A request for “an AI workflow” often contains several different jobs. A form needs to pass data to a CRM. Someone needs to understand what the prospect is asking. A colleague needs enough context to follow up. Each step deserves its own decision.
Here is a practical way to think about that decision, using illustrative marketing workflows. It starts with the work you want done, before choosing a platform.
Use rules when the answer is already defined
Suppose a form includes a country field, and your team has agreed which person handles each country. Sending the enquiry to that person is a rules-based task. The same applies to copying a campaign identifier into the CRM or notifying an owner when a required field is missing.
Write the rule in plain language. If the team can agree what should happen for each input, a conventional integration is a useful first option. Test the awkward inputs too: duplicate submissions, missing fields and values that do not match any agreed category.
Consider AI where language needs interpretation
A free-text enquiry can contain several requests, background information and an unclear next step. AI may help prepare a short summary for a colleague to read. The summary has to preserve the request and make uncertainty visible; it should not turn an ambiguous message into a confident buying signal.
A useful first boundary is to let the assistant prepare information for review. Give the person reviewing it access to the original message. Decide separately whether any part of that interpretation is reliable enough to trigger another action.
Anthropic distinguishes workflows from agents: workflows follow defined paths, while agents decide how to proceed. That distinction helps when reviewing a proposal. Ask which decisions really need to be delegated, and which can stay explicit.
Split a reporting workflow into its parts
For example, collecting agreed figures, applying a calculation and filling a reporting template can follow fixed rules. Preparing a readable explanation is a different task. An AI draft can help express a change, but the team still needs to check that the explanation is supported by the data.
A rise in enquiries does not, on its own, establish why they rose. Keep observations and proposed explanations separate in the brief. Give the reviewer the reporting period, definitions and source figures alongside the draft.
Keep publishing decisions visible
The same distinction applies to a marketing website. Creating a draft page from an approved template and sending it for approval can follow a defined process. Drafting the copy may involve AI. The final publication step should follow the team’s actual approval process.
This is where the website matters as part of marketing infrastructure: the team needs to edit content, review it and move it into the right place without unnecessary handoffs. Adding an assistant should fit that process.
Questions to ask before commissioning the work
- Which steps follow fixed rules, and which involve interpreting information?
- What should happen when information is missing or the result is uncertain?
- Where can a person review the original input and the proposed action?
- Who can change the rules, instructions and source materials later?
- How will we compare total effort, including checking and correction?
Sketch one workflow with these decisions marked. You may find that most of it is an integration, with AI helping at one specific step. That is a useful scope to discuss with a technical partner, and an easier result for your team to evaluate.
See our automation and integration services or read about choosing a first AI workflow.
