Add two Airtable field agents (the current name for AI fields) to your quote-request table: a long-text agent that summarises each request and a single-select agent that tags it from your own list. Airtable publishes no fixed rate per run, so read each agent's credit cost in its Preview; the Team plan includes 15,000 credits per paid seat a month.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowNo run has a figure you can look up in advance. Airtable prices every run on the text going in, the model chosen and the text coming out, so it publishes examples rather than a rate card. The agent's Preview window shows the credit cost for the first record in your view: read it before you switch the agent on. For a quote inbox the credits rarely bite. Tags nobody checked do.
The quote-request table, before any AI goes in
An agent can only summarise what the record holds, so the plain fields come first. Here is the table for an illustrative private dental practice that takes treatment quote requests through an Airtable form on its website:
| Field | Type | Filled by |
|---|---|---|
| Name, Email, Phone | Single line text, Email, Phone number | The form |
| Treatment asked about | Single select (the patient's own choice) | The form |
| Request details | Long text | The form |
| Received | Created time | Airtable |
| AI summary | Field agent, long text | Agent 1 |
| AI treatment tag | Field agent, single select | Agent 2 |
| AI priority | Field agent, single select | Agent 3 |
| Tag checked | Checkbox | Treatment coordinator |
| Staff notes | Long text, read by no agent | Staff |
| Status | Single select: New, Quoted, Booked, Lost | Staff |
Three design choices here save trouble later. First, keep the patient's own "Treatment asked about" choice even though an agent will tag the request. Patients pick whitening when they mean veneers, or "other" when they mean an implant, and the gap between their choice and the agent's tag is worth seeing. A formula field such as IF({Treatment asked about} != {AI treatment tag}, "Compare", "") flags those records for a second look.
Second, give staff a separate Staff notes field that no agent reads. With Run automatically on, an agent re-runs whenever a field it reads changes, so a receptionist typing "rang back, no answer" into Request details would trigger a fresh summary, a fresh tag and a fresh charge, and the summary would start describing your phone calls instead of the patient's request.
Third, AI fields can't appear on Airtable forms, and they don't need to. The form fills the input fields; once the record exists, an agent set to Run automatically fills its own cell. Submit one test request through the live form before trusting it, because that is the path every real request will take. If you're still choosing the intake tool itself, Typeform or Airtable for client intake covers that decision.
Health details turn up in these forms whether you ask for them or not. Ask only what a quote needs, and check your workspace's AI settings and your own data-protection obligations before any agent reads the records.
How do you build an Airtable agent that summarises a quote request?
Building a summary field agent in Airtable takes about ten minutes, in these steps:
- Click the + button next to the last field. In the Field agents section, choose Build a field agent.
- The field type defaults to Long text, which is right for a summary. Write the instructions, inserting fields with + Insert field or by typing { and the field name, so the reference survives if someone renames the field.
- Open the Model dropdown and start with a model marked Low cost. Airtable notes that lower-cost models are weaker at arithmetic and complex tool use; a three-line summary needs neither.
- Read the Preview under the instructions, including its credit figure for the first record, then click Create field.
- Under Settings, turn on Run automatically and click Save. Until you do, the agent only runs when someone triggers it.
These are the instructions I'd start with:
You are summarising a dental treatment quote request for the practice's
treatment coordinator. Read {Request details} and {Treatment asked about}.
Write exactly three short lines:
1. What the patient wants and why, in their own terms.
2. Anything that changes the quote: number of teeth, previous treatment,
budget, payment-plan question, deadline, days they can attend.
3. Which of these the request does not tell us: last check-up date,
recent X-rays, medical history form, preferred days.
Write "not stated" for anything missing. Do not give clinical advice,
a diagnosis or a price. Do not add anything that is not in the request.
Here is an illustrative request, the summary a first draft of the prompt produced, and the fix:
Request details: "Hi, I lost a lower back tooth about two years ago and the gap catches food. I'd like to know about an implant and roughly what it costs, and whether you do payment plans. I was quoted a bridge elsewhere but don't want the teeth either side filed down. I can only come in on Tuesdays or Fridays."
1. Wants an implant for one missing lower back tooth (gap for about two
years, catches food).
2. Prefers not to have a bridge; asks about payment plans; no budget
stated; available Tuesdays and Fridays.
3. Needs an implant assessment and X-ray before quoting; not stated
whether there is bone loss or a recent check-up.
Lines 1 and 2 are exactly what a coordinator needs. Line 3 is the problem: "bone loss" is a clinical idea the patient never mentioned, and "needs an implant assessment" is the AI deciding treatment. That is why the prompt above restricts line 3 to a fixed checklist of missing information. Rerun the Preview after each prompt change and read five or six different records, not just the first.
How do you tag quote requests from a list you control?
To tag quote requests from a list you control, build a second field agent, change the type to Single select and set the options yourself before the first run. Airtable's help page says select agents choose options based on other fields; it doesn't say they invent new ones, so your list is the whole vocabulary. A filled-in option list, with the definitions that go into the instructions:
| Option | Choose it when the patient... | Typical wording |
|---|---|---|
| Emergency or pain | mentions pain, swelling, bleeding that won't stop, or a broken tooth that hurts | "throbbing since Sunday" |
| Implants | wants a missing tooth replaced and mentions implants or asks what replaces a gap | "gap where my tooth was" |
| Clear aligners | wants teeth straightened without fixed braces | "straighter, but nothing visible" |
| Whitening | is only concerned with colour | "yellow, want them brighter" |
| Veneers | wants the shape or colour of front teeth changed with covers | "chipped front teeth, want them even" |
| Crowns and bridges | has a crown or bridge problem, or asks for one by name | "my crown came off" |
| Check-up and hygiene | wants an examination, cleaning or registration | "overdue a check-up" |
| Other | fits none of the above | anything else |
Choose the one option that best matches what the patient is asking for
in {Request details}. Apply these rules in order:
1. If they mention pain, swelling, bleeding that won't stop, a cracked or
broken tooth, or being unable to chew, choose "Emergency or pain",
whatever else they ask about.
2. If they ask about more than one treatment, choose the first one they ask about.
3. "Veneers" is only for front teeth and appearance. A crown that came
off or broke is "Crowns and bridges".
4. If nothing fits, choose "Other". Do not decide from
{Treatment asked about} alone.
The third agent, AI priority, is a second single select with three options: Same day, This week and Routine. Its rules are short: Same day only when the treatment tag is Emergency or pain; This week when the patient names a date within 30 days, such as a wedding; Routine otherwise. Treat both tags as routing, not clinical judgement. Anything marked Same day goes to a person at once, and a clinician decides what it is.
The pattern transfers to other sectors by swapping the list. A physiotherapy clinic quoting employers for on-site sessions might use Desk assessments, On-site clinic days, Injury rehab blocks, Group exercise classes and Other, with the pain rule replaced by "wants to start within two weeks".
When do Airtable field agents re-run, and what do they overwrite?
With Run automatically on, an Airtable field agent runs for new records and again when an input field changes, and it does not overwrite cells a person has edited. The re-run rules decide both your credit bill and whether staff corrections survive, so they are worth knowing before the first busy week:
- With Run automatically on, an agent runs for new records and again whenever one of its input fields changes. Airtable's help page is plain that every agent run consumes credits.
- Editing an agent's instructions does not regenerate cells that already have content. If Run automatically is on, saving the change makes Airtable ask whether to overwrite existing records; you can save without overwriting.
- Airtable never overwrites a cell that a person has edited, even with Run automatically on. Clearing a cell counts as an edit, so a cleared cell stays empty until someone runs the agent on it again.
- Only Editors and above can trigger runs, and only Owners and Creators can change an agent's instructions or model.
For bulk runs, open the agent's field menu and hover over Run field agent. The five choices, and when each one earns its credits:
| Run field agent option | Use it when |
|---|---|
| All cells in the view | Filling a filtered view for the first time, such as last month's requests. Costs one run per record in the view, so filter first. |
| Cells modified by agent | Refreshing AI output after a better prompt, while leaving cells staff corrected by hand alone |
| Outdated cells | Catching up on records whose output Airtable marks as outdated, for instance after inputs changed while automatic runs were off |
| Cells with errors | After a failed batch, such as running out of credits mid-month |
| Cells never modified | Records that have never had a value, typically those created before the agent existed |
There is also a ceiling on length: prompt plus response is limited to about 12,000 words on lower-powered models and about 90,000 on higher-powered ones, and longer responses are cut off. A quote request never comes close; a pasted email thread with a 40-page attachment might.
How many credits do 400 quote requests a month use?
On the page's illustrative figures, 400 quote requests a month use 2,400 credits before re-runs, against 60,000 pooled credits on a four-seat Team plan.
Take the illustrative dental practice above on the Team plan with four paid seats: the practice manager, the treatment coordinator and two receptionists. At $20 per seat a month billed annually that is $80 a month ($96 billed monthly at $24 a seat). Airtable pools Team credits across the workspace, so the base draws on 4 × 15,000 = 60,000 credits a month.
Suppose the Preview shows 4 credits for the summary on a low-cost model and 1 credit each for the two tags; those figures are illustrative, so measure your own. The month then looks like this:
| Item | Sum | Credits |
|---|---|---|
| New requests | 400 × (4 + 1 + 1) | 2,400 |
| Re-runs after staff fix typos in Request details (1 in 5 records) | 80 × 6 | 480 |
| Refresh after a prompt rewrite in week three (Cells modified by agent) | 300 × 4 | 1,200 |
| Month total | 4,080 | |
| Share of the 60,000 pool | 6.8% |
The pool is shared with every other base in the workspace, so a busy inventory base or a lot of Omni questions (each Omni question uses credits too) draws from the same 60,000. To estimate the whole workspace rather than one table, use the method in estimating what Airtable AI fields cost each month.
Three settings move the bill far more than volume does. These scenario rows are illustrative, except where Airtable publishes the figure:
| Scenario | Per request | 400 requests |
|---|---|---|
| Low-cost model, text only (the table above) | about 6 credits | 2,400 |
| Summary moved to a higher-powered model costing, say, 12 credits a run | about 14 | 5,600 |
| Internet access left on for the summary (Airtable's web-search example: 10 a run) | about 16 | 6,400 |
| An attachment-analysis agent reading uploaded letters (Airtable's example: about 200 credits for a 10-page contract) | tens to hundreds | could pass 20,000 |
At four seats even the attachment row fits. The sum changes for a one-seat workspace, which has 15,000 credits: 400 requests with attachment analysis could use the lot. Self-serve add-on packs start at $20 a month for 10,000 credits ($200 a year), with 20,000 for $40 and 50,000 for $100. Before buying one, ask whether the agent needs the attachment at all; a coordinator can open a letter in seconds. Team workspaces get an email alert when usage passes 75% of the limit, which is your cue to look at the scenarios above.
Image attachments add a plan question as well: Airtable's help page says extracting from PNG, JPG and WebP images needs Business or Enterprise Scale with Airtable Labs switched on. If you're weighing that upgrade, Airtable Team versus Business sets out what else changes.
If you want to run credit sums like these in a spreadsheet, try The Complete Excel Course with AI Integration on Coding Liquids.
Why add a tick-box gate before the tags drive anything?
A person should check each tag before it sends emails, assigns staff or feeds a report, and the simplest gate is a Tag checked checkbox. Agents that tag are the easy part; agents whose tags drive actions need a person in between until you've seen how accurate they are. The setup:
- Create a view called To check, filtered to Tag checked is unticked and sorted by AI priority, so Same day requests sit at the top.
- The coordinator reads the summary, glances at the request, corrects either tag by picking the right option, and ticks Tag checked.
- An automation with the When record matches conditions trigger watches for Tag checked being ticked, then routes the record: Send email to the practice manager for Emergency or pain, a different action for everything else.
Airtable's own guidance on that trigger explains the checkbox. The trigger fires every time a record changes into the matching state, and a condition spread over several fields can fire before someone has finished editing. A single ready checkbox changes once, so the automation runs once, at the moment a person has looked. Airtable also advises against multiple select fields in these conditions, because adding a second value can put a record back into the matching state. The Team plan includes 25,000 automation runs a month, so one run per request costs nothing worth counting. More patterns for this kind of routing are in six Airtable automations a small business can copy.
After the first fortnight, audit 50 records properly. An illustrative result:
| What the audit found | Records | Change made |
|---|---|---|
| Treatment tag right first time | 45 of 50 | None |
| Veneers chosen for a loose crown | 3 | Added rule 3 (front teeth and appearance only) |
| "Cracked tooth, chewing on the other side" tagged Implants | 1 | Added "cracked" and "unable to chew" to rule 1 |
| Two treatments, the second one chosen | 1 | Added rule 2 (the first one asked about) |
| Priority wrong | 2 of 50 | Both followed a wrong treatment tag; fixed by the rules above |
At 45 of 50 the gate stays; the one missed emergency is reason enough. When a later audit of 50 shows no missed Emergency or pain tags and few other corrections, you can let routine tags route without the tick and keep the gate for Same day only.
Where the tagging goes wrong in the first month
The cleared cell that never refilled
A receptionist sees "Whitening" on a request that is plainly about aligners, clears the cell meaning to come back to it, and moves on. Because clearing counts as a human edit, the agent never refills it, and the record drops out of every view filtered by treatment. Nobody noticed until a patient emailed nine days later asking whether anyone had read their request. The fixes are a rule (pick the right option instead of clearing, or run the agent on that one cell) and a safety-net view of records where AI treatment tag is empty and Tag checked is unticked.
Two options that overlap
Most wrong tags in the audit came from Veneers and Crowns and bridges overlapping. Agents don't share your team's unwritten definitions, so write them into the instructions, as rule 3 does. If two options keep swapping, the list may be wrong rather than the agent: merge them, or split them by a question patients can answer.
Internet access left switched on
When the internet access setting is on, Airtable evaluates whether each prompt needs the web, and its help page warns this often uses extra credits. A summary of a patient's own words never needs it, and web content has no business in it. Turn it off for every agent in this table.
A quiet model upgrade
Someone with Creator access switches the summary to a higher-powered model because one summary read badly. Every future run now costs more, and nobody notices until the 75% alert arrives. Make model changes a decision: note the Preview's credit figure before and after, and record why in the field description. Who should hold Creator rights at all is covered in the sibling tutorial on Airtable permission levels.
Drafting the reply is a separate job
Once requests arrive summarised and tagged, it is tempting to add a fourth agent that drafts the reply. It can work, but it changes the risk: a summary is read by one coordinator, while a reply goes to a patient and can promise something the practice doesn't offer or quote a price nobody set. If you add one, keep it to a holding reply ("thank you, our treatment coordinator will call you on Tuesday or Friday"), never let it state prices, and put an approval step in front of sending, as described in setting up an approval step for AI-written quotes.
If you want help fitting a setup like this into how your business actually runs, I offer one-to-one AI consultations for small businesses.
Airtable AI fields for quote requests: follow-up questions
Can a field agent add a new tag option when nothing on my list fits?
Airtable's help page does not say it can: it says select agents choose single or multiple select options based on other fields, and says nothing about creating new options, so treat your option list as the full vocabulary. Include an Other option, review what lands in it each month, and add a proper option when the same kind of request turns up several times.
Do read-only staff use up AI credits by opening records?
Viewing existing AI output doesn't run an agent. Credits are used when an agent runs, and only Editors and above can trigger runs. Airtable's billing page adds that Viewers get no allocated credits of their own but can use the workspace's credits through Omni, the built-in assistant, so a busy Omni user can still draw from the pool.
What happens to new quote requests when the month's credits run out?
Agents stop generating until the credits reset at the start of your billing cycle or you buy a pack, from $20 a month for 10,000 credits on self-serve plans. Requests still arrive through the form; they just lack a summary and tag. Keep a view of records with an empty AI tag so nothing slips through while you wait.
Should the summary agent also draft the reply to the patient?
Keep them separate. The summary is internal and read by one person; a reply goes to a patient and should pass an approval step. A third agent can draft a holding reply, but have someone approve every draft before it is sent, and never let it state a price the practice hasn't set.
Further reads
- How to Set Up a Simple CRM in Airtable for a Service Business — Build the wider client tables your quote requests will feed into.
- Airtable Permission Levels: Which Role Should Staff Get? — Decide who may edit agent instructions and who only runs them.
- How to Build an Airtable Inventory Tracker With AI Reorder Notes — Another field-agent build: low-stock flags and reorder notes.
- How to Create Quotes and Estimates in Minutes With AI — Turn the tagged request into a quote faster once it is triaged.
- Airtable vs Google Sheets for Business Data AI Can Use — Check Airtable is the right home for this data in the first place.
- Typeform AI Form Builder: A Quote Request Form for a Fencing Firm — A step-by-step fencing quote request form built with Typeform AI: the prompt, branching questions, photo uploads, estimator alerts and what each plan allows.
Show 2 more related tutorials
- How to Set Up Airtable With AI for Startup Operations — One Airtable base for customers, tasks and invoices, with three AI fields and a plan-limit check before you commit.
- ChatGPT Gmail Integration: Set Up Draft-Only Quote Replies — A translation agency's draft-only quote replies: one event-triggered ChatGPT Work task, the Gmail permission that blocks sending, and a test plan.
Browse all AI-for-business tutorials
Sources: Airtable help pages on using Airtable AI in fields (field agents), Airtable AI billing, the When record matches conditions trigger and custom notifications with automations; Airtable pricing page. All checked 17 Nov 2026.
