Job page

Follow up on quotes you already sent

The hard part was the visit and the estimate. The leak is the silence after you hit send.

This page is about that job first: automated quote follow-up for a small business, without handing a send button to a model. Then the other jobs worth an agent: answers from your process, status lookup, analysis that points attention.

An agent here is not a personality and not a helper. It is a bounded job with a feed, stop rules, and a human who directs it. That is the Do beat of an Intelligent Business Layer.

We do not have a magic close-rate. We will not recycle the internet's favorite follow-up percentage. You already know you stop too early. The work is to make the next touches happen without sounding like a sequence.

What "automated follow-up" should mean

It should mean:

• The quotes you already sent are in one place (sheet, mailbox folder, estimate tool export).
• Status is a list: sent, replied, won, lost, expired.
• A job writes the next message in your voice.
• You send it, or you allow send after you have watched it for a stretch.
• Reply, win, or "not now" stops the sequence. A new quote does not get two voices at once.

It should not mean:

• A new CRM you will not open.
• A voice agent cold-calling your customers.
• A model inventing a discount you did not approve.
• Follow-up on people who already said no, or who never asked.

If the quotes are not in a list, start with cleaning that list. The agent cannot chase rows that do not exist.

A cadence you can actually run

Use a cadence that fits a trade or a shop, not a SaaS onboarding email.

• Touch 0. The quote itself. Clear total, what is included, how to say yes, when it expires if it expires.
• Touch 1. Two to four days later. Short. Did they get it. Any questions. No new pitch.
• Touch 2. About a week after send. One useful line: a scheduling window, a material lead time, a "we can still hold this if you reply."
• Touch 3. Near expiry, or about two weeks if you do not expire quotes. Last note. Permission to close the file.

Stop immediately on any reply, a win, a loss, or "please don't text." A human handles that thread from there.

Four touches counting the quote is enough to write down. If your work is same-week (catering, repair), compress the gaps. If your work is long-lead (addition, custom millwork), stretch them. Write the gaps on the job card. Do not copy a blog.

SMS or email: use the channel you already used for the quote. Do not surprise someone with a new app.

If the leak is quotes sitting after send, hello@cruxibl.com with how you send estimates is enough to talk. Draft-first unless you say otherwise.

Copy in a human voice

Write three drafts yourself once. The agent will reuse the pattern, not a slogan.

Touch 1, SMS

Hi {{Name}}. Chris at {{Shop}}. Just checking you got the estimate from {{Date}} for {{Job}}. Questions are fine. Reply here.

Touch 2, email

Subject: Estimate from {{Date}}: still good to go?

Still holding the {{Job}} estimate at {{Amount}}. Lead time on {{Part}} is about {{Lead}}. If you want the slot week of {{Window}}, reply yes and we lock it. If you went another way, a one-word "no" lets me close it out.

Touch 3

Last note on the {{Date}} estimate. I'll mark it expired on {{Expiry}} unless I hear otherwise. No pressure. I just don't want it sitting in limbo.

Put your real name. Put the real amount. If you would not send it, the agent should not draft it.

How to write a job for an agent

Fill this in before anyone connects a tool. Four jobs use the same skeleton.

1. Name the job in one sentence

"Follow up on sent quotes until reply, win, loss, or expiry."

Not: "Handle sales."

2. Name the feed

Which file, which columns, who updates status. If this is blank, stop.

3. Name the trigger

New row with status sent. Or every morning, rows where next_touch is today and status is still sent.

4. Name the output

A draft SMS, a draft email, or a line on a board. Where it appears. Who sees it.

5. Name the stop rules

Reply, won, lost, expired, do-not-contact, amount over a named threshold needs a human first, customer is also a relative. Whatever is true.

6. Name what it must never do

Never change price. Never offer extra work. Never send without you (until you say otherwise). Never invent a promised date.

7. Name who directs it

You. A named person, not "the office."

We build it. You help hone it for your business. You direct it. Hone means: the first two weeks you edit drafts in red. The agent learns your cuts. Machines handle the retyping. You handle "this one's a maybe, call them."

The other three jobs worth doing

Follow-up is the leak most owners already feel. These three show up right behind it.

Process answers

Staff text you the same questions: late policy, what's included in a cut, how we quote a roof, where the extra keys are.

The job: answer only from a card you approved. Cite the card. If it is not on the card, say "not written down. Ask {{Owner}}."

Do not dump the Drive. Pick the twenty questions you actually get. Write twenty short answers. That is the feed.

Status lookup

The phone rings: is it ready?

The job: look up ticket ID or phone, return stage and promised date, or "unknown." Stages stay a list: received, in progress, waiting, ready, picked up.

If status lives on a whiteboard, photographing the whiteboard is not a system. Type the stage. Then the agent can look it up. A full shop OS is the right buy if you already run one. Do not rip out a working board to impress anyone. An agent on a notebook is for shops that only have the notebook.

Analysis that points attention

Not a chatbot "insight." A list.

• Quotes older than fourteen days, still sent
• Jobs waiting more than X days
• Services that show up together on tickets
• People who used to come in and have not
• Labor versus revenue for last week, next to the week before

The agent points. You decide. No invented benchmarks. Your Tuesday versus your last Tuesday is enough.

What not to automate

Leave these to people:

• Price exceptions and "just this once." The agent has no judgment you would stand behind in the parking lot.
• Angry customers, injuries, anything that could end up in a lawyer's folder. Draft a "we'll call you" if you must. Then call.
• Hiring, firing, payroll disputes.
• First contact with a stranger who did not ask. Follow-up is for people who already got a quote. Cold outreach is a different, often worse, idea.
• Anything you have not run by hand ten times. If you cannot write the steps, you cannot bound the job.
• Sending money, changing bookings, deleting records.
• Medical, legal, or tax advice dressed up as "the system said."

If a mistake would be expensive or cruel, the output is a draft or a flag, never an action.

A week of doing this without software theater

• Day 1. Export sent quotes for the last 30 days. Six fields: name, phone or email, date, amount, status, next touch. Clean that.
• Day 2. Write the three message drafts. Read them out loud. Cut anything you would not say in the doorway.
• Day 3. Touch the five oldest sent quotes yourself. Mark status.
• Day 4. Write the job card (the seven sections above).
• Day 5. Decide draft-first vs auto-send. For almost everyone, draft-first.

You now know whether the leak is "I forget" (an agent can help) or "I hate asking" (an agent will only make you hate the drafts). Both are useful to learn.

When this becomes IBL

A PDF of this page is not an agent. An agent is connected to your quotes, bounded, taught, and directed.

CRUXIBL builds that as IBL, priced in writing. Site-only Lite or Standard does not include it. The website can be a front door later if that is all you need.