AI features: Smart reply suggestions and automation

Clientflow's AI module analyzes your customer communication and suggests replies, actions, and templates — saving you time while delivering consistent service. The AI learns from your choices and adapts to your company's style.

---

What does the AI do?

Clientflow's AI works in several layers:

1. Reply suggestions — generates draft replies to customer inquiries
2. Action cards — suggests concrete actions (rescheduling, cancellation, etc.)
3. Template learning — learns from your replies and builds a growing bank of reply templates
4. Business rules — detects patterns and suggests rules that improve future replies
5. Feedback analysis — analyzes your feedback to find opportunities for improvement

Important: The AI never automatically sends replies to customers. Every suggestion is reviewed and approved by a team member before it's sent.

---

Activation

First time

The first time you enable AI features, a consent dialog appears:

• Confirm that you understand how the AI uses your data
• AI is enabled per company (tenant)
• You can disable AI again at any time

Company context

In Settings → General → Company context you can describe your business in free text:

• What do you do?
• Which services do you offer?
• Special policies (e.g. "We always give a 10% discount on complaints")

This text gives the AI context so the reply suggestions match your company's tone and practice.

---

AI reply suggestions

Automatic generation

When a customer sends a message:

1. The AI analyzes the message automatically
2. Within seconds, a reply suggestion appears below the compose field in the communications module
3. The suggestion is adapted to the channel (email format, SMS format or generic)

What the AI analyzes

Analysis	Result
Intent	What does the customer want? (E.g. rescheduling, complaint, inquiry, cancellation)
Sentiment	Positive, neutral or negative — indicates urgency
Customer history	Previous tasks, subscriptions, communication — provides context
Confidence	How sure the AI is about its suggestion (as a percentage)
Needs human	Whether the inquiry is too complex for an automated reply

Actions on the suggestion

Action	Description
Accept	Inserts the suggestion into the compose field — you can edit before sending
Dismiss	Hides the suggestion
Regenerate	Asks for an alternative suggestion
Save as template	Saves the suggestion as a reusable template for future use

Internal notes

The AI often adds an internal note — a short explanation to you of why it suggested this particular reply. The note is only visible to team members and is never sent to the customer.

---

AI action cards

Beyond text replies, the AI can suggest concrete actions based on the customer's inquiry. Actions appear as interactive cards that can be expanded and edited.

Available actions

Action	Description
Reschedule task	Move a task to a new date, with optional locking type and team-member assignment
Cancel task	Cancel a specific task
Update task constraints	Change time-window preferences for a task
Update subscription interval	Change the frequency of a subscription
Update subscription constraints	Change time constraints on a subscription
Update contact notes	Add or update notes on a contact
Update contact notifications	Change notification settings for a contact
Update lead status	Move a lead to a new column in the pipeline
Open complaint	Log a complaint based on the customer's message

How to use action cards

1. Review — the card shows a summary: "Current date → Suggested new date"
2. Expand — click to see and edit every field
3. Adjust — change details if the AI's suggestion isn't quite right
4. Confirm — click to perform the action
5. Completed — the card shows a green confirmation with a summary

Example: The customer writes "Can we move the window cleaning to next Tuesday?" → The AI shows an action card with:

• Current date: Wednesday April 8th
• Suggested date: Tuesday April 14th
• Team member: Ahmed (unchanged)
• Locking type: Day (suggested)

You verify, edit if needed, and confirm with one click.

---

Feedback and learning

Give feedback

The AI's suggestions get better over time — but only if you give feedback:

• Thumbs up / thumbs down — quick rating of the suggestion's quality
• Written feedback — elaborate on what was good or bad
• "Missing action" — flag if the AI should have suggested an action it didn't

What happens to your feedback?

1. Feedback is classified automatically into categories:
   • Business rule — "We always issue a credit note on complaints older than 3 days"
   • Missing action — "The AI should have suggested cancelling the task"
   • Style — "The tone was too formal"
   • Context gap — "The AI lacked information about the customer's history"
   • Factual error — "Wrong date or price in the suggestion"
2. Business rules with high confidence → created automatically as inactive rules (awaiting your approval)
3. Missing actions → counted and prioritized in the action-gap overview

---

AI learning (Settings)

The Pulse — overview dashboard

In Settings → AI learning you'll find "The Pulse":

• Adoption rate — how often AI suggestions are accepted (target: > 60%)
• Pending rules — number of business rules waiting for your approval
• Action gaps — categorized overview of AI improvement potential

Rules tab

Pending rules: The AI detects patterns in your feedback and suggests rules. Each rule shows:

• Category and description
• Confidence score
• Approve — makes the rule active (it influences future reply suggestions)
• Dismiss — discards the rule

Active rules: List of rules currently influencing the AI's replies. You can:

• Temporarily disable a rule
• Delete a rule permanently
• Max 15 active rules (to keep the AI's prompt size manageable)

Action gaps tab

Shows action types the AI tried to suggest but couldn't. Frequent action gaps indicate where the AI needs new capabilities.

Feedback stream tab

Live feed of all feedback from team members — useful for spotting trends and friction points.

---

Template learning

Clientflow learns from your actual sent replies:

1. Observation: Each outgoing reply is analyzed automatically (every 15 minutes)
2. Synthesis: Every day at 03:00 the AI reviews the observations and:
   • Creates new templates from recurring reply patterns
   • Updates existing templates with better wording
   • Demotes templates that are rarely used
3. Priority: Three template sources:
   • Seed (default) — predefined templates, lowest priority
   • Learned — automatically generated templates
   • Manual (manually saved) — templates you saved yourself, highest priority

---

AI settings

In Settings → AI learning and Settings → General:

Setting	Description	Default
AI learning enabled	Turn automatic learning on/off	Off
Minimum observations	How many replies the AI must analyze before generating templates	10
Synthesis confidence threshold	Minimum confidence for new templates	0.7
Template aging	Days before unused templates lose priority	90
Company context	Free-text description of your business	—
AI spend (USD)	Total AI usage — for budget control	—

---

Automated workflows

Beyond reply suggestions, you can create AI-powered workflows in Workflows:

1. Define a schedule (e.g. "Every Monday at 07:00")
2. Pick data sources (which tables the AI should read: tasks, subscriptions, contacts, etc.)
3. Write a prompt — describe what the AI should generate
4. Pick an output — which inbox the result is sent to

Examples:

• "Generate a weekly summary of every unfinished task and send it to the admin inbox"
• "Identify customers who haven't received service in 60+ days and generate a follow-up message"
• "Build an overview of this week's leads with recommended actions"

Event-based workflows

Event-based workflows trigger automatically when a message is received in an inbox — no AI credits are consumed.

Setup:

1. Pick the inbox the workflow should watch
2. Pick a message type: email, SMS or note
3. Optionally add sender contacts to limit the trigger to specific senders
4. Configure actions that run in order:

Action	What it does
Extract fields	Pulls data out of the message content via regex (e.g. phone number)
Find or create contact	Looks up a contact based on extracted fields — or creates a new one
Write note	Writes an internal note on the contact's conversation
Close conversation	Automatically closes the sender's conversation after every action

Example: A supplier sends an SMS in the format "New customer: Peter Hansen / 4512345678". The workflow extracts the name and phone via regex, finds or creates the contact and writes a note — all without manual intervention.

---