Turbo AI PM
Tools / decision aid

Does this actually need AI?

Someone has asked for AI. Before you scope a model, check the four things that decide whether AI is the right tool here — and what to build instead when it isn't. Then match the problem to the capability that fits it.

Needsnothing — four questions about the problem you have been handed
Givesa verdict (including "not AI yet" and "data first") and the capability that fits

Four questions

Answer honestly. A "no" is not a failure — it usually points at something cheaper that works.

Is there a pattern in past examples that predicts the outcome?

AI learns from examples. If every case is genuinely one-off, or the answer follows a rule someone could write down, there is nothing to learn.

Can you get the data at decision time, not just in a warehouse?

A signal that exists in a monthly export but not at the moment of the decision cannot be used. Permissions, freshness, and latency all count.

What does one wrong answer cost?

This sets whether you need a human in the loop, not whether to build. Cheap and reversible means you can ship early and learn.

Would a simple rule or heuristic get most of the value?

If ten if-statements get 80% of the benefit, that is the better product: cheaper, instant, explainable, and it becomes your baseline later.

Match the need to the capability

If AI does fit, this is the shortest route from what the user needs to the workflow that covers it. The capability you pick decides the metrics, the failure modes, and the kit.

What the user needsLooks likeCapabilityWhere to go next
"Predict a number"Forecast, price, risk score, demandRegression / forecastingError metrics: MAE vs. RMSE vs. MAPE
"Sort this into buckets"Route a ticket, flag fraud, moderate contentClassificationClassifier / triage kit
"Put the best options first"Feed, carousel, "you might also like"Ranking / recommendationRecommendations scenario
"Find the right information"Site search, catalog search, knowledge lookupRetrieval / searchSearch & Discovery scenario · RAG kit
"Answer questions from our docs"Internal knowledge assistant, help centreRetrieval + generation (RAG)RAG kit
"Pull fields out of documents"Invoices, forms, contractsExtractionWorkflow automation scenario · Classifier kit
"Draft or rewrite something"Emails, summaries, code, marketing copyGenerationCopilot kit · Coding assistant kit
"Talk to our customers"Support chat, self-serviceConversational + retrievalCustomer support scenario · Support kit
"Do the whole task for me"Multi-step work across toolsAgent / pipelineAgents & pipelines kit
"Pick the next best action"Offer, message, route, or do nothingDecisioning / upliftDecisioning scenario
If none of these fit, that is a useful result too: the problem may be a rules problem, a data problem, or a workflow problem wearing an AI costume. Write down the rule and ship it.