Service 09
Automate the typing, keep the judgement
A support assistant that answers from your own documents, search across years of files, intake that summarises itself, and draft replies a person approves before they send.
Why people call us about this
The same answers, typed again every day
Staff retype the same five replies, summarise the same forms, and copy details between systems. That is the work worth automating, not the decisions.
- The same questions arrive by email every week
- Enquiry forms get read twice and summarised by hand
- Useful information is buried in years of documents
- Off-the-shelf chatbots answer confidently and wrongly
What you get
What we build
We start with one process and a clear before and after, not a platform rollout.
01
Use case pick
We look for a task that is repetitive, high volume, and low risk if it needs a second pass. That is where this pays off.
02
Assistants on your content
Answers drawn from your own documents and pages, with links to the source so staff can check them.
03
Document search
Ask a question across years of policies, minutes, and manuals and get the passage, not a guess.
04
Intake and summaries
Forms and emails summarised into the fields your team needs, ready for a person to confirm.
05
Connected to your tools
Wired into your inbox, CRM, booking system, or portal so nothing needs copying by hand.
06
Limits and logging
Clear boundaries, refusal behaviour for anything sensitive, full logs, and a human step before anything is sent.
Typical timeline
Three to six weeks for a first useful integration
We run it beside your current process before anyone relies on it.
Step 01
3 to 5 days
Pick the task
We shadow the work, count how often it happens, and agree what good output looks like.
Step 02
1 to 2 weeks
Prototype
A working version on your real content, tested against cases you already know the answers to.
Step 03
1 to 2 weeks
Integrate
Connected to your systems with permissions, logging, and the human approval step in place.
Step 04
2 weeks
Run in parallel
Staff check every output for a fortnight. We tune it, then agree what it can do unattended.
This service, in the wild
AI work inside systems we built
Open a project to read the full story.
AI Integrations
Academic platform, nursing and midwifery colleges
Scholarium
An assistant answering student questions about fees and admissions from the college’s own handbook, with links to the source page.
Intake automation
Retail, Uromi
Computer accessories store
Printing enquiries and uploaded files summarised into a job sheet with page count and finish, ready for staff to price.
Tools and platforms
What we build with
We are clear about what leaves your systems, where it goes, and how long it is kept.
Claude API
OpenAI API
Vector search
Zapier and Make
Your existing CRM
Audit logging
Placeholder quote, replace with a real client
It answers the fee questions we used to answer forty times a week, and it links to the handbook page so nobody argues with it.
Placeholder name
Admin lead, placeholder institution
Questions we get asked
AI questions
Will it make things up?
We build assistants that answer only from your documents and say they do not know otherwise, with a link to the source so staff can verify. That removes most of the risk, and anything customer-facing keeps a human approval step.
Where does our data go?
We tell you exactly which provider processes what, what is retained, and for how long, before anything is connected. Sensitive categories can be excluded or kept on infrastructure you control.
Is this going to replace staff?
Not in the work we do. It removes retyping and summarising so the same people handle more without working later. Decisions stay with a person.
Can you add this to a system you did not build?
Usually, if it has an API or a database we can read safely. We start with a small read-only integration and prove it works before touching anything that writes.
Previous service
Next service
Got a task your team repeats every day?
Describe it in two lines. We will tell you whether it is worth automating and roughly what it takes.