Skip to content
Softgenix

AI Workflows · Workflow Proof £350

AI workflows that do the reading, sorting and drafting.

A lot of office work is reading something and deciding where it goes: an invoice, an enquiry, a delivery note, a claim form. Language models are now good at that reading. We build workflows that use them for it, connect the result to the systems you already use, and keep a person in the loop for anything that needs judgement.
Tell us your workflowHow the £350 proof works
Example: a website enquiry, from inbox to CRM
  1. Ordinary code: Enquiry arrivesWeb form or email, any wording
  2. AI step: AI reads and sorts itType, urgency, product, postcode
  3. Ordinary code: Rules check itIn your service area? Existing customer?
  4. AI step: AI drafts a replyFrom your templates and price list
  5. A person: A person approvesEdits if needed, then sends
  6. Ordinary code: Logged in the CRMContact, enquiry and reply, once
  • AI step
  • Ordinary code
  • A person

Why it matters

Worth doing properly

Most AI pilots stall in the same place. The demo looks impressive, but nobody can say how often it is wrong, what happens when it is, or how it fits into the systems the team actually uses. We start from one workflow, test it on your real examples, and show you how often each field comes out right before anything goes live.

“Our team spends hours reading emails, documents and forms, then typing what they say into other systems.”

The kind of message that starts these projects

What we build

Typical jobs

Document extraction

Invoices, purchase orders, delivery notes, contracts and forms read into structured fields, checked against your rules and posted to the right system.

Inbox and enquiry triage

Emails and web enquiries classified by type and urgency and routed to the right person or queue, with the key details already extracted.

Drafting with approval

Replies, quotes, summaries and reports drafted from your own data and templates, then approved by a person before anything is sent.

Answers from your own documents

Staff ask a question and get an answer drawn from your policies, product data or past projects, with links to the source so they can check it.

Matching and clean-up

Supplier names, addresses and product descriptions matched and tidied so the same thing is recorded the same way in every system.

Review queues

A simple screen where staff see what the AI decided and why, and approve or correct it. Corrections are kept, so accuracy is measured rather than guessed.

Process

How it runs

  1. Pick one workflow

    We look at the process, the systems involved and a sample of real examples, and agree what a correct result looks like for each field.

  2. Test it on your examples

    A working prototype runs over the sample and we report how often it is right, field by field, before you commit to a full build.

  3. Build it into your systems

    Connections to your inbox, storage and business software, with rules, retries, logging and a review queue for anything uncertain.

  4. Run it and keep measuring

    We watch accuracy and running cost in production and adjust prompts or rules as your documents and requests change.

Deliverables

What you end up with

Code, accounts and infrastructure are in your name from the start.

  • A working workflow connected to the systems you already use
  • An accuracy report from testing on your own examples
  • A review queue for results the system is unsure about
  • A log of every item processed and every decision made
  • Running cost per item, so you can compare it with doing it by hand
  • Code, prompts and accounts in your name

FAQ

Common questions

Which AI models do you use?

We choose per task. Usually that means models from OpenAI, Anthropic or Google through their business APIs, and sometimes a smaller open model where cost or data location matters more. Workflows are built so the model can be changed later without a rebuild.

Is our data used to train AI models?

That depends on the provider and the account terms, so we don't give a blanket answer. Each proposal names the providers that will process your data, their terms on training and retention, and how long anything is kept, and we confirm those settings before any of your data is processed.

What happens when the AI gets something wrong?

It will, sometimes, so the workflow is designed around that. Rules check the output, uncertain items go to a person, and nothing involving money or customers goes out without the checks you agree. Every correction is logged, so you can see the error rate.

Do we need to replace our current software?

No. In most cases the workflow sits between systems you already have, such as your inbox, Xero, your CRM or SharePoint.

Worth reading first

Also from Softgenix

Test it on your own examples first.

Tell us what comes in, where it ends up and who touches it on the way. We will reply with the scope and sample we'd propose for a £350 Workflow Proof.

Tell us your workflowWhat's included