Start with the PSA you already run: ConnectWise's AI agents work inside ConnectWise PSA and are priced by quote, while Halo says its AI is included in the per-agent price. Either way, test first on one set of about 60 closed tickets, counting wrong-client errors, routing, priority and summaries before anything else.
Agent names vary between articles. ConnectWise's own pages name Teams, Email, Voice and RMM agents plus service desk agents such as Company, Board, Dispatch and Oversight, and they don't use labels like Triage Agent or Data Analyst Agent, so check any write-up that does. Halo's site now calls its product simply Halo, with a Managed Services module for PSA work. Everything below comes from the two vendors' own pages as of 1 November 2026, so treat each capability as a claim until your ticket test confirms it.
ConnectWise's agents and Halo's AI, as each vendor describes them
ConnectWise splits its AI into agents that take requests in, agents that work the queue and an agent that works on endpoints. These rows come from its AI agents pages.
| ConnectWise agent | What the page says it does | What it runs on |
|---|---|---|
| Teams Agent | Takes a request in a Teams chat; the user can review and edit it before the ticket is created | PSA |
| Email Agent | Reviews emailed tickets, asks the sender for missing details and adds the answers to the ticket | PSA |
| Voice Agent | Answers inbound support calls, gathers details and creates a request with the call transcript; calls can be transferred to an on-call technician by configured criteria | PSA |
| Company, Contact and Agreement agents | Identify the customer, the requester and the service agreement | PSA |
| Urgency and Priority, Board and Dispatch agents | Set urgency and priority, route to a service board, and assign work using ticket context and calendar availability | PSA |
| Summary Line, Insight and Oversight agents | Write a concise ticket summary, find trends and risks, and monitor workflows to flag SLA risks | PSA |
| AI Assistant | Helps technicians troubleshoot, suggests next steps and drafts notes, updates and time entries | PSA |
| RMM Agent | Investigates device issues using live endpoint context; the page says technicians approve any endpoint changes | PSA and RMM |
Halo organises its AI differently: conversational agents people talk to, process agents that act on events, and intelligence built into the ticket. These rows come from Halo's AI page and its Managed Services page.
| Halo feature | What the page says it does |
|---|---|
| End-User Virtual Agent | Works across web chat, Teams, Slack and the self-service portal, triaging and logging requests, guiding self-service and escalating with full context; a Chat Profile sets persona, knowledge scope, tone, escalation logic and channel |
| Fulfiller Operations Agent | Sits inside the ticket and works alongside your technicians |
| Process agents: Dispatch, 1st Line Support, Major Incident, Knowledge, Change Management | Respond to events on a schedule; auto-resolve known issues, flag service degradation, coordinate multi-step changes; each runs within your role-based access controls |
| AI Suggestions | Learns how similar tickets were resolved and suggests categorisation, prioritisation, team routing and SLA-breach warnings inside the ticket, each switched on separately |
| Matched tickets and knowledge; Platform Themes | Surfaces earlier tickets and articles on each new request; uses unsupervised machine learning to surface high-volume patterns |
| Feature library | AI report analysis, satisfaction surveys with sentiment flags, thank-you detection and knowledge base creation from tickets |
| Agent interoperability | Native MCP servers so external agents such as Claude, Copilot and Codex can connect, plus a REST API |
Two differences show up before any test. ConnectWise describes phone calls, chat and email as intake channels, including a voice agent. Halo's AI page names web chat, Teams, Slack and the portal for its virtual agent, and lists call management only as a channel, with no AI voice agent described, so ask Halo how phone requests would reach the AI if your clients mostly ring. And ConnectWise's agents split the ticket-handling job into small named steps, while Halo pairs a handful of larger process agents with suggestions you switch on one at a time. Neither design is better by itself; they fail in different places.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowPricing and trials: quote versus published
| Question | ConnectWise AI agents | Halo |
|---|---|---|
| Price shown | None. The pages offer Request pricing, Get pricing and Talk to Sales | A per-agent calculator with a region and licence-type choice. The figure my fetch returned was in a non-USD currency, so read the USD figure for your size yourself; an FAQ also says to talk to Halo for a published figure |
| What "agent" means | An AI agent that does work | In Halo's pricing, a licensed human user: the page offers named and concurrent licences, which are seat types |
| AI priced separately? | Not stated on the pages I read | Pages say included as standard with no separate AI licence, yet the module list and trial form label Halo AI an optional feature |
| Trial | A 15-day Proof of Value, no credit card, eligibility requirements apply | A free 30-day trial with all features and modules, no credit card |
| How the trial starts | ConnectWise analyses your PSA data and builds a Partner Value Snapshot, prepares the environment and trains the model on your service data, then the 15 days begin | You start a Halo instance and configure it |
| Onboarding | Not described on the pages I read | Onboarding packages are required with purchase at a bespoke cost depending on scope; Halo's MSP FAQ says most MSPs are live in 4 to 6 months, smaller teams faster, with no charge until you are ready |
| What you must already run | ConnectWise PSA, plus RMM for the RMM Agent | Halo as your PSA |
The confusing word in that table is "agent". When Halo's pricing page says per-agent pricing it means seats for your technicians, not AI agents, so a five-technician MSP pays for five seats whether or not it switches the AI on. ConnectWise's pricing is quote-based, and its Proof of Value is the front door to that quote. Ask each vendor for USD pricing for your exact technician count, billing basis and term, and ask Halo to confirm in writing that the AI features you test are included.
Where a small MSP should start, situation by situation
| Your situation | Test first | Why |
|---|---|---|
| Already on ConnectWise PSA, with 6 to 15 technicians | ConnectWise agents | They run inside what you have, so no migration; the Proof of Value starts from your own ticket data |
| On another PSA, with 3 to 8 technicians, and a move is already being discussed | Halo's trial | AI is a reason to look, not a reason to migrate; Halo's own MSP page says go-live typically takes 4 to 6 months |
| On ConnectWise but unhappy with the price | Both | Run the same ticket set through each, then weigh migration cost against what the AI saves |
| Most requests arrive by phone, including out of hours | ConnectWise Voice Agent, plus a question to Halo | ConnectWise's pages describe a voice agent, while Halo's AI page lists call management as a channel but describes no voice agent; an AI receptionist for an MSP is a separate route if your PSA has no voice agent |
| Alert noise from your RMM is the biggest time sink | ConnectWise RMM Agent, and Halo's process agents | Both claim to act on events, so test them on your real alert types, with technician approval switched on |
The wider field of tools, including help desks that aren't PSAs, is covered in the best AI help desk tools for small MSPs. This comparison stays on the two PSA vendors because the PSA is the system of record for tickets, agreements and billing, and moving it is the costliest decision on the list.
Build the ticket set before either trial starts
Pull 60 closed tickets from the last 90 days and write down the right answer for each before any tool sees them. The mix matters more than the count, because a set made only of password resets will flatter every product.
| Ticket type | Count | What it tests |
|---|---|---|
| Password resets and account lockouts | 15 | Speed and safe handling of the simplest work |
| New starter and leaver requests | 10 | Multi-step requests that need the right company and the right agreement |
| Alert-driven incidents (backup failures, offline devices) | 10 | Priority, routing and whether the AI acts on noise |
| Vague requests such as "it's slow" | 10 | Whether the tool asks for detail or guesses |
| Look-alike clients (similar names, shared contacts) | 8 | Wrong-client errors, the costliest mistake |
| Sensitive or urgent (suspected phishing, a director's request) | 4 | Escalation to a human instead of automation |
| Duplicates and follow-ups to open tickets | 3 | Whether it merges or opens a second ticket |
Record the ground truth in a sheet with one row per ticket, so scoring is a lookup rather than a debate.
ticket_id, text_redacted, correct_company, correct_contact, correct_board, priority_1_to_4, ticket_type, expected_next_step, safe_to_automate
T001, "Locked out of email since 8am...", [Client A], [Contact 1], Service Desk, 3, lockout, "Verify identity then reset", Y
T047, "Is this email from our director real?", [Client F], [Contact 9], Security, 1, suspected phishing, "Escalate to senior tech, do not auto-reply", NThe two vendors run trials differently, so agree in writing how your 60 tickets will pass through each. ConnectWise's Proof of Value starts from your PSA data and has technicians use agents in their normal workflows. Halo's trial is a fresh instance. Ask each for a way to submit the test tickets through the ordinary email or chat intake of the trial environment, and record every field the AI fills in or changes. Ticket text contains client information, so redact names and credentials, and check your client contracts before any vendor processes live tickets.
Scoring: the errors that matter most
Not all misses cost the same. A ticket filed under the wrong category wastes a minute. A ticket filed under the wrong client can leak information or bill the wrong company. Set gates before the first ticket runs.
| Metric | Pass mark (my rule of thumb) | Gate |
|---|---|---|
| Company and contact identified correctly | 95% or better | Any wrong-client error in the 8 look-alike tickets stops the test |
| Board or category correct | 90% or better | None |
| Priority within one level | 90% or better | A P1 marked low priority is a fail |
| Summaries with no invented fact | All of them | One invented fact fails that ticket |
| Suggested next steps are safe | All of them | Advice that skips approval for a risky change is a fail |
| Sensitive tickets escalated, not auto-handled | 4 of 4 | A miss stops the test |
This invented scorecard shows how the gates change the answer. The figures say nothing about either product.
| Metric (60 tickets) | Tool 1 | Tool 2 |
|---|---|---|
| Company and contact right | 57 | 55 |
| Wrong-client errors in the 8 look-alikes | 0 | 2 |
| Board or category right | 53 | 55 |
| Priority within one level | 54 | 52 |
| Summaries with an invented fact | 1 | 3 |
| Sensitive tickets escalated (of 4) | 4 | 3 |
Tool 2 files more tickets in the right category, which looks better on a dashboard. It also sent two look-alike tickets to the wrong client and failed to escalate one sensitive ticket, so it fails two gates and the test stops there. Tool 1's single invented summary fact still needs fixing, but it passes every gate. The rule to take from this: choose on the worst thing a tool does, not its average.
One look-alike ticket, followed through both tools
Take a ticket from your look-alike group. An invented example: an email arrives from a contact at "Northfield Dental Group" saying the practice server is down and patients are waiting. The MSP also supports "Northfield Dental Care", and one contact appears on both clients' lists. The right answer is Northfield Dental Group, the managed-service agreement with out-of-hours cover, and priority 1.
On ConnectWise, check each step the pages describe. Did the Company Agent choose the right company, and the Contact Agent the right requester? Did the Agreement Agent attach the agreement with out-of-hours cover rather than a break-fix one? Did the Urgency and Priority Agent score it as priority 1, and did the Dispatch Agent pick an available on-call technician? Does the Summary Line Agent's one-liner say "server down, patients waiting" and nothing invented, such as a cause?
On Halo, check the same outcomes through its mechanisms. Were the category and priority suggestions made inside the ticket, and who accepted them? Did the matched-tickets feature surface last month's server ticket for the right client and not for the similarly named one? If a process agent acted, what does the audit log show about what it saw and what it did? Run the ticket three times, because a tool that picks the right company twice and the wrong one once has a problem you would otherwise find on a busy Monday.
Approval, audit and kill-switch checks
Both vendors say you stay in control. Check what that means in practice, because the wording differs and the gaps are where incidents start.
| Control | ConnectWise, per its pages | Halo, per its pages |
|---|---|---|
| Human approval | Technicians approve endpoint changes made by the RMM Agent; a Teams user can review and edit a request before the ticket is created | Agents act within role-based permissions you define; the page doesn't say whether a person must approve process agents' actions first |
| Log of what the AI did | AI-driven actions can be monitored, tracked and reviewed within service workflows; an Oversight Agent monitors workflows and flags SLA risks | Halo's MSP page lists full prompt and response audit logging |
| Switching an agent off | Not described on the pages I read | Admin-level kill switches are listed on the MSP page |
| Identity and permissions of the AI | Not described on the pages I read | Each process agent runs within existing role-based access controls, which define its identity and permissions |
| Data handling | The AI pages link to a trust page and describe nothing further | Isolated tenant boundaries, data never used to train third-party models, and PII redaction, per the MSP FAQ |
Where a cell says the pages don't describe something, that is a question for the vendor, not a verdict. Ask both for the same five answers in writing: which actions can run without approval, where the log lives and how long it is kept, how to disable one agent without disabling all of them, what permissions the AI runs with, and what happens to your ticket data when you leave. A routine for running a tool in shadow mode before anything touches a customer is in how to pilot AI in shadow mode, and it fits this test well.
A break-even sum for a five-technician MSP
Neither vendor publishes a number you can plug in, so use a formula and your own quote. A technician costs the MSP $42 an hour fully loaded, there are 21 working days a month, and a quote arrives at $150 per technician per month. Both figures are invented.
The break-even is the monthly price divided by the hourly cost, times 60, divided by working days: 150 / 42 x 60 / 21 = 10.2 minutes per technician per day. For five technicians that is about 51 minutes of team time a day, or $9,000 a year in licence cost. If the AI plus the checking it needs saves more than that, it pays; if checking eats the saving, it doesn't.
ConnectWise's own page quotes a partner saying break-even on its AI spend is now about 16 minutes per technician per day. Treat that as a prompt to work out your own figure, not as a benchmark. For Halo, take the USD per-agent price from the calculator at your technician count and run the same formula, remembering that the seat is paid whether or not you use the AI.
The saving only counts when it becomes recovered capacity: more tickets closed per technician, fewer escalations to a senior, or after-hours cover you no longer pay for. How to set pilot success criteria that hold up helps you fix those measures before the trial rather than after.
After a pass: a staged rollout
Passing the gates earns a limited rollout, not a switch-on. Start with the AI proposing and a technician acting, for one week, comparing every proposal with what the technician would have done. Then let it act alone only on the lowest-risk category, which in the sample mix is password resets and lockouts, a quarter of the tickets. Keep sensitive tickets, anything touching a look-alike client and every RMM change on human approval until a month of clean results says otherwise.
Fitting the test into a 15-day and a 30-day trial
The shorter window sets the pace, so plan the ConnectWise Proof of Value first and let Halo's 30 days stretch around it.
- Before day one. Fix the 60 tickets and the gates, agree in writing how each vendor will receive the tickets, and get permission for any live data.
- ConnectWise, days 1 to 3. Compare the Partner Value Snapshot with your own analysis of where tickets go. If ConnectWise's estimate and yours disagree widely, find out why before trusting either.
- ConnectWise, days 4 to 12. Technicians use the agents in normal workflows while someone checks every wrong-client and priority decision daily. Run the 60-ticket set in the middle of this period.
- ConnectWise, days 13 to 15. Score against the gates and ask for pricing for your technician count.
- Halo, weeks 1 and 2. Configure the basics, switch on AI Suggestions one at a time, and push the 60 tickets through email intake, recording each field.
- Halo, weeks 3 and 4. Test the virtual agent on a portal with a handful of staff, check the audit log, and score. Ask for USD pricing and the onboarding package scope.
If you want a general routine for turning ticket tagging and routing on without surprises, AI ticket triage: tag, route and prioritise support requests covers the mechanics, and the gates above are the part specific to an MSP with many clients.
ConnectWise or Halo AI: questions before a trial
Do ConnectWise's AI agents work without ConnectWise PSA?
ConnectWise's pages describe the agents as running inside its platform and PSA workflows, and the RMM Agent as connecting PSA and RMM. They don't describe a version that works on another PSA. If you run a different PSA, ask ConnectWise in writing whether the agents can be used at all before you spend time on its Proof of Value.
Is Halo's AI really free?
Halo's pages say AI is included as standard and that there is no separate AI licence, priced per agent with no tiers. Its module list and trial form, however, label Halo AI an optional feature. That is probably a switch you turn on, but ask Halo to confirm in the quote that the AI features you want carry no extra charge.
How long are the two trials?
ConnectWise describes a 15-day Proof of Value that begins after it has analysed your PSA data and prepared the environment, with no credit card required and eligibility requirements that apply. Halo offers a free 30-day trial with all features and modules included and no credit card. Plan your ticket test to fit the shorter window.
Will either tool resolve tickets without a technician?
The vendors word it differently. ConnectWise's service desk page describes classifying, routing and assisting technicians, and says its RMM Agent's fixes are approved by your team. Halo's page says its process agents can auto-resolve known issues within the permissions you define. Whatever the wording, decide which actions need human approval before you switch anything on.
Can I use real client tickets in a trial?
Only if your client contracts and data-processing terms allow a vendor to process them. Tickets often contain names, device details and sometimes credentials. Redact them for any test you run yourself, and for a vendor-run trial that uses your live data, get the sub-processor and retention terms in writing first.
Further reads
- Is AI Ticket Automation Worth It for a Small IT Support Shop? — Whether AI ticket automation pays off for a small IT shop.
- How Small IT Support Businesses Use AI to Resolve Tickets Faster — How small IT support firms use AI to resolve tickets faster.
- MSP Automation Ideas: Five AI Workflows and Their Set-Up Time — Five MSP workflows with realistic set-up times.
- AI IT Documentation for MSPs: Client Runbooks From Screen Recordings — Client runbooks from screen recordings.
- How to Evaluate an AI Software Vendor: A Small Business Scorecard — A scorecard for judging any AI software vendor.
- How to Run an AI Software Demo So You See the Real Product — How to run a demo so you see the real product.
- Automated Managed Service Provider: Hardware Refresh Plans From Asset Lists — How a small managed service provider groups a client's asset list into replace-now, next-year and later buckets, then builds the budget table.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: ConnectWise AI agents pages (overview, service desk agents, Teams, email and voice agents, Proof of Value), ConnectWise's automated ticket triage page and its June 2026 AI agents post; Halo's AI page, Managed Services page and pricing page.