AI
Pricing an AI product: connect the bill to value without hiding usage risk
AI products can have variable delivery costs and uneven customer usage. A workable price needs to make both sides of the transaction understandable.
THE AI DESK
AI
Artificial intelligence through the lens of products, economics and accountability.
80 stories to explore
AI
AI evaluations: build a test set around the mistakes your customers would notice
A model can look impressive in a demonstration while failing the tasks that determine whether a customer trusts the product.
AI
RAG freshness: the answer is only as current as the knowledge pipeline
Retrieval can give an AI product access to company information. It also creates an operational responsibility to update, remove and permission that information.
AI
AI agent permissions: give the system the smallest useful set of actions
An agent’s authority should follow the task it has been asked to perform. A capable model does not need unrestricted access to be useful.
AI
Prompt injection: why retrieved text must never become the boss of an AI workflow
Documents and webpages can contain instructions aimed at an AI system. Products need to keep source material separate from the authority to act.
AI
Model routing: optimise the cost of a successful task, not the cheapest response
Using different models for different jobs can improve economics, but retries and escalations must be included in the calculation.
AI
Human review in AI products: put judgement where it can still change the outcome
A review step adds value only when the reviewer has enough context, time and authority to intervene.
AI
AdBanana and the harder test for AI fashion photography
Generating an attractive fashion image is one challenge. Producing a dependable catalogue is another. AdBanana’s product proposition puts that distinction in focus.