AI workflow automation connects a language model to your email, documents and business software, so routine multi-step work such as entering orders or chasing invoices is prepared by the system and approved by a person. With Praxen, a pilot on one workflow takes 2–4 weeks and costs £2,900 (£2,030 with the launch offer). Prices exclude VAT.
What AI adds to business process automation
Classic workflow automation moves data that is already structured, such as a form field into a CRM or a paid invoice into a spreadsheet. It stops when the input is an email, a PDF or a phone call, because a person has to read it first.
AI automation covers that reading step. A language model pulls the details out of the message, checks them against your rules and prepares the next action. The ten examples below are generic workflows found in many businesses. Each one names the point where a person approves.
Examples 1–4: sales and customers
Each example says what starts the workflow, what the system does and where a person approves. The tools named are common examples. The same pattern works with most software that has an API.
- Enquiries into the CRM. Web forms and emails are read, the contact and deal are created in HubSpot or Pipedrive with the need, budget and source filled in, and duplicates are merged. A salesperson checks new deals each morning.
- Quote drafts. The customer’s request, your price list and your template become a draft quote with quantities, delivery dates and terms. The salesperson edits it and sends it.
- Call and meeting notes. A transcript becomes a short summary with actions, owners and dates, saved to the CRM record. The person who held the meeting checks it before anything is shared with the customer.
- Customer email triage. Incoming emails are tagged by topic and urgency, routed to the right person and given a draft reply from your knowledge base. An agent approves each reply.
Examples 5–7: finance and documents
Finance workflows get the strictest approval points, because a mistake here costs money directly.
- Supplier invoices. Invoices arriving by email are read, matched to the purchase order and entered as draft bills in Xero, QuickBooks or Sage. Finance approves each draft, and any request to change a supplier’s bank details goes to a person as a possible fraud.
- Overdue invoice reminders. Each overdue customer gets a reminder in your tone with the invoice attached, on the schedule you set. Replies that dispute the invoice or promise a payment date go to a person.
- Contract checks. Incoming contracts are checked against your checklist: payment terms, renewal and notice dates, liability and termination. The result is a one-page summary with clause references. A person reads the contract and decides.
Examples 8–10: operations and people
These workflows cross several systems and teams, so the log of each run matters as much as the automation itself.
- Order entry. A customer emails an order as a PDF or a spreadsheet. The system matches each line to your product codes and the customer’s account, checks prices against their agreed terms and creates a draft order. Lines it cannot match are flagged with the reason, and a person releases the order.
- Weekly report. Figures from your CRM, accounting software and helpdesk are pulled into one page with short notes on what changed since last week. The manager reads and edits it before it goes to the team.
- New starter onboarding. Once a contract is signed, the system requests accounts, sends documents for signature and tracks the first-week checklist. HR sees what is still outstanding for each new starter.
How to choose the first workflow
Score each candidate against the questions below. A good first project scores well on most of them. A workflow with a high cost of error can follow later, once the first one has shown how the approval points work.
| Question | Good sign | Warning sign |
|---|---|---|
| How often does it happen? | Every day or every week | A few times a year |
| How does the input arrive? | Email, PDF, form or call | In conversation, with no written record |
| Can the rules be written down? | One page of rules covers most cases | It depends on who is doing it |
| Where does the result go? | A system with an API, such as HubSpot or Xero | Paper or a personal spreadsheet |
| What does a mistake cost? | It is caught at the approval step | It reaches a customer or a bank before anyone sees it |
| Who owns it? | One person who can explain it and check the results | Nobody, or a different person each week |
Where people stay in charge
The approval points are designed before the build starts, together with the person who owns the workflow.
- Anything that leaves the business, such as an email, a quote or an order confirmation, is approved by a person until you decide that a specific, low-risk category can go out on its own.
- Payments, refunds, discounts and changes to bank details always need a person.
- When the system is unsure, it stops and sends the case to a person with the reason. You set the confidence level in the pilot, based on tests with your own past cases.
- Every run is logged: what came in, what the system read, what it changed and who approved it.
- Data stays in your own accounts and systems. The AI runs on enterprise APIs that do not train on your data, and we sign a data processing agreement before launch.
What it costs and how long it takes
A pilot on one workflow costs £2,900 (£2,030 with the launch offer) and takes 2–4 weeks, including a test on last month’s real cases. Linking several workflows with AI agents starts at £6,500 (£4,550). Ongoing care is £350 a month (£245). Model usage is charged by the providers at cost and runs on your accounts. Prices exclude VAT.
To check payback, multiply the hours a workflow takes each month by the hourly cost of the people doing it. For example, order entry that takes a coordinator six hours a week at £20 an hour costs about £520 a month in staff time. These figures are illustrative; use your own from a normal week. Then compare the result with the pilot price and the monthly care fee.