AI Agents in the Back-Office of a Service Company
Four agents — Qualifier, Proposal, Follow-up, Backoffice — that quietly run the unsexy half of a consulting firm. Here's what each one actually does, and where humans still have to step in.
Most "AI for sales" pitches are theatre — a chatbot that summarizes a meeting, a tool that auto-drafts a follow-up. Useful, but not transformative. The transformation happens when you stop using AI as a feature and start treating it as a small team of agents that each own a part of the pipeline.
The four agents
In a service business, the back-office work that consumes 60% of partner time falls into four buckets. Each one is now a job for an agent.
1. Qualifier
The Qualifier reads every inbound lead — email, WhatsApp message, form submission — and decides:
- Which scope this is asking about
- Whether the budget signal is plausible
- What the urgency looks like (timeline mentioned? deadline?)
- A temperature: hot / warm / cold
It then writes a structured note to the lead and assigns a manager. Humans review the temperature flag in 30 seconds instead of reading the email cold.
2. Proposal
The Proposal agent owns the first draft. It pulls from your scope library — the reusable description of every service you sell — and combines them into a coherent technical proposal in .docx. The team edits, prices, and sends.
3. Follow-up
The Follow-up agent watches every deal that's stuck in "proposal sent" and writes contextual nudges with a humanized cadence (48–180s random delay between sends, exponential gap between attempts). It uses the Message-ID / In-Reply-To headers to track replies, so it knows when to back off.
What it does not do: send anything without your approval. The agent drafts and queues. A human flips the switch.
4. Backoffice
The least sexy and most valuable. Backoffice handles:
- Routing signed contracts to the right project
- Closing won deals and creating projects
- Linking documents to projects automatically
- Maintaining the audit trail
Quiet, boring, mistake-free. This is the agent that prevents the "we forgot to onboard them" disasters.
Where humans still have to step in
Three places, always:
- Pricing. AI can suggest a range from comparables, but the final number is a judgement call.
- Trust signals. The first cold message in a long sequence — the one that establishes credibility — is still hand-written.
- The "no" call. Walking away from a deal that doesn't fit is a senior decision, not an automation.
The compound effect
Each agent is mildly useful alone. The compounding shows up at the end of the quarter:
- Proposal turnaround dropped from 5 days to 12 hours.
- Lead-to-first-touch dropped from 4 hours to under 2 minutes.
- Senior partners spent 40% more time on actual delivery.
The shift didn't come from one thing getting ten times better; it came from four small two-times improvements stacking on the same workflow.
Each tenant gets the four agents pre-configured. Edit prompts, point them at your scope library, and watch the queue clear.
Wire up your first agent