1. Home
  2. AI development
  3. Generative AI

Generative AI

Generative AI that drafts the documents your team writes every week

Generative AI writes a first draft from your data and your template. We build it for the documents a business produces over and over: proposals and estimates, job and inspection reports, client letters, call summaries. Names and numbers come from your systems, and a person edits and approves every document before it goes out.

Get the free audit How a draft gets made
  • Prices published on the site
  • First working version in two weeks
  • Support month to month, cancel any time

A first draft in your format, from your own records

Generative AI here means a language model that drafts text. Pointed at one document type, it can turn a CRM deal and a price list into a proposal, a technician's notes into a job report, or a call recording into a summary with action items. Your team edits a draft that already has the facts in it.

The risk is a model that fills a gap with something plausible. So the design inserts facts from your systems: prices, names, dates, addresses and quantities are copied from your records, and the model writes the sentences around them. A person still reads the wording before it goes out. Where a fact is missing, the draft shows a marked blank.

Set up once, checked on every document

Set up

  1. Collect examples

    20 to 50 good examples of the document from your own files. We read them for what every version has in common.

    Week 1

  2. Write the template and rules

    Required fields, tone, and a list of what the model must never invent: prices, dates, names, legal terms.

    Week 1

  3. Connect the data source

    The draft reads from your CRM, price list or job system, so every number comes from your system.

    Weeks 1 to 2

  4. Test on past cases

    Past cases run through the system, and we compare each draft with what your team actually sent.

    Weeks 2 to 3

Every document

  1. A request comes in

    A form, an email, a status change in the CRM or a button click starts the draft.

    Trigger

  2. The model drafts from your data

    It writes the text around the facts it was handed and leaves a marked blank where one is missing.

    Draft

  3. Automatic checks

    Required fields are present, and the numbers and names in the draft match the source record.

    Checks

  4. A person edits and approves

    Your team fixes the draft and sends it. Nothing goes out without that click.

    Person

On a support plan, the corrections your team makes are collected and used to tighten the template and the checks.

Documents we would draft first

Proposals and estimates

Pulls the customer, scope and prices from the CRM and the price list, and writes the cover text and scope description around them.

Job and inspection reports

Turns a technician's checklist and notes into a finished report in your layout, with the findings in the order your clients expect.

Client letters

Drafts the letter for a status update, a delay or a request for documents from the case record, ready for a person to read and send.

Call summaries

Condenses a call transcript into who called, what was decided and what happens next, then files it on the customer record.

Meeting minutes

Writes the minutes and the action list from a transcript, with an owner and a date beside each action.

Policy drafts from notes

Turns a manager's bullet points into a first draft of a policy or procedure, in the structure of your existing ones, for a person to rework.

Which documents are worth drafting with AI

Worth building when

  • The same document is written ten or more times a month from a similar set of facts
  • Those facts already sit in a CRM, a spreadsheet or a form
  • Staff spend more time assembling the document than judging it
  • A person can review a draft in a few minutes

Something simpler will do when

  • The document is a one-off: writing it yourself is faster than building a template
  • It is a regulated filing such as a tax return, a court paper or a permit application, where one wrong word carries legal weight
  • Only fields change and the text never does: a mail merge or a document template fills it without a model, and costs less

Priced by the document type

One process: $2,500 to $6,000, 2 to 4 weeks

One workflow running end to end, or one internal tool.

Connected systems: $6,000 to $15,000, 4 to 8 weeks

Several systems made to work as one: integrations, roles, reports.

One document type with its template, checks and review screen usually lands in one process. Drafts that pull facts from several systems, for example a CRM, a price list and a job tracker, land in connected systems. The cost of the model at small-business volume is estimated in the audit and billed to your own account.

Care $300 to $500 a month. Active $750 to $1,500 a month. AI model usage, GPU time and hardware are paid to the provider, on accounts in your name.

Get the free audit See all prices

Questions about generative AI for documents

Will the AI make up numbers or facts?

It can, which is why the design keeps facts away from it. Numbers, names, dates and addresses are copied from your records, and an automatic check compares them with the source before a person sees the draft. When a fact is missing, the draft shows a marked blank instead of a guess.

How is this different from asking ChatGPT or Claude directly?

A chat tool can keep shared instructions in a project, but someone still pastes in the facts for each document. We build the template, the rules and the data connection once, so the draft appears in the tool your team already uses, with the same structure every time and the numbers filled from your system. Staff can still use a chat tool for one-off writing.

Which model writes the drafts?

Usually Claude or Gemini, or an open-weight model such as Llama or Qwen on your own server when the text cannot leave. We test the candidates on your past cases and tell you which model the system calls and what data it sees. Evaluation covers how we compare them.

How many examples do you need?

We start with 20 to 50 good examples of the document. Short, uniform documents work with fewer; a document that comes in many variants benefits from more. We also need read access to the system the facts come from.

Can it write in our voice?

Yes, for business documents. The tone comes from your examples and a short style sheet we write with you. If a prompt and examples cannot hold your format, fine-tuning is the next step, and we propose it only after testing shows the need.

Who owns the templates and the system?

Accounts and code repositories are opened in your company's name, and the custom code and templates we write are assigned to you under the contract once paid for; and the system can run on your own server. If it calls an outside AI model, we tell you which one and what data it sees before we build.

Related services

Tell us what's slowing you down.

You get a free written review by email within two business days.

Only your email and the task are required. No passwords or client data needed.

No mobile information will be shared with third parties or affiliates for marketing or promotional purposes.