios per month, maintenance and enhancements, monitoring, SLA during business hours and quarterly coaching.
Public three-month renewable commitment.
Direct public source · reviewed July 18, 2026
Comparability: This is a recurring service and should not be directly compared with a one-off build package.
Focused MVP Factory
Genee
From €8,000 excl. VAT
First testable product version with essential features; the dedicated MVP Factory page publishes the price and delivery time.
Publicly stated delivery time: 2 to 4 weeks.
Direct public source · reviewed July 18, 2026
Comparability: This fast format assumes a simple scope. Features, integrations and expected finish should be confirmed.
MVP on the general services page
Genee
Generally around €15,000
MVP described in the general services FAQ, separate from the MVP Factory entry price.
The general services page states 6 to 10 weeks.
Direct public source · reviewed July 18, 2026
Comparability: The difference from the €8,000 / 2–4 week format reflects different scopes and should be explained in the proposal.
Business application: conflicting public figures
Genee
No single value retained
The two official pages below do not currently publish the same price ranges or delivery times. Each figure remains attributed to its source.
Services page
Complete business application: €30,000 to €150,000
Standard business application: 3 to 6 months
Open Genee services pageHomepage
Custom development from €18,000 excl. VAT; FAQ: complete business application generally €15,000 to €80,000
FAQ: complete business application 1 to 3 months
Open Genee homepageDirect public source · reviewed July 18, 2026
Genee's official sources do not currently show the same ranges. These figures should be clarified before comparing a proposal.
Comparability: Do not merge these ranges or calculate an average. Scope and pricing should be confirmed directly with Genee.
Warning: even for one provider, entry price, delivery time and support level change by format. Compare scope, integrations, acceptance criteria and recurring costs before comparing headline prices.
Benchmarks that should not be confused with an agency proposal
A SaaS subscription buys access to software; a training course buys a learning service. Both may be part of an AI project, but neither covers the same discovery, integration and custom development scope as an agency engagement.
Chatbot SaaS subscription
Botnation AI
From €39 excl. VAT/month
No-code platform subscription; credits, volumes and enterprise plans are separate. Fully delegated creation is quoted separately.
Direct public source · reviewed July 18, 2026
Comparability: A SaaS subscription is not an agency proposal for discovery, integration and custom development.
Open-enrollment AI training
Cegos
€990 excl. VAT
One-day open-enrollment course: “L’intelligence artificielle au service du manager”.
Direct public source · reviewed July 18, 2026
Comparability: An open-enrollment training course is neither an integration project nor custom application or AI-agent development.
These benchmarks are not used to calculate an average AI agency price. They are included only to avoid comparing different categories of spending.
Factors that change the proposal
Ask each agency to state the assumptions used for pricing. This makes it easier to see which part of the cost comes from the business need, existing environment, risk or service level.
Clarity of the need
A documented process and defined criteria reduce discovery time and scope changes.
State of the data
Collection, cleaning, labeling, permissions and updates can represent a major share of the work.
Number of integrations
Every application adds access, constraints, tests, exceptions and dependencies to coordinate.
Risk level
Sensitive data, critical decisions, traceability requirements or high availability increase requirements.
Customization
A configured product is priced differently from a custom application, agent or data pipeline.
Volume and performance
Users, documents, requests and latency influence architecture, model choice and operating cost.
Agency responsibility
Advisory, build, hosting, monitoring, support and service commitments cover different scopes.
Internal handover
Documentation, training and client-team takeover require time but reduce future dependency.
Costs that are often forgotten
A proposal may exclude work that is still necessary for success. Assign every cost line to the agency, client or a third party before approving the budget.
- Business-team time to describe the process, provide examples and validate results.
- Data preparation, cleaning, classification or anonymization.
- API access, application licenses and test environments.
- Model usage, storage, compute, logs and monitoring.
- No-code tools, vector databases, authentication services or connectors.
- Security, legal, procurement and compliance work when the risk level requires it.
- User training, documentation and change management.
- Editorial maintenance of content and regular re-evaluation of answers.
- Support, fixes, enhancements and recovery after an API or model changes.
- Reversibility: export, documentation, account transfer and support for another team.
Compare two proposals without looking only at the total
Reorganize every proposal into the same lines: discovery, data, development, integrations, acceptance, production, training, licenses and maintenance. Add work left to the client. A cheaper proposal may simply transfer substantial effort to the internal team.
Then compare evidence, assigned profiles, timelines, acceptance criteria and assumptions. Ask for low- and high-volume scenarios when usage or data quality remains uncertain.
Questions for normalization
- Are deliverables and acceptance criteria the same?
- Are the same integrations and edge cases included?
- Who prepares data and environments?
- Which tools and volume assumptions drive recurring cost?
- Does maintenance cover fixes, monitoring and change?
Maintenance and operations
Ask for a separate budget for monitoring, fixes, updates, support and enhancements. Model, cloud and software costs should be tied to volume assumptions. Also identify the client-side time required to monitor quality and maintain content.
The agreement should specify covered hours, priority levels, response times and what triggers additional work. A maintenance allowance without scope remains difficult to compare.
Reversibility and exit cost
Changing agency or taking the solution in-house may require exports, code handover, account transfer, documentation and an assistance period. Estimate these items before signing, especially when no-code scenarios or proprietary services are central.
Credible reversibility states data formats, dependencies, usage rights and knowledge to transfer. It protects continuity and makes open and more integrated offers easier to compare.
Budget preparation table
Complete this table before approaching agencies. Better inputs reduce assumptions and make proposals more comparable.
| Area | Information expected | Internal owner |
|---|---|---|
| Business problem | Volume, current time spent, errors, impact and users. | Business / sponsor |
| Initial scope | Included processes, excluded cases and interfaces involved. | Project lead |
| Data | Sources, quality, sensitivity, owners and preparation. | Business / data / security |
| Integrations | CRM, ERP, helpdesk, documents, APIs and environments. | IT |
| Acceptance | Test set, thresholds, critical cases and human validation. | Business / quality |
| Production | Hosting, accounts, logs, backups and support. | IT / operations |
| Recurring costs | Models, cloud, licenses, maintenance and volume assumptions. | Finance / IT |
| Reversibility | Code, accounts, exports, documentation and handover. | Procurement / IT |
Frequently asked questions about AI agency pricing
Can you know an AI agency price before discovery?
An initial budget or entry price can be shared, but a reliable estimate depends on the process, data, integrations, acceptance criteria and expected production level.
Does a cheaper POC reduce risk?
Only if it answers a precise question using representative data. An oversimplified test can postpone integration, security or quality issues instead of reducing them.
Why can two AI proposals have very different prices?
They may cover different responsibilities: data preparation, development, integrations, testing, production deployment, licenses, training or maintenance. Align cost lines before comparing totals.
Fixed price or time-and-materials?
Fixed price fits a sufficiently defined scope. Time-and-materials can suit an exploratory phase when governance, milestones and a budget ceiling are clear.
Which costs recur every month?
Depending on the project: models, cloud, third-party software, connectors, storage, monitoring, support and maintenance. Ask for cost scenarios by volume.
Does AIPartnerLens publish an average AI agency price?
No. The public figures available cover services that are too different to support a useful average or median.
Move from budget to comparable procurement
Use the category page to identify providers, then the selection guide to verify evidence, responsibilities and production conditions. Amounts that are absent from public sources should remain marked as not disclosed.
The audit, POC, MVP or production guide helps verify that a proposal matches the engagement format you actually need.