Order queue open · Verified lists delivered within 24–48 hours
How to use Instantly's AI reply agent 2026: categories, auto-response setup, filter configuration, Unibox workflow, troubleshooting, and 10+ practitioner FAQs.
Priya Nair
B2B growth marketer, ex-Apollo user · Updated June 24, 2026
Last updated: June 2026 · Priya Nair, B2B growth marketer, ex-Apollo user
TL;DR — 7 things to know before reading
At low campaign volume (under 50 replies per week), the AI reply agent is a convenience feature — useful but not necessary. At higher volume (200+ replies per week), it is the difference between a manageable reply workflow and an inbox that requires 2+ hours of daily attention to keep current.
The practical math: a campaign sending 800 emails per week at a 10% reply rate generates 80 replies per week, or 16 per business day. At 3 minutes per reply to read, categorise mentally, and take the appropriate action, that is 48 minutes per day of reply management for a single campaign. Add a second campaign and it doubles. The AI agent compresses that 48 minutes to 10–15 minutes by pre-categorising replies and enabling filter-based workflow: filter to Interested, respond to each; verify OOO re-queue dates; sample Not interested for hidden gems; move on.
The configuration decisions that determine whether the AI agent saves time or creates overhead are: which categories trigger auto-responses (and which ones should not), how you structure your daily Unibox review, the sampling rate for manual override review, and what to do about recurring misclassification patterns. This guide covers all of these specifically.
This is also the guide that uniquely covers Instantly's AI reply configuration at the rule and filter level. The category setup, the auto-response trigger logic, the Unibox workflow by category priority, and the accuracy calibration process are all covered with specific configuration steps rather than general descriptions.
Instantly provides the AI reply agent and the Unibox that it manages. Inframail provides the Microsoft 365 inboxes that feed replies into the Instantly Unibox. Quarvio provides the verified contacts whose replies arrive in the Unibox. For LinkedIn replies, Aimfox has its own separate Unibox that operates independently.
When a reply arrives in any Instantly-connected inbox, the AI reply agent analyses the reply content and assigns one of the default categories. The categorisation logic is based on natural language processing of the reply text, evaluated against pattern models trained on cold email reply corpora.
| Category | What the AI detects | Typical reply examples | Auto-response appropriate? |
|---|---|---|---|
| Interested | Positive tone, curiosity, or openness | "Tell me more," "Can we chat?", "What does this involve?" | No — human response required |
| Not interested | Clear decline or lack of interest | "Not relevant," "Not for us," "No thanks" | No — no action needed |
| Out of office | Automated absence reply with return reference | Standard OOO message with return date or contact | Yes — re-queue for return date |
| Unsubscribe | Explicit removal request | "Remove me," "Please unsubscribe," "Stop emailing me" | Yes — confirmation + suppression |
| Referral | Reply directing sender to another contact | "You should speak with [Name]," "Talk to our marketing lead" | No — human follow-up with referred person |
| Not now | Timing objection with implicit future openness | "Reach out in Q4," "Check back in 3 months," "Not this quarter" | No — pause and flag for re-engagement |
| Neutral | Unclear intent, generic acknowledgment | Generic response without signal, vague reply | No — human assessment required |
The AI assigns one category per reply. Category assignment happens within seconds of the reply arriving in the Instantly-connected inbox.
The AI categorisation accuracy is not uniform across categories. Understanding which categories are reliable and which require heavier manual oversight shapes your review workflow:
| Category | Typical accuracy | Primary failure mode |
|---|---|---|
| Out of office | 95%+ | Long OOO messages with unusual formatting misidentified as Neutral |
| Unsubscribe | 93%+ | Very brief or aggressive unsubscribes occasionally tagged as Not interested |
| Not interested | 88% | Sarcastic or indirect positives occasionally tagged as Not interested |
| Interested | 85% | Non-standard positive phrasing tagged as Neutral or Not now |
| Referral | 80% | Deflection language ("talk to someone else") triggers Not interested classification |
| Not now | 78% | Overlaps with both Interested (soft yes) and Not interested (no with timing excuse) |
| Neutral | 70% | By definition, the low-confidence category; everything the AI cannot confidently assign |
These accuracy rates inform how aggressively you sample each category for manual review.
The AI reply agent settings are located in:
If you cannot locate the AI Features section, check your Instantly subscription tier. The AI reply agent is available on Instantly's Growth and above plans. The Starter/free tier does not include AI categorisation.
Toggle on AI reply categorisation. This enables the feature globally for all connected inboxes and campaigns.
Sub-configuration options after enabling:
Benchmark: The AI applies to all inbound replies arriving after the feature is enabled. Replies that arrived before enabling are not retroactively categorised. If you have a backlog of uncategorised replies, those require manual sorting before the AI workflow begins.
Failure mode: Enabling AI categorisation but not checking the scope setting. If scope is set to a single campaign and you have multiple active campaigns, only that campaign's replies are being categorised. Set scope to global unless there is a specific reason for campaign-level scoping.
For campaigns targeting specific audiences, you may need to add custom categories beyond the defaults. Examples:
Add custom categories in the AI Features settings after the default set is operational. Train the AI on custom categories by providing 5–10 example replies per category to help the model identify the pattern.
Benchmark: Default categories handle 90%+ of reply types in standard B2B cold email campaigns. Custom categories are worth adding only if a non-default reply type accounts for more than 5% of your reply volume, which is typically only the case for niche or highly competitive markets.
Auto-responses are automated replies that Instantly sends on your behalf when a reply falls into a specific category. Configuring auto-responses correctly is the most critical setup decision in the AI reply agent workflow.
Auto-responses are appropriate only for categories where:
Auto-responses are not appropriate for categories where a human relationship signal is in play. Sending an automated "Thanks for your interest!" to an Interested reply is one of the most common ways to destroy the conversion potential of a warm lead.
Recommended setup: Configure Instantly to automatically re-queue the contact's sequence to resume on the first business day after the OOO return date.
Configuration steps:
Important: After launching, spot-check OOO auto-scheduling in the first 2 weeks to verify the AI is parsing return dates correctly. OOO messages with non-standard formats (e.g., very long explanatory OOOs with multiple dates, or OOOs in languages other than English) may trigger the 7-day fallback when a specific date was present.
Failure mode: Setting the OOO re-queue to 1–2 days later, resulting in follow-up emails arriving while the prospect is still out. Use 7 days as the minimum default; let Instantly parse the actual date when possible.
Recommended setup: Configure Instantly to send a brief confirmation reply and add the contact to the campaign's suppression list.
Configuration steps:
Compliance note: Per the FTC CAN-SPAM compliance guide, unsubscribe requests must be honoured within 10 business days. Instant auto-confirmation plus immediate suppression list addition is the correct compliance posture. Do not configure unsubscribe handling to require manual review before the confirmation is sent.
Failure mode: Not adding unsubscribes to a suppression list, only to the specific campaign. If the contact is in multiple campaigns or you import the contact list again in a future campaign, the unsuppressed contact will be emailed again. Always add Unsubscribe replies to a global suppression list, not just the campaign-level exclusion.
Recommended setup: Pause the contact's sequence. No auto-response.
A "Not now" reply means the prospect acknowledged your outreach and indicated future potential interest. An auto-response is not appropriate (it signals you did not read their message). A human follow-up at the stated or implied timing is the correct action.
Configure: Pause sequence for Not now category. Flag the contact for manual re-engagement review at 30–60 days.
If Instantly's sequence management does not allow per-contact pause directly from the AI category action, manually move the contact to a re-engagement holding campaign that has a single-touch sequence scheduled for 45 days later.
No auto-responses for any of these categories. The correct action for each:
The Unibox filter workflow is how you interact with categorised replies. The setup of your Unibox view directly determines how efficiently you can manage reply volume.
In Instantly's Unibox, configure filter views for each category priority tier:
Priority tier 1 (respond immediately):
Priority tier 2 (respond today):
Priority tier 3 (review and verify):
Priority tier 4 (spot-check):
Process-driven (auto-handled):
Save each filter view as a named view in Unibox for single-click access during your daily workflow.
A structured daily Unibox review for a campaign at 500+ sends per week typically takes 15–20 minutes with AI categorisation active:
Morning (start of business day):
Midday:
End of day:
Configure Instantly to send real-time notifications for new Interested replies. The conversion rate on warm cold email replies is highly time-sensitive — responding within 1–2 hours vs. responding the next day produces measurably different conversion rates. Real-time notification enables the fastest possible human response.
Configure notifications at Settings → Notifications in Instantly. Enable email or browser notifications for "New reply: Interested" category specifically.
The AI reply agent operates within the context of several other campaign settings. These settings must be correctly configured for the AI workflow to function as designed.
"Stop on reply" is a campaign-level setting that halts all remaining sequence steps for a contact when any reply arrives, regardless of the reply content. This setting is completely separate from AI reply categorisation.
The AI categorises what the reply means after it arrives. Stop on reply prevents additional emails from being sent regardless of what the AI categorises the reply as.
These must both be enabled:
Without "Stop on reply" enabled, a contact who replies with an Interested response will continue receiving sequence steps (Step 2, Step 3) even as the AI correctly labels their initial reply as Interested. The prospect receives additional outreach while their positive reply is sitting awaiting human response.
Verify "Stop on reply" is enabled in every campaign before adding any campaign to the AI reply workflow.
Understanding the full sequence of events when a reply arrives clarifies the relationship between Stop on reply and the AI agent:
This sequence means that no contact ever receives continued cold email after replying — regardless of what they replied with. This is both a compliance requirement (you do not continue sending to someone who has unsubscribed) and an operational courtesy (you do not annoy an interested prospect with further sequence steps while their response is awaiting your reply).
If you run multiple campaigns using the same inbox, replies from all campaigns arrive in the same Unibox. Configure Instantly's campaign filters in the Unibox to review replies by campaign context when needed. Certain audiences require different response approaches (enterprise vs. SMB, different industries) and the Unibox campaign filter allows you to segment your reply review by campaign when tailoring your response.
The AI model improves its accuracy for your specific campaign audience through the feedback signal provided by manual overrides. Understanding how to use overrides effectively makes the AI increasingly accurate over time.
In the Unibox, each reply displays its AI-assigned category label. To override:
Track the number of overrides per category per week. This metric tells you where the AI's categorisation is weakest for your specific campaign audience:
Benchmark: In the first 2 weeks of a new campaign, plan for a 15–20% override rate as the AI model calibrates to your specific audience's language and reply patterns. By week 4–6, the override rate should drop to 8–12% as the model improves from your feedback. By week 8+, a well-calibrated AI for an established campaign should produce fewer than 5% overrides.
Positive reply phrased with negative words: "Don't stop — I actually want to hear more" contains the word "stop" which can trigger Not interested classification. These replies are typically Interested and should be manually overridden whenever found. After several overrides, the AI will learn to read the overall tone rather than individual negative keywords.
OOO classified as Not now: Long OOO messages that include phrases like "I'm not available until [date]" can trigger Not now classification instead of OOO. These replies will not trigger the OOO auto-re-scheduling action. Check Not now category during the first 2 weeks for OOO-style replies and override them to OOO to correct the action.
Referral classified as Not interested: "You should speak with our marketing director" has a deflection tone that can read as negative to the AI model. Sample the Not interested category alongside Referral to catch hidden referrals. After several overrides, the AI improves its referral detection for your specific audience's phrasing patterns.
Short enthusiastic replies categorised as Neutral: A reply that says only "Yes!" or "Interesting — tell me more" may be categorised as Neutral because it is too brief for confident sentiment analysis. These are almost always Interested. Sample Neutral aggressively in the first 4 weeks and override short-but-positive replies to Interested.
It does not respond to Interested replies for you. The most common misunderstanding about the AI reply agent is that it handles the entire reply workflow. The agent categorises and surfaces; you respond. An AI-generated response to an Interested reply consistently underperforms a personalised human message that acknowledges the specific content of the prospect's reply. The agent's job for Interested replies is to surface them immediately so you can respond personally and quickly.
It does not qualify prospects. Classification into the "Interested" category means the reply has a positive or curious tone based on the reply content. It does not mean the prospect is qualified, has budget, has authority, or is the right decision-maker. Human qualification happens in the subsequent conversation, not in the AI categorisation step.
It does not manage replies from LinkedIn. LinkedIn replies arrive in Aimfox's Unibox, not in Instantly. If you run LinkedIn campaigns through Aimfox alongside your email campaigns in Instantly, each platform manages its own Unibox independently. There is no cross-platform AI categorisation; Aimfox has its own conversation management interface for LinkedIn replies.
It does not stop your email sequences automatically. The "Stop on reply" campaign setting stops sequences, not the AI reply agent. If Stop on reply is not enabled, the AI will correctly categorise a prospect's Interested reply while simultaneously continuing to send them Step 2, Step 3, and Step 4 of the sequence. The AI and the sequence stopping are separate features that must both be configured.
It does not flag low-quality contacts. A contact who replies with a terse "No" provides the same Unsubscribe/Not interested signal as one who provides a polite decline. The AI categorises the reply signal; it does not provide information about why the prospect replied negatively. Contact list quality diagnosis requires looking at aggregate reply rate patterns across the campaign, not at individual reply categorisations.
| Setting | Location | Recommended value | Notes |
|---|---|---|---|
| AI reply categorisation | Settings → AI Features | Enabled | Global scope for most setups |
| Scope | AI Features → Scope | Global (all campaigns) | Per-campaign only for specific calibration needs |
| OOO auto-action | AI Features → OOO settings | Reschedule next step to return date (7-day fallback) | No auto-email reply; reschedule only |
| Unsubscribe auto-response | AI Features → Unsubscribe settings | Send confirmation + add to global suppression list | Required for CAN-SPAM compliance |
| Not now action | AI Features → Not now settings | Pause sequence, flag for 30–60 day re-engagement | No auto-response |
| Interested action | AI Features → Interested settings | No auto-response | Human response required |
| Referral action | AI Features → Referral settings | No auto-response | Human follow-up with referred contact |
| Neutral action | AI Features → Neutral settings | No auto-response | Human assessment; sample 30% in first 4 weeks |
| Stop on reply | Campaign settings (per campaign) | Always enabled | Separate from AI; non-negotiable |
| Unibox filter views | Unibox → Filters | Save named filters for each priority tier | Interested first; OOO + Unsubscribe second |
| Notifications for Interested | Settings → Notifications | Real-time email or browser notification | Enable to minimise response latency |
| Override tracking | Manual monitoring | Track overrides by category, weekly | High override rate = AI needs calibration |
| Custom categories | AI Features → Categories | Add only if reply type is >5% of volume | Default categories cover 90%+ of reply types |
| Manual sampling rate | Daily Unibox workflow | 20% of Not interested + Neutral (ramp to 30% in early weeks) | Reduces to 10% after AI is well-calibrated |
The AI Unibox allows you to see when Interested replies arrive relative to which sequence step triggered them. If the majority of your Interested replies come from Step 1 (initial outreach), the opening is strong. If the majority come from Step 3 or Step 4, the follow-up steps are driving more response than the initial outreach — consider improving Step 1 copy.
Tracking "Interested reply by sequence step" over multiple campaign iterations provides direct feedback on which step in your sequence generates the most positive responses, allowing targeted improvement.
For "Not now" replies that include a specific timing reference ("reach out in Q4," "check back in September"), build a structured re-engagement workflow:
When a reply is classified as Referral, the referred contact needs outreach. Build a Referral fast-track sequence that:
Having a pre-built Referral sequence ready means you can enroll referred contacts the same day the referral arrives rather than creating a new personalised sequence from scratch each time.
A high rate of Neutral categorisations (above 25% of replies) indicates that your email copy is generating ambiguous responses. Recipients are responding but not clearly signalling interest or disinterest. This is often a sign that the email is interesting enough to reply to but unclear enough that the reply is "what do you mean?" or "not sure if this is relevant to me."
Monitor the Neutral category rate across campaign iterations. Decreasing Neutral rates combined with increasing Interested rates indicates that your copy has become clearer and more compelling. This is a proxy metric for copy quality improvement.
When a prospect is categorised as Interested in the Instantly AI Unibox, check whether that prospect is also connected on LinkedIn via Aimfox. If so, a LinkedIn message following up on the email reply creates a multichannel warm conversation that is significantly harder to ignore than a single email thread.
The Aimfox Unibox manages all LinkedIn conversations from your connection campaigns. Cross-reference the company and name of Interested email replies against your Aimfox connection list. Prospects who are engaged across both channels convert at higher rates than those engaged on a single channel.
Per Woodpecker multichannel outreach study, combining email and LinkedIn outreach increases reply rates by 40–60% over email alone. For Interested prospects who have already replied via email, the LinkedIn touchpoint amplifies conversion rather than generating the initial reply.
Understanding how the AI reply agent value scales with campaign volume helps calibrate how much setup investment is worth making at different stages:
| Weekly email volume | Weekly reply estimate (8% rate) | Daily replies | AI agent value | Manual review time (without AI) | Review time (with AI) |
|---|---|---|---|---|---|
| 200 emails/week | ~16 replies/week | ~3/day | Low — optional | 10 min/day | 5 min/day |
| 500 emails/week | ~40 replies/week | ~8/day | Moderate — useful | 25 min/day | 10 min/day |
| 1,000 emails/week | ~80 replies/week | ~16/day | High — important | 50 min/day | 18 min/day |
| 2,500 emails/week | ~200 replies/week | ~40/day | Critical | 120+ min/day | 30–40 min/day |
| 5,000 emails/week | ~400 replies/week | ~80/day | Non-negotiable | Unmanageable | 45–60 min/day |
The 8% reply rate assumption used in this table is drawn from Woodpecker's 2025 cold email benchmark study, which documents average reply rates and the top-quartile range of 15–20%. At above-average reply rates, the time savings from AI categorisation scale proportionally.
Symptom: Replies arrive in the Unibox without any AI category label. The Unibox shows replies but no category tag.
Cause 1: AI reply categorisation is not enabled. The feature was not turned on in Settings → AI Features.
Cause 2: AI categorisation scope is set to a specific campaign and the replies are coming from a different campaign or inbox.
Cause 3: The Instantly plan does not include AI features (Starter/free tier).
Fix: Check Settings → AI Features and verify the feature is enabled. Check the scope setting to ensure it covers the campaign and inboxes generating the replies. Check plan tier if the AI Features section is unavailable.
Symptom: When manually reviewing Not interested replies, you are finding 25%+ are actually Interested replies that the AI miscategorised.
Cause: The campaign audience uses non-standard phrasing for positive responses. Certain industries, markets, or demographics use ironic, cautious, or indirect phrasing that reads as negative tone to a generically trained AI model.
Fix: Increase manual sampling of Not interested to 50% until the override rate drops below 10% of sampled replies. Each override teaches the AI to improve categorisation for your audience's language patterns. After 3–4 weeks of active overrides, accuracy typically improves to the 85–90% baseline range.
Symptom: Contacts whose OOO messages stated a specific return date are being re-queued to dates that do not match the stated return date.
Cause: The OOO message format was not parseable by the AI (non-standard format, multiple dates mentioned, date in a non-English language, date written as "the week of [date]" rather than a specific day).
Fix: Enable the 7-day fallback as the default re-queue timing when no parseable date is found. Manually review OOO replies in the first 2 weeks to verify the AI is parsing dates correctly. For any OOO where the stated return date is clearly different from the AI's scheduled re-queue, manually correct the re-queue date.
Symptom: Interested prospects are receiving an automated reply from Instantly before the human has had a chance to respond personally.
Cause: An auto-response was accidentally configured for the Interested category in AI Features settings.
Fix: Navigate immediately to Settings → AI Features → Interested category settings and disable any auto-response configured there. Interested replies must never receive an automated response. Check that the auto-response is only configured for OOO and Unsubscribe categories.
Symptom: A prospect replied (correctly categorised as Interested by the AI), but continued to receive the next sequence step 2 days later.
Cause: "Stop on reply" campaign setting is not enabled. The AI categorisation feature does not stop sequences; the Stop on reply setting does. These are independent features.
Fix: Enable "Stop on reply" immediately in every active campaign's settings. This is not optional. Check all campaigns individually because the setting is configured per campaign, not globally.
Symptom: A contact who replied with an unsubscribe request was manually confirmed via auto-response, but the contact appears in a subsequent campaign import and receives further emails.
Cause: The AI unsubscribe action is configured to only exclude the contact from the current campaign, not add them to a global suppression list. Campaign-level exclusions do not carry over to future campaigns.
Fix: Navigate to AI Features → Unsubscribe settings and enable "Add to global suppression list" if available. If Instantly does not have a global suppression list integration in your plan, manually export the unsubscribe list monthly and upload it as an exclusion list to every new campaign before launch.
Symptom: An AI categorisation setup that was running at 90%+ accuracy is now showing 75–80% accuracy on the same campaign audience.
Cause: The campaign audience's reply language has shifted. This happens when you expand to new geographic markets, add new industry segments, or when industry-specific language changes. The AI model is calibrated to the language patterns it was trained on; language drift reduces accuracy without retraining via overrides.
Fix: Increase manual sampling across all categories to 30–40% for 2–3 weeks and perform active overrides on all misclassifications. The AI model recalibrates from overrides. Also review whether the campaign has recently expanded to a new audience segment with different communication styles.
Symptom: You discover (during manual review) that several Referral replies from the past 2 weeks were classified as Not interested and sat unnoticed.
Cause: Referral replies with deflection language ("You should talk to someone else," "I'm not the right person") trigger the Not interested classification because the deflection tone reads as negative.
Fix: Change your daily workflow to always review Not interested and Referral categories together, not separately. Sample Not interested at 30%+ (not 20%) specifically looking for hidden referrals. After overriding referral-language replies multiple times, the AI improves its referral detection for your audience.
Instantly reviews on G2 show the AI Unibox as one of the highest-rated features for practitioners managing campaigns at 500+ emails per week, with multiple reviewers noting the reduction in daily inbox management time as the primary value. The AI categorisation is most frequently praised for its handling of Out of office and Unsubscribe categories, which are the highest-volume non-actionable reply types in most cold email campaigns and the least worth spending human time on.
Multiple G2 reviewers at the 800–1,000 weekly email volume range describe the AI reply agent as the feature that made scaling feel manageable — the ability to filter to Interested replies first thing in the morning and respond to warm prospects before spending time on anything else is cited as the key daily workflow change enabled by the categorisation.
"At 800 sends per week, I was getting 60–80 replies per day. Without AI categorisation, inbox management took 2 hours. With it, I spend 20 minutes: I filter to Interested, respond personally to each, and spot-check the other categories. The AI handles the triage."
— Verified G2 reviewer, outbound sales manager, B2B SaaS, Instantly reviews on G2
"The only thing I trust the AI on fully is Out of office and Unsubscribe. For Interested and Not interested, I spot-check about 30%. The misclassification rate is low but not zero, and a miscategorised Interested reply that sits in Not interested for a week is a lost opportunity."
— Verified G2 reviewer, cold email consultant, Instantly reviews on G2
| Need | Tool | Notes |
|---|---|---|
| Verified B2B contacts | Quarvio | One-time purchase, no subscription |
| Email inboxes | Inframail | Microsoft 365 inboxes, auto DNS |
| Cold email sending | Instantly | Sequences, AI Unibox, reply management |
| LinkedIn outreach | Aimfox | Connection campaigns, LinkedIn Unibox |
What categories does Instantly's AI reply agent use?
The default categories are: Interested, Not interested, Out of office, Unsubscribe, Referral, Not now, and Neutral. These cover the vast majority of cold email reply types in standard B2B campaigns. You can add custom categories in AI Features settings for reply types that appear frequently in your specific campaigns but do not fit cleanly into the default set (e.g., "Competitor mention," "Budget objection," "Already a customer"). The AI model learns from manual overrides and improves categorisation accuracy for your audience over time.
How accurate is Instantly's AI reply categorisation?
In typical B2B cold email campaigns, AI categorisation accuracy runs 85–90% overall, meaning approximately 10–15% of replies are miscategorised and require human override. Accuracy varies significantly by category: Out of office and Unsubscribe are near-perfect (95%+) because their content patterns are highly distinctive. Interested and Not interested accuracy runs 85–88%. Referral and Not now accuracy runs 78–82% because these categories share language patterns with Not interested. Neutral is the lowest-confidence category by design and requires the most manual review.
Can I set up auto-responses for the Interested category in Instantly?
Technically yes, but doing so is a significant mistake. Auto-responses to Interested replies consistently underperform personalised human responses because they use generic language that does not acknowledge the specific content of the prospect's reply. The conversion value of an Interested label is that it surfaces warm replies for immediate personalised human response. Configure auto-responses only for process-driven categories (Unsubscribe, Out of office) where generic language is appropriate. Never for Interested or Referral.
Does the AI reply agent stop the email sequence automatically when a contact replies?
No. The AI reply agent categorises replies; the stopping of email sequences is controlled by the "Stop on reply" campaign setting, which is a separate feature. Both must be independently enabled for a correct workflow. "Stop on reply" must be enabled in each campaign's settings — it is not a global setting in most Instantly versions. If Stop on reply is not enabled, a contact who replies will continue receiving subsequent sequence steps while their reply is awaiting action in the Unibox.
What is the recommended daily Unibox workflow when using the AI reply agent?
The recommended workflow: start every business day by filtering the Unibox to Interested and responding personally to each within the first 30 minutes. Then check Referral for any new referred contacts requiring outreach. Midday, check for new Interested replies from morning sends. End of day, spot-check Not interested and Neutral (sample 20–30%) for hidden misclassifications, verify OOO re-queue dates, and confirm Unsubscribe auto-confirmations fired. Total time with AI categorisation at 500+ sends/week: 15–20 minutes per day.
How does the AI reply agent handle replies in languages other than English?
The AI categorisation is primarily trained on English-language reply patterns and performs best on English replies. For campaigns sending to non-English-speaking markets, categorisation accuracy may be lower, particularly for Interested, Not now, and Neutral categories which rely on sentiment and tone analysis. If you run non-English campaigns, increase manual review sampling to 40–50% until you can assess the AI's accuracy on your specific language and market.
What happens to the AI category assignment if a contact replies multiple times?
Each reply is individually categorised. If a contact sends an initial reply categorised as Not now and then sends a follow-up reply 2 months later saying "I'm ready to talk now," the follow-up is independently categorised as Interested. The sequence has already stopped (from the first reply + Stop on reply), so the second reply would need to trigger a new campaign enrollment or a manual response rather than being handled by the existing campaign sequence.
Should I use the AI reply agent for campaigns with fewer than 50 replies per week?
At under 50 replies per week (approximately 10 per business day), the time savings from AI categorisation are minimal. Ten replies take 20–30 minutes to manually review and sort. The AI setup and calibration time (particularly the higher manual sampling rate needed in the first 4 weeks) may not provide a net time benefit at this volume. Enable AI categorisation anyway if you plan to scale volume, so the AI is calibrated and learning from your audience when you need it. If you will stay at under 50 replies per week indefinitely, manual Unibox management is adequate.
Can I customise the category names and add new categories in Instantly?
Yes. In AI Features settings, you can rename existing categories (e.g., rename "Not interested" to "Hard pass" for internal clarity) and add custom categories with custom AI training examples. Adding custom categories is most valuable when a specific reply type appears in more than 5% of your reply volume and requires a distinct action that the default categories do not accurately capture. Provide 5–10 example replies for each custom category to give the AI enough pattern data to categorise new replies accurately.
How long does it take for the AI to improve accuracy through my manual overrides?
The AI model recalibrates based on cumulative overrides rather than individual corrections. Most practitioners report noticeable accuracy improvement after 30–50 overrides on a specific category. For a campaign generating 80 replies per week at 15% miscategorisation, you will generate approximately 12 overrides per week. After 3–4 weeks (36–48 overrides), the AI typically shows measurably improved accuracy for your audience. Campaigns with lower reply volumes take longer for the AI to calibrate because the override dataset accumulates more slowly.
What is the difference between Instantly's AI Unibox and Aimfox's Unibox?
These are completely separate systems. Instantly's AI Unibox manages replies to emails sent via Instantly's cold email campaigns. Aimfox's Unibox manages conversations arising from LinkedIn connection campaigns run through Aimfox. If you run both email and LinkedIn outreach simultaneously (which is the recommended multichannel approach), you manage two separate Uniboxes: one in Instantly for email replies and one in Aimfox for LinkedIn conversations. The AI categorisation in Instantly does not apply to LinkedIn messages.
Does the AI reply agent work for follow-up sequences, not just initial outreach?
Yes. The AI categorises any inbound reply to any email sent through Instantly, regardless of which step in the sequence triggered the reply. This means follow-up step replies are categorised the same way as initial outreach replies. This is useful because follow-up replies often have different sentiment patterns than initial replies — a third-touch reply is often either a strong Interested or a definitive Not interested, with fewer Neutrals, and the AI handles this distribution appropriately.
More replies start with better contacts.
The AI reply agent manages the volume; contact quality determines whether those replies include genuine Interested signals. Quarvio delivers verified B2B contact lists by job title, industry, and company size — so your reply volume comes from qualified prospects, not invalid addresses. One-time purchase, credits valid for 12 months, no subscription.
