An agentic use case is a job, not a feature. It starts with a person — a claim approver, a service engineer, a parts manager — and the part of their week that is really three systems, a spreadsheet and a memory. It names what goes wrong. It lays out the flow an agent runs instead: what it watches, what it computes, what it drafts, and where it stops and waits for a human. And it says what changes for the business. This page states the use cases xMatix runs today in that form, so you can hold them against your own operation and see which ones are yours.
Three labels appear on the cards. Ships today means the engines and agent tools are native to the platform. Built in AI Studio means the use case is assembled in AI Studio from a custom agent, your knowledge and your connected tools — configuration, not a vendor project, but work your team or ours does for your operation. Roadmap marks the one piece we describe but do not yet ship.
The autonomous claim validator
For the claim approver — who validates dealer submissions by matching invoices, service reports, warranty policy and check sheets across systems, by hand, one claim at a time. Ships today
- Manual checks across invoices, service reports and check sheets
- Slow settlement, so dealers carry the cost of the wait
- Inconsistent decisions between approvers
- Cash blocked in claims nobody has looked at yet
- A completed job, return or scheme period generates the claim from its evidence — parts, labour, photos, invoices
- Validation rules check policy, eligibility and consistency deterministically
- Sense triages the queue: which claims will reject on current evidence, and why
- Clean claims are approved in bulk; exceptions carry their reasons to a person
- Settlement posts to the ledger and the dealer is notified
- Approver capacity goes to exceptions, not paperwork
- Decisions become consistent because the rules are written down
- Faster reimbursement, so the channel trusts the programme
- Every decision carries its evidence into the audit trail
Service opportunity discovery
For the service outreach executive — who rings customers about upcoming or overdue maintenance from scattered service logs, usage history and manual checks. Ships today
- Maintenance opportunities missed because the data is spread out
- No tracking of usage hours, asset type or due dates in one place
- Manual effort to work out who to call this week
- Late follow-up means lost service revenue and lower uptime
- The contract book, maintenance schedules and service history are read overnight
- Every asset falling due produces a reminder lead, ranked and queue-ready
- The service-recommendation nudge flags work the record itself indicates
- Sense drafts the outreach and answers why this customer, why now
- The executive approves the call list; bookings land as appointments
- Opportunities identified before the customer notices
- Better asset uptime through maintenance done on time
- Higher conversion because the call carries context
- Analysis time goes to zero; the queue is ready at open of business
The service engineer's assistant
For the service engineer — who troubleshoots by pulling asset history, hunting for the right part and filling check sheets during the visit. Ships today
- Time spent finding the asset's history before the work starts
- Manuals scanned for guidance under time pressure
- Parts identified by memory and phone calls
- Check sheets filled differently by every engineer
- The engineer opens the job on the mobile app; the open record is the assistant's context
- Sense answers from the asset's history, past jobs and the knowledge base — manuals included
- A parts list is drafted from the diagnosis for the engineer to confirm
- Checklist templates standardise the check sheet; readings are captured offline
- The job card, parts and check sheet update the record when the visit closes
- Faster preparation and first-time fix
- Consistent service output across the team
- More jobs per engineer per day
- Less lookup and entry time in the field
Connected assets and telematics
For the fleet or asset manager — whose vehicles and machines already report fault codes, hours and health, into a system nobody in service reads. Built in AI Studio
- Telematics data sits in the provider's portal, disconnected from service
- Fault codes are found at the next visit, not when they fire
- Preventive alerts reach nobody who can act
- Breakdowns that the data predicted still happen
- The integration hub ingests telematics events and fault codes against the asset record
- Rules classify severity; a custom agent explains the code against the manual
- A service recommendation or lead is created and routed to the right technician
- The customer sees the health status and next-service date in their portal
- The visit closes the loop — the fix is recorded against the code that raised it
- Fewer breakdowns and roadside events
- Lower repair cost and downtime
- Higher asset uptime and reliability
- A service relationship that starts before the failure
Opportunity discovery for key accounts
For the key account manager — who scans tender portals, industry sources and social signals by hand to find region- and keyword-matched opportunities. Built in AI Studio
- Opportunities missed because the sources are scattered
- Manual research across sites and regions
- Signals nobody monitors for the keywords that matter
- Lead generation that consumes the selling week
- Your source connectors are registered as tools the agent may call
- A scheduled agent workflow monitors them for contracts and keywords
- Rules filter and prioritise by region, industry and account fit
- The agent scores and summarises each candidate with its evidence
- Leads land in CRM as suggestions; the manager qualifies and accepts
- Key account time spent on high-value leads
- Shorter discovery cycles
- Higher win rate through earlier engagement
- Less research load on the team
The parts manager's reorder
For the dealer parts manager — who reviews stock, predicts demand, checks lead times and decides reorder quantities before placing the order. Ships today
- Stock and demand checks that consume the week
- Quantities that overstock the slow lines and starve the fast ones
- Cash blocked in inventory nobody will sell
- Lost sales when the part is not on the shelf
- Stock, in-transit, open orders and lead times are netted continuously
- Three replenishment engines — min-stock, replenish, forecast — project the need
- Suggested purchase lines are drafted with their arithmetic shown
- The purchase-order nudge flags a line the pipeline already covers, or one still short
- The manager approves; the order is placed and tracked
- Better planning, with every number inspectable
- Lower inventory cost and less cash blocked
- Fewer stockouts, more sales
- Faster cycles with less manual work
The AI-assisted contact centre
For the contact-centre lead — who routes sales enquiries, service bookings, complaints and roadside calls across a central team and the dealer network. Ships today Roadmap: voice agent
- Callers repeat themselves to every person they reach
- Routing by guesswork between the centre and the dealer
- Recordings nobody has time to listen to
- Outcomes — a lead, a booking, a case — typed up after the call, or not
- An inbound call is matched to the customer and vehicle before it is answered
- Routing rules pick the queue, the dealer or the agent by skills, capacity and presence
- The agent sees the record; Sense answers the customer's history on the spot
- The call is recorded and summarised; sentiment and outcome are captured against the record
- The outcome creates the lead, booking, case or dispatch — confirmed by the agent
- Smarter conversations, because the context arrives first
- Seamless handover between the centre and the dealer
- Every call becomes a record, not a memory
- Lower cost per resolved contact
The talking voice agent — AI answering first, humans when needed — is the piece marked roadmap. The routing, recording, analysis and record-creation beneath it ship today, which is why we built the plumbing first.
The collections chaser
For the credit controller — who chases receivables from an ageing report and a phone, and finds out about the dispute after the reminder has gone. Ships today
- Ageing reviewed weekly, so exposure grows between reviews
- Reminders written by hand, inconsistently
- Dealers over limit discovered at month end
- No trail of who was chased, when, and what they said
- The overdue-receivables nudge ages money past terms as it happens
- Credit limits gate new orders at save, not at month end
- Sense answers who is behind and why from the invoices the controller can see
- Reminders are drafted per account, in the right tone, and wait for approval
- Sent, acknowledged and disputed all land on the account's timeline
- Exposure seen daily, not weekly
- Consistent, timely reminders
- Fewer over-limit surprises
- A collections history the auditor can read
The beat that plans itself
For the sales representative and the area manager — who plan the week from a paper beat, pitch schemes from memory and take orders that will be re-typed later. Ships today
- Beats planned once and followed loosely
- Scheme slabs missed by a case or two
- Outlets going quiet unnoticed until the month closes
- Orders taken offline and lost in the gap
- Visit plans generate from route schedules and learn from execution history
- Suggested order quantities are computed per outlet from its own pattern
- The scheme-gap nudge shows the shortfall to the next slab and what it is worth
- The dormant-customer nudge flags an outlet quiet against its own rhythm
- Orders are taken offline, spoken or typed, and sync as real orders
- Plans that are followed because they fit the day
- Scheme spend that converts to volume
- Outlets recovered before they churn
- No re-typing, no lost orders
The shelf that scores itself
For the merchandiser and the trade marketing manager — who judge planogram compliance and share of shelf from photos nobody measures. Ships today
- Compliance eyeballed, and argued about
- Share of shelf estimated, never measured
- Photos filed and forgotten
- New SKUs invisible to the scorer until someone retrains it
- The rep photographs the shelf during the visit — offline capture included
- A vision pipeline detects products and facings; the score is computed, never guessed
- Compliance and share of shelf post to the visit and the outlet
- Gaps become tasks; trends become nudges to the area manager
- New SKUs are taught from a handful of reference photos and a reindex
- Compliance that is a number, not an opinion
- Share of shelf tracked outlet by outlet
- Merchandising spend argued from evidence
- A model that keeps up with the range
The administrator who configures by asking
For the tenant administrator — who turns a business request into fields, layouts, rules and permissions, and worries about getting it wrong in production. Ships today
- Every small change is a ticket and a wait
- Configuration knowledge concentrated in one or two people
- Changes made straight into production
- No record of why a rule exists
- The administrator describes the change in plain language — in xMatix, or from Claude or ChatGPT
- Sense Control follows the platform's own task recipes and previews the change set
- Each step waits for approval; questions come back as questions
- The change is applied in a sandbox and promoted with undo and drift detection
- The conversation is the change record
- Changes in hours, not sprints
- Configuration knowledge shared with the copilot
- Production protected by preview and promotion
- Every change explains itself
Your business, inside the assistant you already use
For everyone — who already works in Claude, ChatGPT or Claude Code and has been copying numbers out of the business to get them there. Ships today
- The assistant is where the thinking happens; the data is elsewhere
- Exports pasted into prompts, stale by the time they are read
- No permissions, no citations, no audit on what left the building
- IT cannot see it, so IT cannot govern it
- An administrator enables external assistants and sets the write ceiling
- The user pastes one endpoint into their assistant and signs in as themselves
- The assistant discovers entities, queries, runs reports and dashboards, searches knowledge
- Writes, if allowed, run under the user's own permissions
- Every call is audited; the administrator can disconnect any session
- Answers grounded in live, permission-checked data
- No more exports, no more paste
- IT gets a switch, a ceiling and a log
- The assistant your people chose, on the data your business governs
How to read your own operation in these cards
Start with the person, not the technology. List the jobs where one role reconciles several systems by hand, and write each one as the cards above are written: the problems, the flow you would want, the impact you would measure. Then sort them on the autonomy ladder. Work that is arithmetic over your own records — ageing, netting, scheduling, scoring against a rule — can run alone tonight. Work that needs language and retrieval can be drafted and put in a queue. Work that moves money keeps a human on the trigger. The use cases that pay back first are almost always the ones where the deterministic part is large and the approval is a single click. Choosing your first agentic AI use case is the longer version of that method.
Common questions
Which of these use cases run without a human involved?
The deterministic engines: claim generation from evidence, service-lead generation from contracts and schedules, replenishment suggestions, visit-plan generation, shelf scoring and the scheduled nudge sweeps. They run on schedules and put their output in a person's queue. Approving a claim, sending a reminder, placing an order or changing configuration is always a human click.
What does "Built in AI Studio" mean in practice?
The use case is assembled from platform pieces rather than shipped as a finished screen: a custom agent with its own instructions and model, knowledge collections it may search, your connected systems registered as tools it may call, a schedule if it should run unattended, and a budget. It is configuration your team or ours completes for your operation, governed the same way as the built-in agents.
Can we start with one use case and add others later?
Yes, and we recommend it. The engines and agents share one data model, one permission model and one audit trail, so a second use case reuses everything the first one set up. Most organisations start with the case whose deterministic part is largest — claims, replenishment or service leads — and add the assistant-led ones once the team trusts the queue.
How do these use cases stay within our security model?
Every read runs as a user — the asking user for interactive work, a narrowly scoped execution account for scheduled agents — through the same record security, field permissions and restriction rules as every screen. Writes are approval-gated, budgets are enforced before the model is called, and every run is audited. The governance contract is public at how Sense earns trust.
Do we need our own AI models or keys?
No. The platform meters usage against your organisation's credit allowance on the models it provides. Organisations that prefer to bring their own model keys can, per model, and are then metered on their own account.