Agentic workflows

An agentic workflow is not a traditional “if this, then that” automation. It’s an AI agent that reads a form submission, thinks about it, and decides what to do — in the same way a human assistant would.

---

Traditional automation vs agentic workflow

	Traditional automation	AmazingForms agentic workflow
How it works	Fixed sequence: trigger → step 1 → step 2	AI reads submission, reasons, decides actions
Personalisation	Merge tags (fill-in-the-blank)	Fully written by AI in your voice
Branching	Explicit if/then rules you define	Agent decides dynamically based on content
Handles edge cases	Only what you anticipated	Agent can handle novel inputs gracefully
Setup	Configure each step manually	Describe what you want in plain English

---

Creating a workflow

In the AI builder

Describe what should happen after submission in the same prompt as your form:

“If the project budget is under £1,000, send a polite ‘not a fit’ reply. Otherwise, send a personalised response in my voice asking for a discovery call and add them to my CRM as a hot lead.”

The AI builder will generate both the form and the workflow. You can review and edit each action in the workflow editor before publishing.

In the workflow editor

Open an existing form → Workflow tab → Edit workflow.

The workflow editor shows your actions as a visual sequence. You can:

• Add, remove, or reorder actions
• Edit the prompt for AI-written emails
• Set conditions on individual actions
• Add a Wait step to delay actions

---

What the agent does on each submission

When a form is submitted, the agent runs automatically:

1. Reads the full submission, including all field values
2. Assesses the submitter’s intent, urgency, and fit based on the answers
3. Decides which configured actions to execute (and in what order, if conditional)
4. Writes any personalised content (email replies, CRM notes) tailored to this specific submission
5. Executes the actions: sends emails, updates CRM, posts to Slack, fires webhooks, etc.
6. Logs its reasoning and actions to the submission’s AI Log

This entire process typically completes in 3–10 seconds after submission.

---

Viewing the agent’s reasoning (AI Log)

For any submission, you can see exactly what the agent reasoned and decided:

1. Go to AmazingForms → Submissions
2. Click the submission you want to inspect
3. Click the AI Log tab

The log shows:

• The agent’s assessment of the submission
• Which actions it decided to execute and why
• The content it generated (e.g. the email it wrote)
• Any errors or retries

This is useful for auditing, debugging, and understanding why the agent behaved a certain way.

---

Writing effective workflow prompts

The workflow prompt is the plain-English instruction you give the agent. Here are patterns that work well:

Lead qualification example

Read the project description and budget fields.

If the budget is under £500 or the project description is vague (fewer than 20 words),
send a warm but honest reply explaining we're not a good fit for smaller projects
and suggest some DIY alternatives. Tag the contact as "not-a-fit".

Otherwise, send a personalised reply in a friendly, direct professional tone.
Reference specific details from their project description to show I've read it carefully.
Ask for a 20-minute discovery call and include my Calendly link: https://calendly.com/me/discovery.
Tag the contact as "hot-lead" and move them to the "Qualified" stage in the CRM.

Booking enquiry example

Send a personalised confirmation email thanking them for their enquiry.
Reference the service they selected and the date they requested.
Let them know I'll confirm within 24 hours.
Add a CRM note summarising what they've booked and any special requirements they mentioned.
Tag them as "pending-booking".

Tip

Include a short sample of your writing style in the workflow prompt — a sentence or two. The agent will match your tone. Example: “Write in a warm, direct tone. No corporate jargon. Short sentences.”

---

Model routing

Mode	Models used	Cost	Best for
Managed AI (default)	Cost-efficient Flash-class models	Included in Pro plan credits	Most workflows — fast, good quality
BYOK	Your choice (GPT-4o, Claude Sonnet, etc.)	Your API bill	High-stakes workflows, frontier quality, privacy, high volume

Switch to BYOK in Settings → AI → API keys. You can set a different model per workflow in BYOK mode.

---

Timeouts and retries

• The agent typically completes in 3–10 seconds
• Complex workflows with many actions may take up to 30 seconds
• If the agent times out, the submission is always saved first — no data is lost
• Failed workflows appear with an error badge in Submissions
• You can manually re-trigger a failed workflow from the Submissions screen: open the submission → Re-run workflow

---

Fair use credits

Managed AI includes a credit allocation with every Pro licence. Credits are consumed per workflow run, weighted by complexity (number of AI actions, length of generated content).

The allocation is generous for typical usage. If you’re running high volumes (thousands of AI-processed submissions per day), consider BYOK mode for cost predictability.

View current usage: Settings → AI → Usage dashboard