AI EngineeringAgents & Agentic Workflows
Agent Opportunity Blueprint
A ranked map of which of the client’s workflows in one function are worth turning into agents, each scored for automation candidacy, with ROI models for the top three built on the client’s own figures, so they can choose what to build first and know the likely cost and return before they spend on it.
Agent Opportunity Blueprint
$8K–$15K
- Duration: 1 week
- Tier: Diagnose
- Format: Fixed scope
You leave withWorkflow-to-agent map, automation candidacy scores and top-three ROI models
Book a Free SessionSound familiar?
Our team does the same multi-step job hundreds of times a month across three or four systems. Someone said an agent could do it and I have seen the demos, but I cannot tell which of our workflows would work as an agent and which would be a mess. I also cannot take ‘it will save a lot of time’ to Finance. I want to know which one to try first, what it would cost and what we would get back, before I pay anyone to build anything.
What you leave with
Workflow inventory and maps
Format: PDF and editable diagrams
How we know it’s acceptableEach workflow has steps, systems, exceptions and approval points confirmed by the person who does the work.
Candidacy scoring
Format: Spreadsheet with the rubric
How we know it’s acceptableEvery score has a one-line reason. The workflow owner has agreed or disputed it at the readout, and disputes are recorded.
ROI models for the top three
Format: Spreadsheet
How we know it’s acceptableEvery input is labelled client-confirmed or estimated. Sensitivity is shown. Finance has seen the cost-per-hour assumption.
Findings report and recommendation
Format: PDF, readable by a non-technical executive, plus the readout
How we know it’s acceptableNames one workflow to build first with a scope draft usable as the Agent Prototype Sprint scope, and names what not to build.
Also included
- A workflow-to-agent map: which steps an agent would do, which stay with a person, and where a human must approve.
How it runs
1 week
Day 1
Kickoff and inventory
Confirm the function, the workflows and an owner for each. Collect process documents and volumes. Gate: if documentation or practitioner time is not there, stop and re-scope to two weeks or three workflows.
OutA workflow inventory with data gaps listed.
Days 2 and 3
Walkthroughs
One session per workflow with the person who does it.
OutStep-by-step workflow maps with exceptions, systems and approval points.
Day 4
Score and model
Score each workflow against the rubric, choose the top three, build the ROI models from client figures.
OutScored candidates and draft models.
Day 5
Findings and readout
Client corrects figures at the readout.
OutFindings report, final models, the recommended first build and its scope draft.
Is this right for you?
This is for you if
- Buyerthe person who owns the cost or throughput of the workflows in question: an operations, finance or customer service leader, or a CTO sponsoring an agent initiative.
- Also has to say yesthe people who actually do the work, because the blueprint is only as honest as their description of the exceptions. Whoever owns the systems the workflows touch, who knows which have APIs. Finance, for the cost-per-hour and volume figures the ROI models rest on. Most failed blueprints fail because the practitioners were never in the room and the manager described the happy path.
This is the wrong answer if
- Nobody can describe how the work is done today, or the people who do it are not available. The week becomes process discovery and we absorb it. Map the process first, or start with the AI Opportunity & Roadmap Sprint.
- The workflow is high volume and genuinely deterministic. A script or rules engine is cheaper and easier to audit than an agent, and the honest output is ‘do not build an agent’.
- The client has already chosen one workflow, and has an owner, volumes and examples for it. Choosing is what this programme sells. Go straight to the Agent Prototype Sprint.
- The client wants a view across the whole company, or governance and risk. That is the AI Opportunity & Roadmap Sprint.
- Every candidate workflow needs an agent to take actions nobody will approve or reverse. The blueprint will conclude ‘not yet’, which is a valid result but not what the client is expecting to buy.
What’s included, and what isn’t
In scope
- one function
- up to five workflows
- one walkthrough per workflow of up to two hours with the person who does the work
- candidacy scoring
- ROI models for the top three with an indicative build-cost range
- one findings report
- one 45-minute readout
Out of scope
- building or prototyping any agent (that is the Agent Prototype Sprint)
- a fixed quote for the build
- access to production systems or live data beyond what a screen-share needs
- vendor or platform selection
- a second function
- process redesign beyond agent candidacy
- a governance, security or compliance assessment (that is the AI Opportunity & Roadmap Sprint)
Before we start
Before day one
- A named function and a list of candidate workflows.
- A workflow owner for each, and two hours booked from the person who does the work in days 2 and 3.
- Rough volumes and time per case, or agreement that we estimate and label them.
- A Finance contact reachable during the week for the cost-per-hour figure.
- A way to see the process: documents, or a screen share. Confidentiality terms in place.
- A named decision-maker for the readout, and agreement that the output is a basis for a decision, not a quote.
What we’ll need from your team
- The people who do the work: about two hours per workflow across days 2 and 3, up to ten hours across several people for five workflows.
- The function leader: one hour at kickoff, 45 minutes at the readout, and 30 minutes to correct figures.
- A Finance contact: about 30 minutes.
- A systems contact: 30 to 60 minutes on integrations and APIs.
Pricing
What we need to give you a quote
- The function and the candidate workflows (up to five).
- Rough monthly volume and time per case for each.
- The systems each touches and which have documented APIs.
- Whether process documentation exists, and who does the work today.
- Whether we need access to live systems or a walkthrough on screen share is enough.
- Data sensitivity.
- Which cost-per-hour figure Finance will accept.
- Who attends the readout.
After we hand over
What we measure
- Blueprints delivered
- Share that lead to a paid prototype within 90 days
- Share where the top result was ‘do not build’
- Where a build follows, modelled task time against measured task time
The client measures: the before figures in the blueprint, which are volume, minutes per case and error rate, confirmed by their Finance contact.
What comes next
- Agent Prototype Sprintfor the top-ranked workflow, using the scope draft from the report.
- AI Opportunity & Roadmap Sprintif the readout shows other functions have the same question, or governance becomes the real issue.
- Vertical Agent Launchif a BigStep vertical accelerator matches the workflow.
- Nothing.If no workflow is a good candidate, that is the finding and the report says it.
Questions buyers ask us
Then you may not need the choosing part. If you can name one with an owner, volumes and examples, go straight to the prototype sprint. We will tell you that on the scoping call.
For three to five workflows in one function, it is, provided the process is documented and the people who do the work give us about two hours per workflow. If that is not true, we say so on the scoping call and this is the wrong programme.
This is a fair objection. The models are only as good as the volumes and costs you give us. Every input is labelled client-confirmed or estimated, and we show what has to be true for the return to hold. It is a basis for a go or no-go decision, not a forecast.
For ideas, you can. What takes the week is scoring each workflow against its exceptions, systems and approval points, and producing an ROI case Finance will read.
The document decides whether you spend several times that on a build. ‘Do not build this one’ is a valid and useful result.