The decision before "which consultancy" is usually "do we build, buy, hire, or partner." Each path has a real shape. The table below names what differs across time-to-production, ownership, lock-in, and how change is handled. No invented score-card; just the factual axes a CTO actually weighs.
| Dimension | Off-the-Shelf AI Tool Generic SaaS, no customisation, lowest control. | Traditional Dev Agency External team, headcount-priced, sequential reviews. | In-House ML Team Full ownership, slowest to start, highest fixed cost. | remilink Architect-led, AI-native delivery, client-owned artifacts. |
|---|---|---|---|---|
| Time to Production | 1–4 weeks | 4–6 months | 6–12 months | 3–5 months |
| Ownership | Vendor-owned | Agency-dependent | Full ownership | Client-owned |
| Lock-in | Very high | High | None | None |
| Human Oversight | Minimal | End-of-sprint only | Team-dependent | Every gate |
| Accuracy Guarantee | Vendor SLA | None | Internal targets | Gate-validated |
| Change Management | Vendor roadmap | Change orders | Agile / ad-hoc | Milestone-based |
| Best fit | Standard problems with no domain edge. Sales pipelines, generic chatbots, vanilla document Q&A. | Software projects with no AI specificity, where headcount-driven delivery is acceptable. | Companies with AI as a core competency, willing to absorb 6–12 months of hiring and ramp. | Mid-market initiatives that need production AI in months, not quarters, with the option to take ownership in-house at handoff. |
Right answer when the problem is generic enough that a packaged tool already covers it. Use the tool.
Sounds simple until you actually price it. You need more than headcount:
Most teams underestimate all four.
Solves all four. You get a team that already exists, with cross-disciplinary expertise across LLMs, RAG, agents, computer vision, MLOps, and data engineering.
When your problem touches more than one of those areas, two or three of our architects and engineers review it together. No extra line on the proposal.
If your ML problem is genuinely small enough that an architect-led engagement is overkill, we will say so on the feasibility call. The point of this table is to make the choice easier, not to market against the alternatives.
The feasibility call is free. Two hours with a senior AI architect. We will tell you which path fits your project, not just argue for ours.