Nothing auto-sends
Client-facing and risky actions are hard-gated — a human releases every one. It’s wired into the product, not a toggle you can forget to turn on.
AI operations desk for MSPs and IT service teams
Handle more tickets per tech without hiring another one
A team of specialized AI employees works your queue, alerts, and QBRs — reading each ticket, grounding in that client’s history, and drafting the reply, script, or report. Your tech approves in seconds instead of writing from scratch, and nothing client-facing sends until a human releases it.
Sign up free and a sample client, a starter roster, and a draft at the approval gate are waiting at first login — no card, no rollout project.
Draft · Tier 1 Tech
Northwind TradersDrafted reply
Re: Can’t connect to VPN after password change
Hi Dana — thanks for flagging this. Your VPN client is still caching the old credential. Here’s the 90-second fix: sign out of the GlobalProtect client, clear the saved password, and sign back in with your new one…
KB: VPN resetticket #4821client profile
Approve Edit Reject
Nothing sends until you approve.
Human-gated
nothing client-facing auto-sends
Isolated
per-client workspaces
Audited
full forensic log
~60 sec
to your first draft
The math you’re already losing
Your best techs are billing senior rates to write the same L1 reply again
You feel this every Tuesday. The cost is a math problem you haven’t put numbers to yet — and all three of these are real. None of them have to stay true.
- The L1 queue never empties: password resets, onboarding and offboarding, mailbox rules, printer and Wi-Fi tickets your team has solved a hundred times.
- Every repeatable ticket costs real minutes twice — read it and find the client context, then write the reply, then document it so the next tech doesn’t start from zero.
- You can’t hire your way out fast enough: a new tech runs $60–90k+ fully loaded before they touch a ticket, then months to ramp on each client’s quirks.
- After-hours and overflow coverage is held together with on-call goodwill, and slipped SLAs cost you in credits, renewals, and reputation.
- And the thing that keeps you off AI: the fear it emails a client something confidently wrong, on your letterhead, and burns a relationship you spent years building.
How a ticket actually moves
The agent does the first 80%. Your tech approves the last 20%.
One walk-through shows both where the minutes come from and where the human stays in control — because it’s the same five steps. The drafting is where the time is saved. The gate is where nothing goes rogue.
- 01
Intake
A ticket or alert lands. The Dispatcher reads it, identifies which client it belongs to, and routes it to the right agent.
- 02
Grounding
The agent pulls that client’s history, docs, and past fixes from their isolated workspace — the answer for that client, never a generic one.
- 03
Draft
It writes the actual next move — a customer reply, a remediation script, an escalation package, a QBR section — in your team’s voice.
- 04
Gate
It stops at an approval gate with its sources attached, so review is a glance, not an investigation.
- 05
Release
A human approves, edits, or rejects in Slack, web, or SMS — and only then does anything send or run.
Reject it and the reason becomes a lesson the agents remember, so the same miss doesn’t come back next week. Your tech reviews in seconds instead of writing and documenting from a blank page.
Why the savings don’t blow up in your face
AI never goes rogue on a client — a human releases every move
Your biggest objection, answered before the upside. Approval-native isn’t a setting you can forget to switch on — it’s how the product is built. A wrong draft is a reject that never left the building, and the agents get more aligned to how your desk works, not less.
Per-client isolation
Every ticket, doc, memory, approval, and tool call is scoped to one client workspace, so a draft for Client A can never reach into Client B’s data or systems.
Fully auditable
A forensic log shows what each agent did, what context and tools it used, and who approved it — so you can show a client exactly what happened in their environment.
You set the gates
Decide which actions need sign-off, which agents may draft, and raise the bar for anything touching a production system or a client mailbox.
Do the math yourself
See the time saved per ticket — with a calculator you control
No fabricated stats. No email gate. TandemDesk assists — it does not replace techs — so the unit of value is minutes reclaimed per assisted ticket, multiplied by your real volume. On conservative defaults a typical desk lands near breakeven on reclaimed L1 work alone; the return climbs from there based on how repeatable your queue actually is. Every input is yours to change.
Illustrative model · your numbers
4
18
50%
Floor we use is 35%. Most desks are more repeatable than that.
6 min
After a human reviews and approves — the gate time is already subtracted.
$32
21
Reclaimed-time value vs. plan cost
1.2×$420/mo net
$2,419/mo of reclaimed tech time on the Operations plan ($1,999/mo).
76 hrs
Tech hours reclaimed / mo
0.44 FTE
FTE-equivalent freed
$2,419
Reclaimed value / mo
$5,042
Net value / year
756 assisted tickets/mo of 1,512 total.
Start free and run your first draft- Hours, dollars, and FTE are the same reclaimed time shown three ways — not separate gains. Don't add them.
- Reclaimed time is capacity, not a cash refund — it shows up as deferred hiring, less overtime, or more clients per tech.
- Counts only reclaimed handle & documentation time. After-hours coverage, SLA penalties avoided, and new-client revenue are real but excluded — upside, not headline.
The team you’re renting
Seven specialized roles, priced like software, working like staff
Start with three roles on Pilot to prove one workflow. Run the full seven on Operations for one flat price — not seven salaries. Every role drafts; your team reviews and releases.
01
Dispatcher
Triages and routes every ticket and alert, identifies the client, and keeps handoffs from drifting.
02
Tier 1 Tech
Drafts the repeatable fixes and customer replies that clog your L1 queue — the bulk your seniors shouldn’t be touching.
03
Tier 2 Tech
Deeper diagnostics, scripts, and clean escalation packages so handoffs don’t lose context.
04
vCIO
Turns ticket history into QBRs, client health, roadmaps, and renewal briefs that usually eat a delivery manager’s week.
05
CISO
Maintains risk registers and policy reviews, and turns findings into accountable remediation work.
06
Security Engineer
Triages alerts, enriches incidents, and drafts containment timelines.
07
Service Manager
Watches SLA drift, queue load, and handoffs so nothing quietly breaches.
Works with the stack you already run
Sits alongside your PSA and RMM — not a rip-and-replace
You keep your system of record. TandemDesk handles the drafting and triage beside it, reaching your tools through native connectors plus Nango, MCP, OpenAPI, and HTTP actions. Wiring intake and approvals is the whole first setup — no migration project, no retraining off the tools your team already knows.
Bring your own stack: native connectors plus Nango, MCP, OpenAPI, and HTTP actions reach the tools you run today.
Connect ticketing, RMM, identity, email, and Slack so agents read real context and route approvals where your team already lives.
Keep your system of record — TandemDesk drafts and triages beside it, with no data migration and no rebuild of your workflow.
Scope every connector per client, so credentials and actions stay inside the right tenant.
Live in 60 seconds
A sample client and a draft waiting to approve at first login
Feel the whole loop before you connect a single tool. No big-bang rollout, no long onboarding project to get the first win.
01
Sign up free and land on a sample client with a starter roster and a draft already at the approval gate — approve, edit, or reject it on the first screen.
02
You feel the whole loop — read, ground, draft, gate, release — before you connect a single tool.
03
Start with one client and three agents (Dispatcher, Tier 1, vCIO) to prove one workflow, then expand at your own pace.
04
No big-bang rollout, and no autonomy is ever switched on quietly — you enable each agent and each action on purpose.
Priced against a hire, not a seat
One flat plan costs a fraction of the tech it frees up
Pricing tracks usage and rollout stage, so spend lines up with the volume you actually assist. Plug in your numbers to see which plan pays for itself fastest.
Pilot
start here$799per month
Prove the savings on one queue before a wider rollout.
- Best for
- One queue, one team lead, up to 5 clients
- Includes
- 3 active AI employees
- Dispatcher plus two technician or advisory agents
- Up to 5 client workspaces
- 500 assisted actions per month
- Ticketing, docs, email, and Slack connectors
- Approval queue, audit log, and spend controls
Operations
most common$1,999per month
Run the full roster across service, security, and advisory work.
- Best for
- Active MSP desk, up to 25 clients
- Includes
- 7 active AI employees
- Full roster for support, security, advisory, and service management
- Up to 25 client workspaces
- 2,500 assisted actions per month
- RMM, identity, security, and vCIO workflows
- SLA risk, client health, and workflow tuning
Enterprise
custom rolloutCustomstarts at $4,999/mo
Custom rollout, deeper controls, and unusual systems for larger MSPs.
- Best for
- Multi-team or high-volume MSP
- Includes
- Custom roster and usage limits
- Expanded client workspace options
- Custom connector and workflow design
- SSO, role controls, and evidence exports
- Priority implementation and workflow reviews
- Procurement, security, and compliance support
Straight answers
The questions an MSP owner actually asks before trusting AI on the desk
Can TandemDesk ever email a client on its own?
No. Client-facing and risky actions are hard-gated — a human approves, edits, or rejects every one before anything sends or runs. It’s wired into the product, not a toggle you can forget to turn on. A wrong draft is a reject that never left the building.
Will my clients’ data mix together?
No. Every ticket, doc, memory, approval, and tool call is scoped to a per-client workspace with no cross-client access. A draft for Client A cannot reach into Client B’s data, credentials, or systems — and the audit log shows exactly what context each agent used.
Is this replacing my techs?
No. TandemDesk drafts the reply, the notes, the script, and the escalation, then stops at the approval gate. Your techs review and release in seconds instead of composing and documenting from scratch. The result is freed capacity — more tickets and more clients on the same headcount — not headcount cuts.
Where do the “minutes saved” numbers come from?
They’re your inputs in the calculator, not borrowed stats. “Minutes reclaimed” is time-to-compose-from-scratch minus the time a human still spends reading, editing, and approving the draft — so the review step is already subtracted, not added on top. The defaults are labeled illustrative and set deliberately low, so on conservative numbers a typical desk lands near breakeven on reclaimed L1 time alone; the return scales with how repeatable your own queue actually is.
What if a draft is wrong?
You reject it with a reason, nothing sends, and the agent records the correction as a lesson it won’t repeat. Over time the drafts get more aligned to how your desk actually works.
Does it work with my PSA and RMM, and how fast can I start?
Yes — through native connectors plus Nango, MCP, OpenAPI, and HTTP actions, working alongside what you already run. There’s no migration project and no rebuild of your workflow. Sign up free and approve your first draft in about a minute; wiring your real intake and approvals is the first setup step after that.
What does it actually cost compared to a tech?
Operations is $1,999/mo for the full seven-role roster and up to 25 clients — roughly a third of what one fully-loaded tech costs you each month, against capacity across your whole desk. A new tech runs $60–90k+ fully loaded before they touch a ticket. Pilot starts at $799/mo to prove it on one queue; Enterprise starts at $4,999/mo for SSO, role controls, and evidence exports.
Run your numbers, then run a ticket
See the math on your queue, then approve your first AI draft
More tickets per tech. More clients without more headcount. Every client-facing move still human-approved. Begin with one client and a starter roster — no rip-and-replace, no rollout project, no auto-send risk.
Most teams land on Operations once the loop is proven; Enterprise buyers can talk through scope for SSO, role controls, and evidence exports.