Most businesses do not need a bigger AI strategy before they start. They need one workflow that is clear enough to build.
That is the point of this field guide. It helps you choose the first business agent by looking at work that already happens inside the firm.
You are not trying to automate the whole operation. You are trying to find one repeated workflow, score it honestly, and write a brief that a builder can act on.
You can download the field guide or use the method below.
Start with the work, not the tool
The weak version of an AI project starts with a product demo.
Someone sees a tool that writes, searches, summarizes, or clicks through a screen. The team then tries to find a problem for it.
The stronger version starts with a workflow:
- What happens every week?
- What waits on the owner?
- What gets copied from one system into another?
- What gets assembled from the same template?
- What creates avoidable back-and-forth?
Those questions produce better agent ideas than asking which model or app to use.
The four first-agent patterns
A good first business agent should fit one of four clean patterns.
Inbound triage
Something arrives and a person reads it, classifies it, routes it, and decides what happens next.
Examples include intake emails, broker submissions, support requests, RFP questions, and new lead enquiries.
The first agent does not need to make the final call. It can summarize, classify, flag missing facts, and draft the next step.
Repeat document assembly
The business creates the same type of document repeatedly, with different facts each time.
Examples include monthly client reports, advisory summaries, engagement letters, meeting notes, and owner statements.
The agent gathers the inputs, follows the structure, and produces a reviewable draft.
Cross-system reconciliation
Two systems hold related facts and a person keeps them aligned.
Examples include CRM to accounting, calendar to billing, project tracker to invoice, spreadsheet to dashboard, and inbox to case management.
The agent spots differences and prepares the fix. A person still approves the update at first.
Owner-routed approval
A small decision waits on the owner because the context is scattered.
Examples include scope changes, fee approvals, follow-up wording, refund decisions, and vendor sign-off.
The agent creates the decision card: facts, options, risks, and a recommended response.
Run the 30-minute sprint
Block 30 minutes and do this with a pen, spreadsheet, or shared document.
Minutes 0 to 5: list the work
Write down 5 to 10 workflows from the last month.
Do not filter too early. You want the real operating mess on the page.
Minutes 5 to 15: match the patterns
For each workflow, write the closest pattern:
- triage
- document assembly
- reconciliation
- approval
- none
If the answer is none, keep it on the list but do not force it into an agent build.
Minutes 15 to 25: score
Score every workflow from 1 to 5 on four dimensions:
| Dimension | What it measures |
|---|---|
| Frequency | How often the work happens |
| Time per run | How much time it consumes, including checking and handoff |
| Owner involvement | How much senior attention it needs |
| Pattern fit | How cleanly it matches one of the four patterns |
The highest score is not always the safest first build, but it usually shows where the business is carrying too much manual work.
Minutes 25 to 30: pick
Pick the highest total unless there is a clear safety, privacy, or customer-risk reason not to start there.
If two workflows tie, pick the one with lower customer-facing risk.
Write the first-build brief
The winning workflow needs a brief before it needs a tool.
Use this structure:
Workflow:
Trigger:
What starts the workflow?
Inputs:
What does the agent need to read?
Rules:
What can the agent decide safely?
Output:
What should the agent produce?
Human review point:
Where does a person approve, edit, or reject the output?
Systems touched:
Which apps, folders, inboxes, databases, or spreadsheets are involved?
Success measure:
How will we know this worked?That brief is the useful artifact. It lets you build internally, hand the job to a technical teammate, or bring the workflow to Enterprise DNA for a scoping conversation.
Choose the build path
Use the score to decide what happens next.
| Total | Read this way | Next step |
|---|---|---|
| 12 or below | Useful, but probably not first | Keep looking |
| 13 to 15 | Good candidate | Run a small version one |
| 16 or above | Strong candidate | Scope properly before building |
The first version should be narrow. It should read known inputs, produce a structured draft or decision card, and stop at a human review point.
Where Enterprise DNA fits
If the winning workflow touches customers, revenue, approvals, reporting, or owner bottlenecks, it is probably operating-layer work.
That is where Omni by Enterprise DNA fits. We help scope, build, and manage the agent inside the way the business actually runs.
Bring the completed brief to a scoping call. The useful conversation starts when we can look at the real workflow, the real handoff, and the real review point.