Enquiry triage and reply
Read incoming enquiries, classify by service and urgency, draft a reply with the right questions, and route anything unusual to a named person. Response time drops from hours to minutes without anyone watching an inbox.
Agentic automation
Agentic automation that knows when to ask a person
Software agents are good at the repetitive middle of a process: reading, checking, drafting, chasing and filing. They are poor at judgement. We build workflows that use each for what it is good at.
An agentic workflow is a process where software takes a sequence of steps on its own — reading an enquiry, checking availability, drafting a quote, booking a slot, chasing a payment — and escalates to a person when the case falls outside its instructions. It is not a chatbot bolted onto a website. It is the plumbing between the tools you already pay for.
The value is rarely in the clever part. It is in removing the retyping, the waiting for someone to check an inbox, and the customer who goes elsewhere because nobody replied until Thursday. Those are unglamorous wins, and they are the ones that change how a business feels to run.
Use cases
Where we usually start
Four workflows that pay for themselves quickly in a small or mid-sized business.
Quoting and follow-up
Turn a site visit or a completed form into a structured quote from your own price book, send it, and follow up on a schedule. Margin stays visible because the agent uses the same rules every time.
Onboarding and compliance
Collect the documents you need, check they are complete and in date, chase the missing ones, and file everything in the right place. Exceptions go to a person with the context attached.
Invoicing and reconciliation
Raise invoices from completed work, match payments, flag part-payments and overdue accounts, and produce a weekly list of what needs a phone call rather than another email.
Guardrails
What every workflow we build includes
The difference between an agent and a liability is the boundary around it.
- A written scopeWhat the agent may do, what it must not do, and what it must escalate. Written for the team, not for a developer.
- Approval pointsA person approves anything that commits money, makes a promise or affects a customer's rights.
- An audit trailEvery input, suggestion, edit and send is recorded with a timestamp and an owner.
- Failure behaviourIf a system is down or a model is uncertain, the workflow stops safely and tells someone — it does not guess.
- A kill switchOne action disables the automation and returns the process to manual, without losing the queue.
- A measured baselineThe current time, cost and error rate, recorded before anything changes.
Before and after
What changes in practice
Qualitative, because we will not invent a percentage for a process we have not measured in your business.
| Stage | Typical today | With a well-built workflow |
|---|---|---|
| Enquiry arrives | Sits in a shared inbox until someone checks | Classified and acknowledged within minutes |
| Information gathering | Two or three emails to get the basics | One structured reply asking for exactly what is missing |
| Quote | Written from memory at the end of the day | Drafted from your price book, reviewed and sent |
| Follow-up | Happens when someone remembers | Scheduled, logged, stopped when a reply arrives |
| Exception | Absorbed silently or dropped | Escalated with context to a named person |
Automation is not the same as AI
Automation is not the same as AI
- Read the AI implementation pageWhen a model is the right tool, and when ordinary software is cheaper and better.
- See the governance approachData boundaries, approvals and the audit trail.
- Read a reference buildHow a structured quoting workflow is put together.
Questions
The things people ask first
Is this an AI chatbot?
Usually not. A chatbot is one interface to one model. An agentic workflow is a sequence of steps across several systems, only some of which use a model. The customer may never see the AI at all.
Will it work with our existing CRM or practice software?
If it has an API, a webhook or a usable export, yes. If it does not, we can often work through email or a shared database. The one thing we will not do is pretend an integration exists when it does not.
What happens when the model is unavailable?
The workflow falls back to manual and tells the team. It does not queue silently or invent an answer. Availability and fallback are part of the design, not an afterthought.
How do you stop it sending something embarrassing?
Anything customer-facing either goes through human approval or is restricted to a tightly defined template with a review step. We test on real historical cases before go-live, including the awkward ones.
Related
Let’s find out whether technology is the answer
Tell us what you are trying to achieve. We will tell you honestly whether we can help, what it would take, and roughly what it would cost — before anyone signs anything.