# AI project planning: realistic timelines [Skip to content](#lm-inhoud)Network/NL[EN](/en/)[Hubhub.llmnet.nlCompare models by task, language, cost and license.](https://hub.llmnet.nl/en/)[Communitycommunity.llmnet.nlPrompt techniques, patterns and system prompts.](https://community.llmnet.nl/en/)[APIapi.llmnet.nlLLMs robust in software: rate limits, routing, structured output.](https://api.llmnet.nl/en/)[Consultancyconsultancy.llmnet.nlIntroducing AI in an organization, from pilot to production.](https://consultancy.llmnet.nl/en/)[Newsnieuws.llmnet.nlDevelopments in AI, interpreted for the Netherlands.](https://nieuws.llmnet.nl/en/)[Benchmarkbenchmark.llmnet.nlMeasure AI quality yourself, for your own tasks.](https://benchmark.llmnet.nl/en/)[Jobsvacatures.llmnet.nlAI roles, salaries and career paths in the Netherlands.](https://vacatures.llmnet.nl/en/)[Learnleren.llmnet.nlAI concepts in plain language, from beginner to builder.](https://leren.llmnet.nl/en/)[Guidegids.llmnet.nlRun AI privately on your own Mac, PC, NAS or home server.](https://gids.llmnet.nl/en/)[Directorydirectory.llmnet.nlMapping the AI ecosystem: tools, models, companies.](https://directory.llmnet.nl/en/)[Radarradar.llmnet.nlSignals from X, research and communities for indie developers.](https://radar.llmnet.nl/en/)[Appsapps.llmnet.nlReviews of AI apps and open-source repos, with tips for people who build their own.](https://apps.llmnet.nl/en/)[llmnet.nl — main site](https://llmnet.nl/)[](https://x.com/intent/post?url=https%3A%2F%2Fconsultancy.llmnet.nl%2Fai-projectplanning-realistische-doorlooptijden&text=AI-projectplanning%3A%20Realistische%20doorlooptijden)[](https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fconsultancy.llmnet.nl%2Fai-projectplanning-realistische-doorlooptijden)[](https://www.reddit.com/submit?url=https%3A%2F%2Fconsultancy.llmnet.nl%2Fai-projectplanning-realistische-doorlooptijden&title=AI-projectplanning%3A%20Realistische%20doorlooptijden)[](#)[](https://x.com/intent/post?url=https%3A%2F%2Fconsultancy.llmnet.nl%2Fai-projectplanning-realistische-doorlooptijden&text=AI-projectplanning%3A%20Realistische%20doorlooptijden)[](https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fconsultancy.llmnet.nl%2Fai-projectplanning-realistische-doorlooptijden)[](https://www.reddit.com/submit?url=https%3A%2F%2Fconsultancy.llmnet.nl%2Fai-projectplanning-realistische-doorlooptijden&title=AI-projectplanning%3A%20Realistische%20doorlooptijden)[](#) # Planning realistic timelines for AI projects Published on consultancy.llmnet.nl | Category: Project approach Planning an IT project naturally involves uncertainty, but when implementing artificial intelligence (AI) and Large Language Models (LLMs), timelines consistently run longer than anticipated. Where traditional software development largely revolves around building predefined functionality, an AI initiative is characterized by experimentation, iteration, and strong dependence on external conditions. In this article, we break down the time factor in B2B AI projects. Rather than looking at the estimated development hours for an algorithm, we look at the actual timeline the organization goes through. The focus is deliberately on the phases and organizational friction that initial plans almost always underestimate. ## Why the timeline of AI projects deviates A common pitfall when drawing up a project plan is assuming that technical delivery marks the end of the project. An AI model that works well locally or in a test environment often represents only a fraction of the total timeline. The real delays arise not from writing the code itself, but from the interaction between the technology and the existing organization. Gathering the right data, obtaining legal sign-off, and adapting employees' way of working take far longer in practice than building the technical integration. ## The eight structurally underestimated phases To arrive at a realistic timeline, the entire process must be broken down into phases. The time indications below should be regarded as a general rule of thumb (not an empirical measurement or fixed standard) and depend heavily on the size of the organization and the complexity of the domain. ### 1. Arranging access to data and systems Obtaining API keys, database access, and the necessary authorizations within an enterprise environment is rarely a matter of hours. Internal tickets often need to be opened, security teams have to give approval, and service level agreements (SLAs) with internal administrators need to be worked out. Rule of thumb for timeline: 2 to 6 weeks. ### 2. Cleaning up and structuring data quality A model is only as good as the data it is fed. Cleaning up unstructured documents, correcting incomplete metadata, and consolidating fragmented knowledge sources require intensive manual work and coordination. For a deeper look at this step, see the article on [data quality for AI](https://consultancy.llmnet.nl/en/datakwaliteit-voor-ai) is worth consulting. Rule of thumb for timeline: 3 to 8 weeks. ### 3. Security and privacy review (DPIA) When company-specific data is processed by an AI model, data protection and privacy are of critical importance. Carrying out a Data Protection Impact Assessment (DPIA) and having it reviewed by a Data Protection Officer (DPO) or Security Board significantly delays the start of actual development. See the guide on the [AI risk analysis and DPIA](https://consultancy.llmnet.nl/en/ai-risicoanalyse-dpia). Rule of thumb for timeline: 4 to 10 weeks. ### 4. Procurement and contracting Signing SaaS contracts, data processing agreements (DPAs), or purchasing dedicated compute capacity requires involvement from the legal and procurement departments. Negotiations over liability, data locations, and IP rights take a lot of time. Rule of thumb for timeline: 3 to 8 weeks. ### 5. Setting up and validating evaluation How do you know whether the AI's output is correct, safe, and useful? Building a robust evaluation set (golden dataset) with manually verified examples requires constant coordination with subject-matter experts. See also the section on [evaluating AI systems yourself](https://benchmark.llmnet.nl/en/zelf-evalueren-raamwerk) on the benchmarking platform. Rule of thumb for timeline: 2 to 5 weeks. ### 6. Integration into existing processes and software The AI model needs to land in employees' daily workflow, for example within the CRM or ERP system. This calls for custom API integrations, setting up a management environment, and monitoring. You can read more about technical monitoring at [observability and logging](https://api.llmnet.nl/en/observability-en-logging) on the API platform. Rule of thumb for timeline: 4 to 8 weeks. ### 7. User training and change management End users need to learn to trust the system while also understanding its limits. Organizing workshops, drafting prompt guidelines, and addressing resistance call for a phased rollout. Rule of thumb for timeline: 2 to 6 weeks. ### 8. Management after delivery and model maintenance The project doesn't end at go-live. Models require continuous monitoring for drift, changing source documents, and user feedback. Transferring ownership from the project team to a standing management organization takes time. Rule of thumb for timeline: 2 to 4 weeks. Project phase | Indicative timeline (rule of thumb) | Primary delay factor | 1. Access & authorizations | 2 - 6 weeks | Internal IT tickets, permissions structure | 2. Data quality & cleanup | 3 - 8 weeks | Missing metadata, messy data sources | 3. DPIA & security review | 4 - 10 weeks | Legal review, privacy frameworks | 4. Procurement & contracting | 3 - 8 weeks | Negotiations over the DPA and liability | 5. Setting up the evaluation set | 2 - 5 weeks | Availability of subject-matter experts | 6. System integration | 4 - 8 weeks | Legacy APIs, release cycles | 7. User training | 2 - 6 weeks | The organization's capacity for change | 8. Handover to management | 2 - 4 weeks | Setting up monitoring and support | ## Organizational delay factors (non-technical) Experience shows that in more than 70% of cases, the real delays are attributable to the organization rather than the technology used. A tightly written algorithm grinds to a halt if the organization doesn't move along with it. - Waiting for approvals: Many projects depend on steering committees that meet only once a month. A single missed decision round immediately pushes the entire schedule back by four weeks. - Availability of subject-matter experts: The specialists who need to assess and validate the AI's output often already have a full workweek of regular operational tasks. Their scarce time quickly becomes a bottleneck. - Change freeze periods: Many organizations have strict periods during which no changes may be made to production systems (for example, around the holidays or at the end of a quarter). - Holiday periods: Summer and winter holidays cause delays that go beyond employees simply being absent; resuming decision-making often takes twice as long. ## Managing by decision points and time ranges Using a strict, linear schedule with a fixed end date is almost guaranteed to lead to disappointment in AI projects. A more realistic alternative is to plan around decision points (go/no-go phases) and use a time range instead of a hard delivery date. Instead of promising that a system will go live on October 1, project leadership might instead communicate a time range: "Go-live will take place between week 40 and week 44, depending on DPIA approval in phase 3." Assumption: We assume that management is willing to steer based on phased approval, rather than sticking to a rigid, predetermined end date. After each phase, the project is evaluated. If it turns out that data quality is insufficient to reach the required accuracy, the option is used to fix the data first before the integration phase starts. You can read more about this phased approach in the article on the [AI pilot in 30 days](https://consultancy.llmnet.nl/en/ai-pilot-30-dagen) and the detailed explanation of the transition from [pilot to production](https://consultancy.llmnet.nl/en/pilot-naar-productie). ## Signs that a schedule is failing A project schedule rarely grinds to a complete halt overnight; it's a gradual process. There are several early warning signs that indicate the planned timeline is no longer realistic: - Unanswered questions at the legal department: Has a draft agreement or DPIA been sitting with Security or Legal for more than two weeks without substantive feedback? Then delay is unavoidable. - Subject-matter experts exceeding the feedback deadline: Are the subject-matter experts unable to validate the model outputs within the agreed deadline? This is the first indicator of stagnation in phase 5. - Assumptions about data availability turn out to be wrong: As soon as it turns out that data needs to be manually extracted from PDF files instead of being available through a directly accessible API, the original time estimate no longer holds. For properly scoping the project, consult the guide for [AI project scoping](https://consultancy.llmnet.nl/en/ai-project-scoping). - Shifting test environments: If setting up the test or acceptance environment is repeatedly postponed due to IT management priorities, the integration phase comes under pressure. ## Checklist: Questions to ask beforehand for a realistic schedule Ask the questions below before setting the final timeline for the AI project. A negative answer immediately indicates where extra time needs to be built in. - [ ] Is there already an approved data processing agreement in place for the intended AI vendor or cloud platform? - [ ] Is the Privacy Officer / DPO involved, and has time been reserved in their quarterly schedule for a DPIA? - [ ] Is the required data directly accessible through a documented API, or do exports still need to be created? - [ ] Are the subject-matter experts formally allocated a fixed number of hours per week to validate the system? - [ ] Is it known when the planned release windows and any change freezes of the surrounding IT systems will take place? - [ ] Is the decision-making body (steering committee) set up to make decisions based on phased decision points? By taking this organizational reality into account when drawing up the [AI implementation plan](https://consultancy.llmnet.nl/en/ai-implementatieplan-mkb) , you avoid surprises and build a predictable, manageable process. By Ivo Donker - compiled with AI assistance (Claude & Gemini) - Last updated: 2 August 2026