What project-based means

In a project engagement, you hire a shop to deliver a defined outcome: a prototype, an integration, a model pipeline, a feature. The scope has a start and an end. The shop owns planning and delivery of that scope.

Browse shops that offer it on the project-based engagement model page.

When it fits

Project work suits problems you can describe clearly:

  • A proof of concept to test whether a use case works with your data.
  • A single feature, such as document search or a support assistant.
  • A migration or integration with a clear finish line.

It struggles when requirements will change weekly or when the product has no natural end. In those cases, look at the dedicated team guide. If you only need extra hands inside your own team, see the AI staff augmentation guide.

Scope AI work differently

AI projects carry uncertainty that ordinary software doesn't. You can specify a screen. You can't fully specify how well a model will perform on your data until you try. Good scoping accounts for this.

Define success as a measurement. "Answers questions about our policies" is vague. "Correct on a test set of real questions, reviewed by your team" is something both sides can check. Agree on the test set early.

Stage the work. A common shape is discovery, a small prototype on real data, then a build phase. Each stage ends with a decision to continue, change course, or stop.

Name the unknowns. Data quality, access to systems, and approval from security or legal are the usual blockers. List them in the statement of work. Our statement of work guide lists what to include.

Set change rules. Decide how new requests are handled and who approves them. Without this, scope drifts or disputes follow.

What to check in a shop

  • Relevant case studies. Same modality and similar constraints to yours.
  • Who builds it. Ask for the names of the engineers, not only the account lead.
  • A real reference call. Speak to a past client directly.
  • How they handle "this won't work." A trustworthy shop tells you when a simpler approach beats a custom model, even if it shrinks the project.
  • Handover. Documentation, code access, and a walkthrough for your team at the end.

Region and size filters help narrow candidates quickly. For a full vetting process, see how to choose an AI dev shop.

After delivery

Models drift and data changes. Ask what happens after the project ends: monitoring, retraining, bug fixes, and who answers questions. If the answer is "nothing," plan internally for it, or consider moving to a longer arrangement.

Checklist

  • [ ] The outcome can be described in one paragraph.
  • [ ] Success is defined as a measurement on a test set.
  • [ ] The work is staged, with a decision point between stages.
  • [ ] Data access and approvals are listed as dependencies.
  • [ ] Change request process is written down.
  • [ ] You have spoken to a past client.
  • [ ] Handover and post-launch support are defined.
  • [ ] You compared shops on the project-based page.