Fluxx Tech
At 50 messages a day, one capable person handles everything and the business feels responsive. At 500, that same setup fails in a predictable sequence: replies slow down, enquiries are missed, answers become inconsistent, and nobody can say what is happening.
The fix is not just hiring. It is structure.
What breaks, and in what order
Volume | What breaks first |
50/day | Nothing; one person copes |
100/day | Evenings and weekends; messages pile up overnight |
200/day | Consistency; two people answer differently |
300/day | Ownership; conversations fall between people |
500/day | Everything, unless automation and structure exist |
The transition point for most businesses is between 100 and 200. That is when to build the structure, not at 500 when you are already drowning.
Work out how many people you need
Rough maths, with your own numbers:
Messages per day: 500
Share handled fully by automation: 60% → 200 need a human
Average human handling time: 4 minutes → 800 minutes
Working minutes per agent per day (allowing for breaks
and other work, roughly 70% utilisation): ~330
Agents needed: 800 ÷ 330 ≈ 2.5
Then add for peaks: volume is not spread evenly. If 40% of messages arrive between 7 pm and 10 pm, you need coverage then, not an even roster.
Measure your real handling time before trusting this. Most businesses guess two minutes and find it is six.
The three automation layers
Each layer removes work from the next.
Layer 1 — Deflection. Information people can get without a conversation: highlights, menu, prices in the bio, a catalogue link, an FAQ in the auto-reply. This reduces how many messages arrive at all.
Layer 2 — Automated answers. Keyword rules and an FAQ bot for the top 15 questions. Free, instant, always correct. This should handle the majority.
Layer 3 — AI. For unexpected phrasing and longer questions, with handover. See our AI agents guide.
Only what survives all three reaches a human. Getting layers 1 and 2 right is cheaper and more reliable than hiring.
Shift design
• Cover the hours messages actually arrive, which is usually evenings and weekends, not 9 to 5
• Overlap shifts by 15 minutes for handover of open conversations
• A named owner for every conversation, carried across the shift change
• An away message with a realistic reply time outside hours
• One person on “first response” duty during peaks, others on resolution
Keep answers consistent
With five people answering, consistency becomes the main quality risk.
• A canned reply library, maintained by one owner. See our templates article
• A single source of truth for prices, stock and policy, and a rule that nobody quotes from memory
• A short daily note to the team on anything that changed: offers, stock, timings
• Weekly review of ten random conversations
• One person who approves changes to standard answers
The failure mode is two agents quoting different prices to the same customer on two channels. The customer notices.
Escalation that works
• Clear triggers: complaint, refund, high value, angry language, repeat contact
• A named escalation owner per shift, not “the manager”
• A time limit: if unresolved in [X] hours, it goes up
• A daily review of escalated conversations
• Authority to resolve: an agent who must ask permission for every ₹200 adjustment creates delay and frustration
Give agents a small discretionary limit. It pays for itself in resolution speed.
Tooling, in order of need
1. Shared inbox across WhatsApp and Instagram, with assignment
2. Automation: keyword rules, FAQ, comment-to-DM
3. Contact records with history across channels
4. Tags and statuses
5. Reporting: response time, volume by hour, per-agent load
6. Canned replies
7. Lead and order tracking
8. AI layer, once the rest is working
Adding AI before a shared inbox exists is the common sequencing mistake.
The metrics to run operations on
Metric | Reviewed |
Messages per hour of day | Weekly, for rostering |
First response time | Daily |
Unassigned conversations | Live |
Open conversations older than [4] hours | Live |
Automation containment | Weekly |
Fallback themes | Weekly, to add rules |
Conversations per agent | Weekly |
Enquiry-to-customer rate | Monthly |
Revenue per conversation | Monthly |
Open conversations older than four hours is the single number to put on a screen. It catches almost every operational failure in real time.
Quality does not have to fall
Growing volume usually degrades experience. What prevents it:
• Automation absorbing the repetitive questions, so humans have time for real ones
• Agents named in their messages, so conversations feel personal
• A record of history, so customers never repeat themselves
• Realistic promises in away messages
• Enough staff at peak, not on average
A 90-day scaling plan
Phase | Focus |
Days 1–15 | Measure: volume by hour, handling time, top questions, current response time |
Days 16–30 | Build layers 1 and 2: deflection and keyword automation |
Days 31–45 | Shared inbox, assignment rules, canned replies, roles |
Days 46–60 | Roster to match the volume curve; hire if the maths says so |
Days 61–75 | Escalation rules, quality reviews, reporting dashboard |
Days 76–90 | Add AI for the remainder; review and adjust |
Frequently asked questions
At what volume should we hire? Only after automation is doing its job. Many businesses hire to answer questions a keyword rule could have handled.
Can one tool handle all of this? One shared inbox covering WhatsApp and Instagram with automation and reporting covers most of it. Separate tools per channel recreate the problem.
How do we keep quality while growing? Canned replies, one source of truth, named agents, and a weekly review of real conversations.
