Automation that takes the repetitive work. People still decide.
We automate the jobs your team does by rote: sorting enquiries, chasing paperwork, drafting replies and building the Monday report. Rules and schedules do most of it. AI joins in only where it helps, answers from your own documents, and a person approves anything that matters.
A Monday morning, before anyone logs in.
By the time your team arrives, the routine work is done and the one decision that needs a person is waiting for them. Every result can be traced back to the rule or document behind it.
- Sort 23 new enquiries by service and urgency3 marked urgent and moved to Sam Ellis’s queueDone
- Send this week’s appointment remindersConsent checked before every send · 41 sent, 2 held backDone
- Draft replies to common questions9 drafts from your own policies, each showing its sourceDone
- Summarise last week’s feedback into themes4 themes · car parking mentioned most oftenDone
- Update the Monday reportFigures refreshed and sent to managers at 07:30Done
- Reply to Amira Khan’s complaintDraft ready · complaints always go to a personNeeds your approvalApproveEdit
Leo Hart askedCan a client rebook after missing two appointments?
Yes, once. After a second missed appointment the client may book one more time, and their keyworker is told.
Source: Appointments policy v4, section 3.2
Leo Hart askedWhat mileage rate do we pay for home visits?
I don’t know. That isn’t in the documents I can see. Finance can tell you, or an admin can add the expenses policy to my sources.
No source found · not guessed
The guardrails come first, not after something goes wrong.
Every AI feature we build follows the same path, from what data it may see to who can switch it off. Your data protection officer and IT team can check each step.
- Data in
It sees only what it needs
- The fewest fields possibleEach task gets the fields it needs and nothing else. Names stay out where they are not needed.
- UK GDPR and a DPIAWe set the lawful basis with you and complete a data protection impact assessment before go-live.
- Never used for trainingProviders process your data under business terms that rule out training their models on it.
- The model
It answers from your documents
- Grounded answersThe assistant answers from your policies and handbooks, not from whatever it learned on the internet.
- Sources shownEvery answer names the document and section it came from, so staff can check it in seconds.
- “I don’t know”If the answer is not in your documents, it says so instead of guessing.
- Checks
It is tested before anyone relies on it
- Evaluation before go-liveWe test it against a set of your real questions and agreed answers, and share the results with you.
- Rules around itPlain rules block actions it must never take, such as texting someone without consent.
- Everything loggedEach question, answer and action is logged with its sources, so any result can be traced.
- A person
People stay in charge
- Approval where it mattersAnything that changes a record, spends money or contacts a client waits for a person to approve it.
- Corrections feed backWhen staff flag a wrong answer, we fix the source document, so the mistake is not repeated.
- A kill switchAn admin can turn any AI feature off in one click, and the work drops back to your normal process.
Rules first. AI where it earns its place.
Most repetitive work follows rules you could write on a card. Rules are cheap, fast and predictable, so we use them first. We add AI only for what rules cannot do well, such as reading free text or summarising feedback.
Rules and schedules
Predictable, testable, and most of the work.
- Enquiry routingNew enquiries from forms, email and texts are sorted by service, location and urgency, and land in the right team’s queue with an owner.
- Follow-ups and chasersPayment chasers, follow-up check-ins and appointment messages go out on time, with consent and opt-outs checked before every send.
- Reports that build themselvesThe weekly or monthly report is refreshed and sent on time, from the same figures your dashboard shows.
- Checks that run overnightOverdue tasks, expiring stock, missed appointments and data-quality problems are found each night and put in front of the right person.
AI, with a person approving
Only where reading or writing language helps.
- Replies drafted from your policiesCommon questions get a draft reply built from your own policies and templates, with the source shown, ready for a person to check and send.
- Reading forms and documentsDetails are lifted from referral forms, invoices and PDFs into the right fields. Anything uncertain is flagged for a check, never guessed.
- Feedback summariesHundreds of comments become a short list of themes, each with real quotes, so managers can see what is changing week to week.
- A staff help assistantStaff ask in plain English and get answers from your handbooks and policies, with the source, or an honest “I don’t know”.
On the platform we built for nine NHS services, staff use an AI assistant that answers only from an approved knowledge base. No client data is sent to the model. Read the case study
Often paired with integrations that feed it, reporting and dashboards, the business system it works inside and our security pack.
AI and automation questions
Something else? hello@echticstudios.com
Is our data used to train AI models?
No. We choose a provider for each project under business terms that rule out training, and name it in your data processing agreement. Each task is sent only the fields it needs, and your DPIA records where each provider processes data.
What if the AI gets something wrong?
It is built so a wrong answer gets caught. Answers show their sources, the assistant says when it does not know, and anything that changes a record, spends money or contacts a client waits for a person to approve it.
Do we need AI at all?
Often not. Many jobs are handled better by rules and schedules, which are cheaper and fully predictable. We tell you which parts of a task need AI, if any, before we quote.
Can we switch it off?
Yes. Every AI feature has a kill switch that your admins control. Switching it off sends the work back to your normal process, so nothing stops.
Which AI providers do you use?
Business-grade AI providers, named in your data processing agreement, and we build so you can switch between them. If your organisation has already approved a provider, we use that one.
How do we start?
With one task that eats your team’s time every week. We automate it, measure it against the old way, and only then choose the next one.
Tell us the job your team does every Monday.
A free consultation, then a fixed quote for one task first. Each next step is quoted on its own, and you can stop after any of them.