Frontline DualPathThe console that answers the phone and makes the first touch decisive.
Frontline sits on top of Autotask with an AI co-pilot. The person who picks up the ticket can understand it fast, fix what they can on the call, and send the rest to the right tier with the context attached. Everything they do is saved back to Autotask.
Knowledge that curates itself, and a second QoL pass.
The knowledge queue now runs itself, and the rest of the November QoL review has shipped: undo, one search box, a link back to Autotask, and tidier admin pages.
AI knowledge curator
Frontline drafts knowledge from resolved work, but nobody had time to review the queue, so roughly 5,900 drafts were waiting. An AI curator now decides them automatically. Each draft is judged at temperature 0 against a strict rubric and compared with the three closest entries already published, so duplicates are turned away. The curator approves, rejects, or approves with tightened wording, and approvals go into the shared knowledge base.
- People stay in control. Admins can still approve or reject anything pending. The queue has Pending, Approved and Rejected tabs with counts, and every decision shows who made it, why, and when.
- Less junk to start with. The new version of the drafter no longer files bookkeeping facts or synthetic rows.
- Kill switch. One setting stops the curator without a rebuild.
On the board
- Undo. After you change a status or reassign, the confirmation offers Undo. It isn't offered for closures, follow-up statuses or escalations.
- One search box. It filters the list as you type and searches all tickets. Type an exact ticket number and press Enter to open it.
- Open in Autotask. A link in the ticket header takes you straight to the ticket in Autotask.
- Escalating an unassigned ticket asks first, and names the tier and queue it's going to.
- Recent tickets in the palette. ⌘K with nothing typed lists the last 8 tickets you opened.
- Select toggle. Bulk checkboxes appear only when you turn them on.
- Plain error messages. Board alerts explain what went wrong in a sentence, not an API error code.
Insights
- One period control across KPIs, Accounts and Time, from 7 days to 1 year.
- Long windows are complete. KPIs now read up to 25,000 facts. Before, a 90-day view was quietly cut off at about 6,100 rows.
- Sortable Time ledger. Ticket Review keeps its team-table sort in the URL.
- Plain-language labels. Hover over a metric term to see what it means.
Admin & automation
- One Feature access page for VoIP, Scout and Networks. Old links redirect.
- Tenants. Shared tabs and a breadcrumb, and the list can be filtered and sorted by status.
- JML inbox and portal accounts. Search, a customer column and Load more. New portal accounts pick queues from a checklist.
- Reorder rules. Move triage and approval rules up or down.
- Automation bindings are edited with the workflow's own run form. Raw JSON is still available under Advanced.
- Stacked dialogs. Esc closes only the one on top.
Earlier · 6 Nov 2026 · the first operator quality-of-life pass
On the board
- Command palette. Press ⌘K / CtrlK to jump to any page or action.
- Shortcut help. Press ? to see every key. The list is built from the board's own keymap, so it's always accurate.
- Keyboard ticket navigation. j/k, Enter, /, t triage, n note, e escalate, c copy number.
- Failed notes keep your text. If a note fails to post, it stays in the box so you can retry without retyping.
- Triage waits have a limit. A slow triage stops waiting after a set time, and you can cancel it sooner.
- Failures say so. If a list can't load you see "Couldn't load · Retry", not an empty list that looks like there's no work.
- CSR drafts are protected. Your edits aren't overwritten, there's a recovery path when a save result is uncertain, and blocked follow-ups explain why.
- Alerts don't vanish. When toasts stack up, they expand instead of quietly dismissing ones you haven't seen.
Everywhere
- Views live in the URL. Board, KPIs, Time, Accounts, Networks and VoIP filters survive a reload, and you can share them as a link.
- Freshness chips. Data views show how old their data is. Dashboards show whether the feed is Live, Reconnecting or Polling.
- Live connection pill. The console shows whether the real-time connection is actually up.
- Copy chips. Copy ticket numbers and IDs in one click.
- Shared CSV export and table sorting. Exports are guarded against spreadsheet formula injection.
- More accessible. Device status doesn't depend on color alone, and the KPI quadrant dots work from the keyboard.
- Networks and VoIP. Data ages update live, the tenant and site are kept in the URL, and you can sort the device table.
Safety nets
- Confirm before you can't undo. Portal-account delete, MFA reset, passkey revoke, password reset and role removal now ask first.
- DPIGA "Switch to sync" shows the impact from the last preview before it applies.
- Unsaved-change guards on onboarding settings, automation workflows and tool policies, so navigating away doesn't silently discard work.
For leads & admins
- Knowledge queue. Filter by queue scope and category, then approve or reject in bulk.
- Rule lists show the order rules run in, are searchable, and show who edited each rule last.
- Tool policies mark unsaved cards, show results inline, and can be filtered.
- Automation shows JSON errors inline and lets you filter runs.
- Access pages link straight to Roles & access and Tenants.
Every ticket costs a senior hour it shouldn't.
The fix isn't hiring more tier-1 staff. The fix is making the first touch decisive, so simple work gets closed at the front desk and hard work arrives already triaged.
Before
- Tickets sit until a tech reads them from scratch.
- Password resets, known errors, and "who even gets this?" questions take up T2/T3 time.
- Context gets gathered again at every handoff, and the customer explains the problem twice.
- Time gets logged from memory at the end of the day, so the numbers you manage by are guesses.
With Frontline
- One click turns a raw ticket into a structured read plus a suggested next move.
- Scoped tools let the first touch actually fix things: passwords, Teams presence, endpoint checks.
- An escalation carries the tier, target queue, roles and reason, so the next tech picks up where the last one stopped.
- Notes, time and KPIs record themselves as part of the work.
Queue → Read → Triage → Work → Act.
This loop is the backbone of Frontline. Every tool exists to help you finish it on the first touch, and Autotask stays the system of record the whole way through.
Queue
Every Autotask queue you can access, each with a live open-ticket count. Newest tickets show first.
Read
A normalized ticket: status, priority, SLA pressure, company, contact, and the full activity stream with who said what to whom.
Triage
One click gets you a summary, urgency, sentiment, what to ask the caller, a do-it-now fix and a suggested escalation.
Work
Talk to the customer with a co-pilot that already knows this ticket. Collect what you need before you hang up.
Act
Add a note, change status, reassign, escalate, or run a tool. Each one is saved to Autotask a second later.
Nothing lives only in Frontline. Autotask stays the source of truth. If the AI gateway goes down, the whole loop still works without it.
Illustration with fictional data. Three zones: queues · list · detail. The command deck runs along the bottom.
Minutes come back at every step.
Frontline doesn't ask anyone to work harder. It removes the steps that used to slow tickets down. Each row below is a specific mechanism that's in production today.
One click turns a raw ticket into a decision.
Triage doesn't return a paragraph to skim. It returns labeled sections you can act on. Learn them once and you can read any triage at a glance.
Triage prepares you for the call. It isn't an order.
Skim the activity stream first, so triage confirms your read instead of surprising you. If it's wrong, don't follow it. You decide.
Sections fail independently
If one section errors, the others still appear. You can re-run triage, lean on the ticket chat, or work without AI. The normalized ticket and every write-back action keep working.
Ticket-scoped chat
A slide-out assistant that already knows this ticket. Use it for wording, next steps, or reading an error message. Responses stream in.
A triage console that grew into the operator's whole cockpit.
The core is small and the tools around it cover a lot. Every capability answers the same question: can we close this on the first touch instead of escalating it?
Triage & routing
Ticket board
Live queue counts, a normalized ticket view, a full activity stream, resizable panes and configurable columns.
Helps: you see the freshest work first, with no Autotask codes to decode.
Agentic triage
A structured read, a do-it-now fix, risk flags, draft note and reply, and a suggested tier and queue.
Helps: you know what to do in seconds instead of minutes.
Tiered escalation
T1, T2, T3 and Engineering. One action moves the queue, records the tier and adds a reason note.
Helps: context goes with the ticket, so nobody rebuilds it.
Calls & intake
CSR call workflow
A focused dashboard: find the property and caller in Datto, see existing issues, record the facts, and choose Resolved on the call or Escalate to CSS.
Helps: routing is handled for you, and repeat callers update their open ticket instead of creating a duplicate.
Technician call desk
A browser phone with live transcription. Frontline pulls up related tickets and runbooks as the call goes and drafts the subject and description.
Helps: the ticket is mostly written by the time you hang up.
Caller verification
Push-approval or voice-keypress challenge before any sensitive action. SMS and email codes don't count.
Helps: you can act quickly without getting social-engineered.
Do-it-now tools
M365 actions
A single multi-tenant Entra app with per-action scopes. Password reset is the canonical tool and can reach the customer's parent and child tenants.
Scout
AI-proposed, operator-approved diagnostics on the customer's actual device. No one gets a blank remote shell.
JML
Joiner, mover and leaver flows that mirror a real colleague's access. Destructive steps need two keys.
Comms & network
Teams presence and DND, the Grandstream VoIP fleet, and Meraki/UniFi network health.
Customer-facing safe doors
Secure Send
Seal a secret and email a view-limited link with a code, MFA and an expiry. Never paste a password into an email again.
Info Request
An AI-drafted form the operator approves and sends. The customer's answers go straight into the ticket.
Customer portal
M365 sign-in, ticket self-service and live chat with the CSR.
Notes as email
Send a public note as a real email, routed so Autotask doesn't send it twice.
Automation & autonomy
Automation
A visual workflow builder over Graph, OneDrive, SharePoint and Frontline tools. It has customer bindings, approvals, and durable execution with full review history.
Autopilot rolling out
An autonomous agent for the Frontline-Triage queue. It gathers info, runs tools when it's confident, asks a human when it isn't, and parks finished work for review.
Workbench
A direct assistant console scoped to one customer, with artifacts and documentation (runbooks, repos, the vault) close at hand.
Knowledge curator new
An AI curator works through the knowledge queue. It approves durable guidance into the shared knowledge base and rejects duplicates and filler. Admins can override any decision.
Run the business
Impact & Effort
Value delivered rather than ticket count. Each ticket is scored once, at completion, against a fixed rubric.
Ticket Review
Each tech's tickets, stats and what's stuck, with saved AI assessments judged against the operator's role.
Account Pulse
Each customer at a glance: health, response rate and outreach.
Time & billing
Effort is captured automatically and shown on a per-operator billing balance sheet.
Dashboards
Boards that you and the assistant build, with live updates and source freshness shown.
Ticket Game
A team board and scorecards with tunable scoring, which you can preview before rolling it out.
Your shift, step by step.
You don't need to memorize every tool. You need the loop. When a tool fits the ticket, triage puts it in front of you.
Start your shift · 1 min
- Sign in through Cloudflare Access. CSR accounts land on the CSR dashboard. Everyone else lands on the ticket board.
- Scan the queue counts on the left. Look at the totals rather than counting rows.
- Open your assigned queue. The list is sorted Created ↓, so you're looking at new work, not the oldest tickets.
- Press j/k to move through tickets and Enter to open one.
Find any ticket · 10 sec
- Press / to jump to the search box above the list. There's only one search box now.
- Type a title fragment. The list filters as you type, and Frontline also searches all tickets in Autotask.
- Know the number? Type the exact ticket number, including T-numbers, and press Enter to open it.
- Press Esc to leave the search box.
Run triage and read it · 30 sec
- Skim the activity stream first, so you know the story.
- Press t (or Run AI triage). Sections appear as each one finishes. You can cancel a wait that's taking too long.
- Check the Summary against your own read.
- Read Data to collect. That's your list of questions for the call.
- Look at Do-it-now resolution. If it says a CSR can resolve it and the confidence looks right, use the customer script.
- Check Risk flags. If they fire, plan to escalate.
Write it back: notes, status, reassign · 20 sec
- Press n to write a note, or apply the drafted note or reply from triage and edit it first if you need to.
- Choose internal or public. A public note can go out as a real email with Send as email.
- Set the status, or reassign. Assign me adds you or makes you primary. Picked the wrong one? Press Undo on the confirmation.
- Check the ⏱ chip on your note. If the time estimate is wrong, edit or split it.
- Need something only Autotask has? Use Open in Autotask in the ticket header.
Escalate with context · 15 sec
- Collect every Data to collect item while the caller is still on the line.
- Press e to see the escalation options. Triage has preselected the tier, roles and target queue.
- Check the tier against the tier map.
- Write or accept the reason, then confirm. One action moves the queue, sets the tier and adds the reason note. If the ticket has no one assigned, Frontline asks first and names the tier and queue it's going to.
Take a call as a CSR · the whole call
- On the CSR dashboard, press Start call.
- Find the property and select the caller from Datto. First name, nickname and last name work in any order. Use Caller is not listed only as a last resort.
- If Frontline shows an existing issue, ask whether this is the same problem. A related call updates the open ticket.
- Record what's happening and the troubleshooting you've done. If there's a surge alert for the property, check with CSS before repeating troubleshooting.
- Review the subject and description Frontline drafts and correct any facts that are wrong.
- Choose Resolved on the call with the confirmed fix, or Escalate to CSS. Frontline handles the routing.
- Confirm the saved ticket number, then start the next call.
Take a new issue as a technician · the whole call
- Open New issue from the rail (
/intake). Enable the phone, then answer or dial. Transcription starts when the call connects. - Choose the customer and select the real Datto/Autotask contact, or enter an unlisted caller's name.
- As the call goes, Frontline brings up related recent tickets and published runbooks, along with suggested next checks.
- Generate the subject, description or both from your notes and the transcript. Review the drafts and apply them. Live updates never overwrite fields you've edited.
- Pick the queue (the matching tier queue is preselected) and priority, then create the ticket. The transcript links automatically.
Verify the caller, then reset a password · 2 min
- Triage offers M365 password reset once the tenant is consented and you're allowed to run it.
- Open Verify caller. Choose a push (Authenticator) or voice keypress method and trigger it.
- Continue only if the result is approved. Denied, timeout or error means stop.
- Run the reset. If it creates a temporary password, it's shown to you once and never stored. Read it to the caller.
- The action is saved to the ticket as an AI-assisted note attributed to you, and recorded in the tool audit log.
Diagnose a device with Scout · 3–5 min
- Open Scout from the ticket or the rail and pick the customer's device.
- The AI proposes the diagnostic checks to run.
- Review each proposal and approve it. Destructive checks need an explicit approval step.
- Watch the results stream in, then save the findings to the ticket.
Onboard or offboard someone (JML) · 5 min
- Open the JML blade from the ticket. It also runs from the standalone console and from portal intake.
- For a joiner, pick a peer to mirror. Frontline suggests access based on a real colleague.
- Review the proposed steps.
- Destructive leaver steps need two keys. No single click offboards a person.
Send a secret safely (Secure Send) · 30 sec
- Open Secure Send from the ticket.
- Paste the secret and choose its gates: access code, MFA, expiry and view limit.
- Send. The customer gets a magic link, and Frontline enforces every gate.
Get details from a customer (Info Request) · 30 sec
- Open Info Request from the ticket. The AI drafts the questions based on what's missing.
- Edit the form, then approve it.
- The customer gets a magic link and fills it in. Their answers go straight into the ticket.
Four tiers, color-coded everywhere.
Triage usually proposes the right tier. Knowing the map lets you double-check it.
Service Desk
Password resets that need admin rights, lockouts, MFA re-enrollment, standard installs, printers, and known fixes.
MSP-T1Systems
Endpoint, Intune, M365/Exchange/SharePoint admin, AD/Entra, app config, and backup or restore. Also anything T1 tried and couldn't fix.
MSP-T2Infrastructure
Network, firewall and VPN outages, servers and hypervisors, site-wide impact, DNS/DHCP, federation, and changes that need change management.
MSP-T3Engineering / Security
Suspected security incidents, production platform, legal or compliance, executive or VIP impact, and critical outages with no known fix.
Engineering-EscalationsMeasure the value delivered, not the ticket count.
The actions that resolve tickets also produce the numbers you manage by. No one has to follow a separate time-tracking routine.
Impact & Effort
Each ticket gets an Impact score and an Effort score from 0 to 100. Both come from fixed reference scales and are scored once, at completion, at temperature 0, ignoring the customer's name and size. That's why you can compare one period with another.
Effort uses hours logged, the number of interactions and how long resolution took, plus an AI read of complexity. Impact uses priority and SLA outcome, plus an AI read of business impact.
Ten password resets and steering a customer out of a security incident don't look the same here, and they shouldn't.
Break it down by queue, customer, work category and staffing tier at /kpis.
Helpdesk digest
A daily or weekly email summary for managers. It only reads from Autotask, never messages techs, and has a hard API budget and a kill switch.
Ticket Review
Each technician's open tickets, stats and what's stuck, with saved AI assessments that build on the previous one.
Account Pulse
Customer health, response rate and outreach, so account conversations start from facts.
The AI recommends. You execute.
The AI can suggest anything. On its own, it can only do things that are scoped and logged, and anything destructive needs a second confirmation.
Destructive = double-confirm
Offboarding, MFA resets, deleting accounts and similar actions need a deliberate second confirmation, so nobody fat-fingers a termination.
Everything under your name
Tool runs are saved as AI-assisted notes attributed to the operator and recorded in a tool audit log. The log protects operators. It isn't there to watch them.
Least privilege per action
A password reset has password permission and nothing else. Writes across tenants need an explicit target inside the customer's family.
SSO in front of everything
Cloudflare Access protects the whole site. Every /api/* call is checked against a signed JWT.
Server-side only
Autotask, AI and M365 credentials never reach the browser. One-time passwords are shown once and never stored.
Fail-closed by default
Automation, Autopilot, the digest and the knowledge curator each have a global off switch. In dev-bypass mode, Automation refuses every live action.
How it's built
Every release ships with a written record listing the tests run, the rollback image and anything that wasn't verified. The 7 Nov 2026 release passed 10,106 tests with zero type errors.
The whole job on one card.
The loop
Queue → Read → Triage → Work the call → Act
The rule
The AI recommends. You execute. Destructive actions take two keys, and everything runs under your name.
Before you act on a caller
Verify them with a push tap or voice keypress, not an SMS or email code. Only approved means go ahead.
When you escalate
Include the tier and reason so the context goes with the ticket. T1 Service Desk · T2 Systems · T3 Infrastructure · Engineering/Security. If you're unsure, choose Engineering.
Safe doors
Never paste a secret into an email: use Secure Send. Never start a "tell me…" email chain: use Info Request.
Tools
Reach for one when triage suggests it: password reset · Scout · JML · Teams/VoIP.
Time
Time records itself on the ⏱ chips. Correct it when it's off. You don't need to stop and log it.
Things people ask first.
Can I break something?
Not anything you can't undo. Destructive actions need a second confirmation, everything is logged, and Autotask is always the source of truth. Run the loop and trust the guardrails.
Triage says escalate, but I think I can fix it. Who's right?
You are. Triage is a recommendation. If the fix is within a CSR's scope (guided, low-risk, with the caller on the line), go ahead. If it needs admin, server or network access, or a security judgment, escalate.
What if the AI is slow or down?
Triage sections run independently, so one can fail while the others still appear. Long waits can be cancelled. If the AI gateway is down, you still have the full ticket and every write-back action.
Does this replace Autotask?
No. Frontline is a faster surface on top of Autotask. Every note, status change, reassignment and escalation is written back. Reporting, billing and contracts stay where they are.
Can the AI take actions on its own?
In the operator console, no. It recommends and a person executes. There are two narrow exceptions. The knowledge curator decides which drafted guidance gets published to the knowledge base. It never touches tickets or customers, admins can override it, and it has its own kill switch. Autopilot, which is rolling out, is limited to its own queue, runs behind a kill switch, treats credential issuance as human-only, and parks finished work for review. Its autonomy is earned per category and is lowered automatically when results slip.
How do we prove it's paying off?
Watch the share of Quick wins in Impact & Effort, the efficiency rate, SLA attainment and median first response at /kpis. Check Ticket Review for escalations that stall. The data is recorded when each ticket is completed, so you can compare any two periods.
Where do I get help?
Your team lead or shift lead first. Escalate tickets by tier. Questions about Frontline itself go to your lead or the Frontline admin.