Quicker replies for residents and customers, and a person always close by.
Our first users are the staff who take the calls: suggested answers, the case history and a summary on screen. Self-service comes later and only where it is safe, with firm limits and a handover that passes the whole conversation to a colleague.
For councils, State agencies and companies in Ireland that answer residents and customers on every channel, in Irish and English.
A query comes in on whatever channel the person chooses. The system works out what they want, which language they use and whether the subject is sensitive, then retrieves their case. Sensitive queries, or ones the service does not cover, go to a specialist with everything attached. For the rest, the assistant prepares an answer and suggests next steps, which a member of staff edits and sends. The CRM keeps a record of each query, and a sample is checked for quality.
Before and after
How the day changes for your staff
Today
With the system
Today
Today: Hunting for answers mid-call
To deal with one question, agents jump from the CRM to the case system to the knowledge base and back through old emails.
With the system
With the system: The case is already on screen
Before the agent has even said hello, the case status, earlier contacts and the matching knowledge articles are in front of them.
Today
Today: Notes after every call
Each call closes with minutes spent writing up notes and picking reason codes, usually from memory.
With the system
With the system: The assistant drafts, staff decide
Answers and call notes are prepared for the agent, who changes what is needed, sends it and remains accountable.
Today
Today: Chatbots with no way out
Scripted bots go round in circles on questions they cannot answer, so by the time someone reaches a person they are annoyed and must explain everything again.
With the system
With the system: Always a way through
When self-service hits its limits, or the person simply asks, a colleague takes over with the whole conversation and the reason for the query.
Today
Today: Irish speakers wait longer
Queries in Irish wait for the few staff who can take them, so the service depends on the rota.
With the system
With the system: Irish and English from the start
Answers are prepared in plain words in the language the person used, Irish or English, and a member of staff reads each one before it goes out.
Workflow
The workflow, step by step
Support for agents comes first: the assistant prepares, a person sends. Self-service for clearly bounded questions follows later, with the same safeguards and the same route to a specialist.
Illustrative workflow
7 steps · 2 human checkpoints
Legend
System
AI agent
Person
Human checkpoint
Exception path
Service assistance: illustrative workflow and specialist route
Read the diagram as text
The usual route begins with a query on any channel, moves through detecting the reason and background and the assistant preparing an answer, and reaches an agent who checks and sends it. The query is then logged in the CRM and may be sampled for a quality check.
A member of staff reads every draft and approves any suggested step before anything goes out: this is where a person steps in on the usual route.
Queries flagged as sensitive, complex or outside the service leave the usual route once the reason is detected and go to a specialist with all the background. They return to the route at the logging step.
01
Step 1: Query on any channelSystem
Calls (transcribed as they happen), web chat, email, messages from an online account and scanned letters land in a single queue. Where the request calls for it, the person is identified through their account login or MyGovID.
02
Step 2: Reason and backgroundSystem
The system detects what the person wants, their language and how urgent it is, and flags subjects that always go to a person, such as complaints, vulnerability or legal disputes. Case status is fetched over authenticated APIs.
03
Step 3: Route to a specialistPerson
Sensitive or complex queries, and those the service does not cover, are placed in a specialist queue with the transcript, the detected reason and the case data, so the person never has to start from the beginning.
Exception path: Sensitive, complex or not covered· Returns to step 6
Human checkpoint
The assistant does not give legal advice and does not decide anyone’s entitlements. The service owner, not the model, sets the rules for routing to a specialist.
04
Step 4: Prepare answer and next stepAI agent
Using approved knowledge only, the assistant writes an answer in plain words and in the person’s language, and suggests steps it may prepare, such as booking an appointment, updating an address or sending a status update.
05
Step 5: Agent checks and sendsPerson
The agent edits the prepared answer, accepts or turns down the suggested steps and replies on the channel the query came from.
Human checkpoint
No answer goes out and no step is carried out until a member of staff approves it. Their edits are kept to make future drafts better.
06
Step 6: Log and closeSystem
A summary, the reason for contact and the outcome go into the CRM or case system, and the conversation is kept for the agreed retention period.
07
Step 7: Check qualityPerson
Team leaders read a sample of conversations and drafts, note problems and revise knowledge articles. Every change passes the test set before it is released.
Components
What we would put in place
01
Channel and phone system links
Links to the contact-centre platform, web chat, email and online accounts, with calls transcribed live and a single conversation history across every channel.
Case data is retrieved only once the person is authenticated, and only for that person. The drafting assistant has a short list of steps, each with its own permission, and has no direct write access to records.
Human review
With assisted replies, a member of staff approves every answer and step. In self-service the routing triggers are written down: sensitive subjects, low confidence, the same question asked again, or a plain request to speak to someone.
The assistant gives no legal advice and makes no decision on entitlements, payments or claims.
Auditability
We log conversations, drafts, edits, suggested and completed steps, together with the model and prompt version, so any complaint can be traced from start to finish.
Data protection
Transcripts and logs are kept according to retention rules set with your data protection officer under the GDPR and the Data Protection Act 2018. Personal data is masked in reporting and never trains outside models.
AI transparency
Under the AI Act, people are told when they are dealing with an AI system, and they can ask for a person at any time. An answer sent by a member of staff is still your organisation’s answer.
Measures
How we would judge it
These measures are set with you in discovery and compared against the same queues before the assistant goes in.
How we would judge it
Metric
Why it matters
How we would measure it
01Resolved at first contact
Why it mattersThe plainest sign that the person got what they needed and did not have to come back.
How we would measure itRepeat queries about the same case within an agreed period, from CRM data.
02Time spent after each call
Why it mattersNotes and reason codes are where help saves time without touching the conversation.
How we would measure itWrap-up time per query from the contact-centre software, by queue and by channel.
03Quality of handovers
Why it mattersA handover only works if the specialist does not have to start over.
How we would measure itHow specialists rate the background they receive, and the proportion of handovers that need a call-back for missing details.
04Satisfaction
Why it mattersA quick reply counts for little if the person feels nobody listened.
How we would measure itShort surveys after each query and the themes in complaints, compared by channel and by language.
No targets are set before a baseline exists.
Rollout
How we would introduce it
Phase 01
Discovery
Reasons for contact, channels, languages and systems. We choose frequent, low-risk reasons and list the subjects that must always reach a person.
Exit criteria
Reasons and routing rules signed off
Baseline recorded for each queue
Data protection review under way
Phase 02
Pilot of assisted replies
Background panels, drafts and summaries for one team on one or two channels. Staff send every reply.
Exit criteria
Drafts reach the agreed quality bar
Agents trained and their feedback gathered
Logs and retention working
Phase 03
Self-service under supervision
Chosen reasons answered directly in web chat or the online account, with handover to the same team and all the background.
Exit criteria
Handover tested from start to finish
People told when an AI system is answering
Self-service results checked against complaints
Phase 04
Extend
Further reasons, channels and languages, each clearing the same test gate.
Resident queries to county and city councils about housing, planning and payments, with MyGovID login for status checks and plain answers in Irish and English.
Questions about policies and claims at insurers, or accounts at banks and credit unions, with strong authentication and complaints and vulnerable customers always sent to a person.
Questions about appointments, charges and paperwork at hospitals and community services, with any medical question sent to qualified staff.
Illustrative example
Resident queries at a county council, in Irish and English
Situation
A county council answers residents about housing, planning applications and payments by phone, email and web chat. Queries in Irish wait for a small number of staff, and writing up notes takes a large share of every shift.
System
Inside the existing CRM console, the assistant prepares answers and call notes in the resident’s language from one set of knowledge articles kept in Irish and English. Self-service in web chat handles status questions after a MyGovID login.
Human control
Staff send every assisted reply, and Irish answers are read by staff who work through Irish. Complaints, housing emergencies and vulnerable residents always go to a person, with the conversation attached.
What we would measure
Wrap-up time and resolution at first contact by language, and how specialists rate the handovers they receive.
Ireland
Two languages, one standard of service
Public bodies in Ireland have duties under the Official Languages Act, so a resident who writes in Irish should get the same quality of answer and the same route to a person as one who writes in English. We design one assistance layer with knowledge articles and routing rules kept in step across both languages, and MyGovID login where a query needs the person’s own record.
The AI Act obliges you to tell people when they are dealing with an AI system, and in Ireland S.I. No. 636 of 2023 sets accessibility requirements for in-scope consumer services, enforced by the CCPC, ComReg and Coimisiún na Meán. Both shape the design from the first workshop, together with GDPR retention rules for transcripts and logs.
That is not the aim. We begin with assisted replies, so agents spend less time searching and typing and more time talking to people. Self-service is only added for clearly bounded questions, and a person is never more than one step away.
02How do you decide between self-service and a handover?
We design the handover before anything else. Each self-service reason has written exit conditions and passes over all the background. We track the quality of handovers and repeat queries, not only how many conversations the assistant completes.
03Can it answer people in Irish as well as English?
Yes. The assistant prepares drafts in the language the person wrote in, in plain words. For public bodies with duties under the Official Languages Act, Irish and English knowledge articles are kept in step, and staff who work through Irish review the Irish answers. Our own team works with you in English.
04Must we tell people they are dealing with an AI system?
Yes. The AI Act requires that people are informed when they interact directly with an AI system, and in Ireland the Regulation of Artificial Intelligence Act 2026 sets up the AI Office of Ireland to coordinate supervision. We build the notice, the option to reach a person and the logs in from day one.
05Will it fit our current CRM and contact-centre software?
That is the plan. Where we can, we build inside the console you already use, for example as components in Salesforce Service Cloud, and connect to your phone system and case systems through their APIs.
06What is a good first step?
Look at why people contact you. The most common low-risk reasons make a sound first scope for assisted replies. Read how we handle AI Agents & Automation.
Show us your service channels
Send us your main reasons for contact, your channels and your languages. We will suggest where assisted replies help first and where a person must stay at the front.