Short answer: choose a partner that starts with your business problem rather than a technology, can show software it has shipped and supported, will ground AI in your own data with human review where it matters, and is clear about ownership, cost and what happens after launch.
AI has made it easy to build an impressive demo. It has not made it easy to build something your team relies on every day. The partner you choose makes most of the difference.
1. Do they start with the problem or the technology?
A good partner asks how your work runs today before recommending a tool. Expect questions about your workflows, systems, data, volumes and the decisions people make.
Watch for: a proposal that names a model or platform before anyone has looked at how your business operates.
2. Have they shipped and supported real software?
AI features live inside real products and systems. They need authentication, integrations, testing, monitoring and maintenance like any other software. Ask to see products the partner has built that are in daily use, and ask who supports them now.
Watch for: a portfolio of prototypes and demos with nothing in production.
3. How will they use your data?
Useful business AI works from your own documents, records and policies. Ask:
- Which of our data will the system access, and how is access controlled?
- Will our data be used to train any external model?
- Where is data stored and processed?
Watch for: vague answers, or no mention of access controls and permissions.
4. Where do people stay in the loop?
Every AI system makes mistakes some of the time. A responsible partner designs for that: limits on what the AI can do, human approval for important actions, escalation when the system is unsure, and testing on real cases before launch.
Watch for: promises of full automation with no review step, or claims that the system will never be wrong.
5. What happens after launch?
AI systems change after they go live. Your data and products evolve, and model providers update their models. Ask how the partner monitors quality and cost, how often they review the system, and what ongoing support looks like.
Watch for: a plan that ends at launch day.
6. Who owns the work, and how is it priced?
Ownership of code, models, prompts and data should be agreed in writing before work starts. Pricing should be clear about what is fixed, what is variable, and what ongoing costs, such as model usage, you will carry.
Watch for: unclear ownership terms, or estimates that leave out running costs.
7. Will they tell you where AI is not the answer?
Sometimes a workflow is better served by simple automation, a better form, or a process change. A partner worth hiring will say so.
A practical way to start
Rather than committing to a large project, start with a short, well-defined first step: a strategy engagement that maps your opportunities, or a single workflow with a clear success measure. It shows you how the partner works before you expand.
Birbals has designed and built software since 2003. Our AI Strategy & Roadmap is designed to be that first step, and you can see products we have built and still support on our Work page.