AI agents · Small businesses

AI automation for small businesses: the agent drafts, you approve

For a small business, AI pays off when an agent takes one repeating workflow, such as quotes, supplier invoices or order-status questions, prepares the work and leaves the decision to you.

Discovery is a fixed fee, credited to your first months of monthly care

For
Machine shops, metal workshops, subcontractors and distributors in Israel
The agent
Drafts quotes, invoice entries and follow-ups from your own past jobs; you approve in one tap
First step
A Discovery of one workflow, credited toward your first months of monthly care

Where the time goes

Where do the hours go in a small business?

Five places where owners and office managers spend their week, and what an agent could take off their desk.

  • Every quote eats part of the day

    Read the drawing, find a similar job, work it out, while the customer waits. A slow reply loses the job. An agent drafts the quote; you approve it.

  • Quotes sent, and nobody followed up

    There's no time to chase every open quote. An agent reminds the customer on time and records whether the job was won or lost, and why.

  • Order-status questions

    Calls and messages that break up the day. An agent answers from the job sheet and passes you only the orders that are running late.

  • Supplier invoices typed in by hand

    Copying lines and checking them against the order, one by one. An agent extracts the lines, matches them to the order and asks you when something doesn't add up.

  • Chasing material prices from suppliers

    Calls and emails to 2–3 suppliers for every material order. An agent sends the requests and compares the answers; you decide.

Quoting

RFQ-to-quote automation: from request to approved quote

The agent reads the request and the drawing, finds similar past jobs, estimates material, setup and machine time from your rates, and drafts the quote. You approve, edit or reject it before anything is sent.

Illustrative walk-through

Today

The owner, by hand, often in the evening

  1. Reads the request and opens the drawing
  2. Searches old folders for similar past jobs
  3. Works out material, setup and machine time
  4. Writes the quote and sends it when there's time
  5. Remembers, or doesn't, to follow up days later

With the agent

  1. AgentPDFReads the email and the drawing: material, dimensions, tolerances, finish, quantity
  2. AgentFinds similar past jobs and the prices you quoted for them
  3. AgentPDFEstimates material, setup and machine time from your rates; drafts the quote with reasoning
  4. YouYou approveYou approve, edit or reject it in one tap
  5. AgentEmailThe quote is sent and logged, and a follow-up reminder is set
Who drafts the quote
Today: The owner, from scratch
With the agent: The agent drafts; the owner approves
When the customer hears
Today: When there's time, sometimes days later
With the agent: As soon as you approve the draft
What's kept for follow-up
Today: Whatever the owner remembers
With the agent: Every quote logged, follow-up set automatically

Control and privacy

How does the agent protect your drawings and prices?

Illustrative scenario, not a client project: a CNC machine shop doing turning and milling. This is how a quoting agent is built, and what it never does.

  • Drawings can stay with you: the agent runs on a local model on your own server, or uses cloud models under terms that don't let the provider train on your data.
  • No quote and no price goes to a customer until you've approved it.
  • The agent shows how it reached every figure, and flags fields it isn't sure about for you to check.
  • During Discovery, 50–100 past jobs are cleaned up to serve as the reference set and the test set.

Around the quote itself, the agent can take on smaller workflows too: following up a quote that got no answer, with a record of whether the job was won and why; answering order-status questions from the job sheet; taking in supplier invoices against the order; and asking several suppliers for material prices, while you decide who to order from.

The framework is chosen per workflow, for example Dify for a simple document flow, or LangGraph for a longer flow with several stages. An agent can connect to, for example, email, Excel or Google Sheets, WhatsApp Business and your accounting software's API. These are examples of systems that can be connected, and feasibility is checked during Discovery.

Conceptech also builds machines and test stations and works with mechanical drawings, so we read a drawing the way production people do. Industrial automation and custom machines

Illustrative scenarios

Illustrative scenarios for small businesses

Two scenarios based on common processes: a workflow built around one business, and a ready-made package that suits many similar businesses.

  • Illustrative example

    Small CNC machine shop

    Quotes from drawings, approved by the owner

    Every request for a quote means reading a drawing, finding similar jobs and calculating, usually in the evening. A slow reply loses the job.

    1. RFQ arrives by email
    2. Drawing read
    3. Similar past jobs
    4. Draft quote
    5. Owner approves(human approval)
    6. Sent, follow-up set

    The aim: Quotes go out the same day instead of after several days, and no price is sent without approval

    Typical plan: Growth

  • Illustrative exampleTemplated package

    Café, single branch

    Daily supplier order from till data

    The owner puts the supplier orders together at the end of a shift, from memory and scraps of paper.

    1. Sales from the till
    2. Current stock
    3. Draft supplier order
    4. Owner approves(human approval)
    5. Order sent

    The aim: Orders go out on time without an evening at the desk, with fewer shortages and less overstock

    Typical plan: Templated package

Illustrative scenarios based on typical processes, not client projects.

What a Discovery includes

More workflows

What else can an AI agent take on in a small business?

Any of these can be the first or the second workflow. In each one the agent prepares the work, and you approve what matters.

  • Supplier-invoice automation

    The agent extracts the lines, matches them to the order and delivery note, and drafts the entry. On a mismatch it stops and asks you.

  • Following up unanswered quotes

    After a period you set, the agent drafts a reminder to the customer and records whether the job was won, and why. You approve the wording.

  • Order-status answers

    The agent answers the customer from the job sheet, and passes you every order that's running late and every question about a new date.

  • Comparing supplier quotes

    The agent builds the material list, asks several suppliers and compares price and delivery date. You decide who to order from.

  • Taking in and routing requests

    The agent classifies a new request, extracts the details and routes it to the right person with a draft reply for approval.

What fits

AI agent, ready-made tool or chatbot: which fits you?

Sometimes a ready-made tool is enough, and then it's the better choice. This comparison helps you decide before you invest.

  • A ready-made SaaS tool: quick to start, and a good fit when your workflow looks like everyone else's. It works from a generic model, not from your past jobs and rates, and your data sits with the vendor.
  • Rule-based automation, such as n8n, Make or Zapier: excellent for the fixed parts, like moving data between systems. It can't read a drawing or handle an unstructured email.
  • A WhatsApp or chat bot: answers common questions. It doesn't draft a quote from a drawing or work with your documents.
  • An agent built on your own data: reads your drawings and documents, works from your past jobs and rates, can run on your own server, and you approve every key step. The right choice when the workflow is specific to your business and takes real time.

Often the answer is a mix: automation for the fixed parts, and an agent for the parts that need judgement. And if a ready-made tool is enough for you, we'll say so during Discovery.

Method

How do we start, and what happens until the agent is live?

Five steps, each with a gate. In the first period nothing reaches a customer without you seeing it, and autonomy grows only on criteria we agreed in writing.

  1. 01

    You start here

    Discovery

    We map one process using 20–100 real examples, design the agent's flow and estimate the return on investment.

    No agent yet

    Gate: You decide whether to go ahead with setup

  2. 02

    Baseline

    We measure how long each quote or invoice takes today, how many there are each week, the error rate and how long customers wait for an answer.

    No agent yet

    Gate: You approve the baseline and the targets

  3. 03

    Build

    We build one flow at a time, in a test setup. Anything irreversible gets an approval step, and every action the agent takes is logged.

    Agent drafts in a test setup

    Gate: The agent passes the test set on real past jobs

  4. 04

    Shadow mode

    The agent drafts quotes alongside you, but nothing is sent to a customer. Each draft is compared with the quote you actually sent, for a period set in the Discovery report.

    Runs alongside your team, sends nothing

    Gate: Written criteria, for example: at least 70% of drafts approved without edits, and zero critical errors

  5. 05

    Go-live

    The agent drafts, sends after approval and follows up automatically, and you approve every price. Autonomy widens only while most drafts are approved without edits.

    Agent acts; you approve key steps

What you get at the end of each step
Discovery
Discovery report: process map, flow design, ROI estimate and a fixed quote
Baseline
Baseline figures the agent is measured against
Build
A working agent in a test setup, with an action log
Shadow mode
A side-by-side comparison of agent drafts and your quotes
Go-live
Monthly report: tasks done, time saved, errors caught at approval

Pricing

How is an AI agent priced for a small business?

Per workflow, not by the hour: a fixed price for a defined result, agreed in writing up front.

Each workflow has a one-time setup and a fixed monthly care fee. Monthly care covers hosting and running the agent, monitoring, prompt and model updates, and small changes. You pay the language-model usage directly to the provider, on your own account, so you see every charge.

You get the fixed price for the workflow in writing after the Discovery, before any build work. The Discovery fee is credited toward your first months of monthly care if you go ahead with setup.

What an AI agent costs a small business

Discovery

What do you get from a Discovery?

A fixed-fee, fixed-scope first step: one process, one report you can act on, and the fee is credited toward your first months of monthly care if you continue.

What you bring

  • One process owner who knows the work and can go through examples with us
  • 20–100 real past examples, such as emails, quotes or invoices, sanitised if needed
  • Read access to, or exports from, the documents and systems the process uses
Fee
Fixed and agreed up front, in a written proposal
Credit
Credited toward your first months of care, if you go ahead with setup
Data
NDA on request; examples can be sanitised
Monthly care
Cancel any time with one month's notice, no minimum period

Discovery is a fixed fee, credited to your first months of monthly care

Discovery report
  1. 01Process mapHow the process runs today, step by step: who does what, in which systems
  2. 02BaselineTime per task, weekly volume, errors and response time to customers
  3. 03Agent flow designSteps, approval points, systems, model, and whether it runs in the cloud or locally
  4. 04ROI estimateHow much time the agent is expected to save, against the cost of setup and care
  5. 05Draft go-live criteriaWhat must hold in shadow mode before the agent starts acting
  6. 06Fixed quote for setup and careA fixed price for the workflow, set before any build work starts

AI for small businesses: FAQ

Does AI make sense for a small business with a small team?

Yes, when there's a workflow that repeats every week and has past examples, such as quotes or supplier invoices. You start with one workflow, not the whole business. The Discovery checks whether the workflow is worth an agent, and if a ready-made tool is enough, it says so.

AI agents for larger manufacturers

Can the agent read PDF drawings?

In most cases, yes. Models that read images and documents extract material, dimensions, tolerances, finish and quantity from a drawing, and fields the agent isn't sure about are flagged for you to check. Discovery tests this on your real drawings, before you commit to setup.

Who decides the price in the quote?

You do. The agent drafts from your rates and similar past jobs, and shows how it reached every figure. No quote is sent until you've approved, edited or rejected it.

Our customers' drawings are confidential. Where do they go?

They don't have to leave your business at all: the agent can run on a local model on your own server. Where cloud models fit, we use them under terms that don't let the provider train on your data. We sign an NDA on request when the project calls for it.

Do we need to replace our software?

No. An agent can connect to what you already use, such as email, Excel or Google Sheets, a shared folder, your accounting software's API or WhatsApp Business. These are examples, not a list of past projects, and feasibility for your systems is checked during Discovery.

Connecting to Priority and accounting systems

How does supplier-invoice automation work?

The agent reads an invoice that arrives by email, extracts the lines and matches them to the order. When everything matches, it drafts the accounting entry. When the price, quantity or item differs, it stops and asks for your approval.

How much does an AI agent cost a small business?

It's priced per workflow: a one-time setup plus monthly care, and you pay the model usage directly. The amount depends mainly on the number of systems, the state of your data and the approval flow, and it's set in writing after the Discovery.

What makes up the cost of an AI agent

Our quote history is messy. Is that a problem?

It's common, and it doesn't stop you from starting. During Discovery, 50–100 past jobs are cleaned up: what was asked, what was quoted and what was won. They serve the agent as its reference set, and serve us as the test set before go-live.

Pick the workflow that tires you most, and start with its Discovery

Tell us which workflow takes most of your time. An engineer will reply, and we'll check together whether it suits an agent. The Discovery fee is credited toward your monthly care.

Usually a reply within 3 working hours, by email or phone

Talk to an engineerWhatsApp(opens in a new tab)