Free, no sign-up to start: the plan opens as a live, editable Gantt chart. 17 tasks across 5 phases, about 15 weeks from start to finish with the parallel work shown.
What's in the template
| Phase | Task | Duration | Starts | Depends on |
|---|---|---|---|---|
| Frame | Problem framing and success metric | 5 days | Day 1 | Nothing (starts first) |
| Frame | Data inventory and access requests | 15 days | Day 6 | Problem framing and success metric |
| Frame | Privacy and governance review | 8 days | Day 6 | Problem framing and success metric |
| Data | Data extraction and pipeline | 10 days | Day 21 | Data inventory and access requests, Privacy and governance review |
| Data | Labelling guidelines | 4 days | Day 6 | Problem framing and success metric |
| Data | Labelling and quality checks | 15 days | Day 31 | Labelling guidelines, Data extraction and pipeline |
| Data | Exploratory analysis | 6 days | Day 31 | Data extraction and pipeline |
| Model | Baseline model | 5 days | Day 37 | Exploratory analysis |
| Model | Feature engineering and experiments | 20 days | Day 46 | Baseline model, Labelling and quality checks |
| Model | Model selection | 3 days | Day 66 | Feature engineering and experiments |
| Evaluate | Offline evaluation and error analysis | 6 days | Day 69 | Model selection |
| Evaluate | Bias and robustness checks | 5 days | Day 69 | Model selection |
| Evaluate | Stakeholder review of results | 2 days | Day 75 | Offline evaluation and error analysis, Bias and robustness checks |
| Deploy | Serving and integration | 10 days | Day 77 | Stakeholder review of results |
| Deploy | Shadow or A/B test | 14 days | Day 87 | Serving and integration |
| Deploy | Monitoring and retraining plan | 5 days | Day 87 | Serving and integration |
| Deploy | Full rollout | 2 days | Day 101 | Shadow or A/B test, Monitoring and retraining plan |
Durations are examples to replace with your own estimates, not benchmarks. Some work runs in parallel, so the plan takes 102 days instead of the 135 it would take one task at a time. The 11 darker bars are the critical path: a day lost on any of them moves the finish date.
About this machine learning project
Machine learning projects are rarely late because of modelling. They are late because data access took six weeks, labels were inconsistent, or nobody planned how the model would be deployed and monitored. This template front-loads data access and labelling, builds a simple baseline early, and treats deployment and monitoring as real work.
It suits a supervised learning project in a business setting, such as classification, forecasting or ranking, run by a small team of data scientists and engineers over about four months. If you are using an existing model through an API rather than training your own, shorten the data and experiment phases and keep evaluation, integration and monitoring.
The structure follows the common pattern of framing, data, baseline, iteration, evaluation, deployment and monitoring.
Who it's for: Data science leads, ML engineers and product managers planning a model-backed feature.
How to use it
- Use this template. Click Use this template to open the project with framing, data, model, evaluation and deployment as lanes and every link already drawn.
- Start access requests now. Set the data access task to your organisation's real lead time. It is often the longest single task in the plan.
- Time-box the experiments. The experiment phase is a fixed box. Decide in advance what result is good enough to move on, and write it in the task description.
- Share with stakeholders. Share a read-only link so product and business owners can see when results will be ready for review.
Data first, baseline early
In this plan the critical path runs through data access, the pipeline, labelling and the experiments. A simple baseline model is built as soon as exploratory analysis is done, before labelling is finished. The baseline tells you early whether the problem is learnable and gives the experiments something to beat.
The privacy and governance review runs beside the access requests rather than after them. Both usually involve other teams, and a late governance finding can invalidate data you have already prepared.
Deployment is not the end
Serving and integration, a shadow or A/B test, and a monitoring and retraining plan all come before full rollout. A model that is accurate offline can still fail on live data, and drift starts the day it is deployed. Keeping these tasks on the plan makes the true finish date visible from the start.
Planning tips
- Write the success metric in business terms as well as model terms.
- Version your datasets and record which version each experiment used.
- Put error analysis on the plan; it is where most of the next improvement comes from.
- Agree who owns the model after launch before you launch it.
Frequently asked questions
- What are the phases of a machine learning project?
- Problem framing, data collection and preparation, baseline and experiments, evaluation, deployment and monitoring. This template includes all of them with realistic dependencies.
- How do I plan experiments that might fail?
- Time-box them. The experiment task has a fixed duration, and the model selection task after it is where you decide whether the result is good enough or the scope needs to change.
- Can I use this for a data science project without deployment?
- Yes. Delete the deploy lane and end the plan at the stakeholder review of results.
- Can I see who is doing what?
- Yes. Each task has an owner field, and the resources view shows each person's workload.
Related templates and guides
- Software Development Project Gantt chart template
- Data Migration Plan Gantt chart template
- API Integration Project Gantt chart template
- IT Project Plan Gantt chart template
- Sprint planning on a Gantt chart
- Microsoft Project alternative
- Free Gantt chart maker, no sign-up
- Critical path calculator
- Free timeline templates
- All Gantt chart templates
- Gantt chart examples
- How to make a Gantt chart
- Excel Gantt chart template
The free plan saves one project. Downloads from the editor (PDF, CSV) and more projects are on Solo, $12/month.