Buyer guide

What an AI Discovery for One Manufacturing Workflow Should Deliver

Why a discovery comes first

Many AI projects in manufacturing stall after an impressive demo. The demo used clean sample data; the real workflow depends on views with hidden filters, plant-specific exceptions and rules that live in people's heads. A short discovery exists to surface those dependencies before money is spent on a build.

Questions a discovery should answer

  1. Which decision are we improving, and who owns it?
  2. Where does each input come from, and is that source complete for the question being asked?
  3. What rules and exceptions apply, and who can confirm them?
  4. What is missing or unreliable today, and does it block a pilot or only limit it?
  5. What should a pilot prove, measured how, on which cases?
  6. What will the assistant be allowed to do, and what stays with people?

If a proposal does not commit to answering these, it is probably selling a build rather than a discovery.

Deliverables you should receive

Deliverable What good looks like
Workflow and source map Each step linked to screens, code, views and tables, with open questions marked
Coverage and data-quality notes Population, filters, time windows, units, known gaps
Knowledge or rules model Objects, relationships, rules and owners in a reviewable form
Risk and responsibility notes What the AI may read, suggest and never do
Pilot proposal Scope, architecture options, effort, acceptance cases, stop conditions

Every item should be readable by your team and by another vendor. If the outputs only make sense inside one vendor's platform, you are buying lock-in.

Typical timeline for one workflow

  • Day 1: scoping call, access request, agreement on what a useful result is.
  • Week 1: walkthrough with the owner and experienced users, tracing the workflow through systems, first source map.
  • Week 2: confirm meanings and rules, document gaps, draft the knowledge model and pilot proposal, review session.

Budget

Prices vary by region and vendor size. For a one-workflow discovery, CoWin Future typically quotes US$700–1,400; a focused pilot that follows is typically US$3,500–7,000. Larger consultancies often package multi-workflow discovery programs at much higher fixed fees. What matters more than the number is whether the deliverables are specific and reusable.

Red flags in proposals

  • No named workflow or decision owner, only "AI strategy".
  • A fixed tool or model chosen before the data has been seen.
  • No mention of missing data, evaluation cases or human review.
  • Deliverables described as "insights" or "a roadmap" without concrete artifacts.
  • A request for production write access during discovery.

How to prepare

Bring one workflow, the person responsible for it, a redacted sample or screen recording and a definition of an acceptable result. That is enough to start. If you would like a second opinion on a proposal you already have, or want to scope a discovery with us, describe the workflow here.

Questions & answers

How long should a discovery for one workflow take?

Typically one to two weeks of calendar time. Longer engagements usually mean the scope covers several workflows or is waiting on access and owner time.

What does it usually cost?

It varies widely by vendor and region. At CoWin Future, a one-workflow discovery is typically US$700–1,400, quoted as a fixed price after a short scoping call.

Should a discovery include a working prototype?

Not necessarily. A small throwaway prototype can help test a risky assumption, but the main value is clarity: which data, rules and decisions a pilot depends on, and how it will be judged.

What access does the vendor need?

Usually a walkthrough with the system owner, read-only schema or view definitions, a redacted data sample and time with one or two experienced users. Production credentials are not needed.

LET’S START A CONVERSATION

Let’s start with your workflow.

Tell us what you want to improve and which systems you use. We will reply within one business day with questions, a view on fit and a practical next step.

Discuss a project