BPAI / RESOURCES
Better questions. Better AI.
Practical guides for people commissioning, building and using AI.
Help shape BPAIBUYER GUIDE / 01Before you hire an AI practitioner
- What problem will the work solve and how will success be measured?
- Can the practitioner demonstrate relevant work and their own contribution?
- What data is used, where does it go and who can access it?
- What are the limitations and where is human review needed?
- What are the implementation, licence, support and ongoing costs?
- Who owns the work, what documentation will you receive and who supports it after handover?
- How are problems reported, changes tested, and the service switched off or replaced?
PRACTITIONER GUIDE / 02Show the value of your work
- Describe the problem and how things worked before the project.
- State your role and the scope of your contribution.
- Separate what was implemented from what remained experimental.
- Record testing, data controls and human oversight.
- Support outcomes with evidence, dates and limitations.
- Obtain client permission or anonymise appropriately.
PLAIN ENGLISH / 03AI in plain English
A few useful terms, without the jargon. Understanding the language helps you ask better questions and make informed decisions.
- Generative AI
- AI that produces content such as text, images or code. Its output can be useful, but accuracy and suitability still need checking.
- AI agent
- A system that uses AI to work through tasks and may use tools or take actions. Its permissions, limits and human oversight need to be clear.
- Automation
- Using software to carry out defined steps. Not every automation needs AI, and a simpler approach may be the better choice.
- Model
- The underlying system used to generate outputs or make predictions. Different models suit different tasks and have different limitations.
- Hallucination
- An AI output that sounds convincing but is incorrect or unsupported. Confident wording is not evidence of accuracy.
- Human oversight
- People checking, approving or intervening in an AI-supported process, with clear responsibility for decisions and outcomes.
- Evaluation
- Testing a system against agreed criteria, including useful results, errors, risks and cost, before and after deployment.
- Prototype and production
- A prototype tests an idea. A production system is used in real work and needs appropriate security, monitoring, ownership and support.
PROJECT GUIDE / 04Your first AI project
Start with a useful business problem, not a tool. This six-step guide can help you shape a manageable first project.
- Define the problemChoose one task worth improving. Describe who does it, what is difficult and what a better outcome would look like.
- Measure your starting pointRecord today's time, cost, quality or error rate. Agree how to judge improvement, including running costs.
- Check the data and risksIdentify the information involved, who may use it and where it may go. Decide what must remain private and which decisions need human approval.
- Run a small pilotStart small with suitable test data and clear limits. Compare AI with a simpler non-AI approach where appropriate.
- Evaluate before expandingTest realistic cases and failure scenarios. Review results, reliability, user experience and risks against the original criteria.
- Agree ownership and supportClarify who monitors it, fixes problems and approves changes. Keep a backup way of working and review whether the solution continues to deliver value.
General planning guidance, not an assessment or endorsement.
Use the buyer checklist