Key Takeaways
- To choose omnichannel customer service software, start from a 90-day map of how your accounts actually contact you, then test six pass/fail requirements in a two-week trial with real conversations.
- Channel lists don't separate vendors. What happens at the switch does: Gartner found 62% of channel transitions are high-effort, and smooth ones save about 4 minutes of rep time per journey.
- The record model is the hardest requirement to fake in a demo. Only 35% of service leaders say customer data is fully integrated with their service tools (HubSpot, 2024).
- B2B teams need to test shared Slack and Teams channels explicitly. Support for them varies from native to direct-messages-only to integrations that only notify your own team.
- Decide on voice before the shortlist. It splits the market more cleanly than any other requirement.
- Ask every vendor about ownership and roadmap. In 2026 one major vendor in this category agreed to be acquired, and another bought an AI company.
Carmen's help desk contract renews in 90 days. Her team supports 600 B2B accounts. The largest 40 have asked for shared Slack channels, and six of them live in Microsoft Teams. Her CFO wants to know why the AI pilot answers chat but not email. Three vendors have sent decks, and every deck has the same slide: a ring of channel logos around one inbox.
The slide is true for all three. It's also useless, because the logo ring doesn't show what happens when the engineer who opened the issue in Slack is followed up by the admin who emails about it two days later. That's the test that decides whether the next three years go well, and no deck runs it for you.
How to choose omnichannel customer service software: map your channels before you meet a vendor
Choosing omnichannel customer service software starts with your own data, not a feature list. Pull 90 days of conversations and build a channel map by account. It takes an afternoon. It changes every conversation you'll have with vendors.
Business buyers already spread across channels. Salesforce's 2023 survey found business buyers engage with companies across 10 channels on average. Your map shows which of those actually carry your volume, and where your accounts switch.
| Measure (last 90 days) | Why it matters for the choice |
|---|
| Conversations by channel | Which channels must be native, and which can wait |
| Accounts using 2+ channels on one issue | How much the record model will matter |
| Contacts per account who reach support | Whether you need account-level history, not just person-level |
| Accounts with a shared Slack or Teams channel | Whether customer-facing Slack and Teams are requirements |
| Share of conversations by phone | Whether voice is a requirement or a nice-to-have |
| Top 20 questions, and where the answers live | Whether one knowledge source already exists |
Bring the map to every demo and ask the vendor to show your top three switching patterns, not theirs.
Six requirements for omnichannel customer service software, as pass/fail tests
Feature scores hide the failures that matter. Write each requirement as a test that passes or fails on your own data.
Requirement 1: Every channel on your map is native
Native means the conversation arrives in the same inbox, on the same record. Replies go back to the customer in the same channel. Integrations that only create a ticket or notify your team don't count. Shared channels are where this fails most. Zendesk's Slack channel, for example, is built for direct messages, and “one to many (multi-conversation) is not supported”. That's a fine design for one-to-one requests and a failing grade if your largest accounts use shared channels. Our guide to Slack Connect customer support covers what shared channels need.
Requirement 2: One record per account, not per person
Open an issue as one contact and follow up as another contact from the same account, in a different channel. Pass means the second rep sees the first conversation without searching. Fail means an admin has to link them, or the platform only groups by account when you turn on a setting and ask reps to check it. Why omnichannel depends on one record per account explains the difference.
Requirement 3: One knowledge source for every channel and the AI agent
Ask one product question through the AI agent, email and a messaging channel. Pass means the same answer, from the same article, in all three. Fail means the bot answers from its own content and agents from saved replies. The help center holds a third copy. Every added channel then becomes another place for answers to drift. If the AI agent is part of the purchase, how to choose an AI agent for customer service covers grounding and permissions in more depth.
Requirement 4: The handoff carries the whole conversation
When the AI agent or a first-line rep hands off, the next person should get the full conversation, not a summary field. This is where Gartner's 4 minutes come from: “reps aren't asking customers to repeat information they have already provided.”
Requirement 5: Voice is decided, not assumed
If phone carries a real share of your volume, native telephony and call routing are a hard requirement. They narrow the field fast. If it doesn't, don't let a contact-center feature set decide a B2B messaging purchase. Either answer is fine. Leaving it open wastes two demos.
Requirement 6: Your AI tools can act on the data, within permissions
More teams now want their own AI assistants to read and act on support data. Ask whether the vendor's MCP server is generally available, whether it's read-only or read-write, and what limits apply. The answers vary widely: Freshdesk's went generally available in September 2026 with read and write tools, while Help Scout's is read-only by design.
Run a two-week trial with ten tests
Demos show the happy path. A trial with your own conversations shows everything else. Use a real account with its permission, or rebuild one from last month's history.
| # | Test | Pass looks like |
|---|
| 1 | Email, then Slack or Teams, same account, different contacts | Second rep sees the email thread immediately |
| 2 | Same question via AI agent, email and messaging | Same answer, same source article |
| 3 | AI agent escalates mid-conversation | Human sees the full AI conversation |
| 4 | Connect a real shared customer channel | Threads come in with replies going back to the thread |
| 5 | Update one article | Every channel and the AI agent use the new version the same day |
| 6 | Pull one account's history for a CSM | Every channel, every contact, one view |
| 7 | Route by account tier, not just by channel | Enterprise accounts reach the right team whatever channel they use |
| 8 | Report repeat contacts across channels | A number you can trust, per account |
| 9 | Price a real month of AI volume | A forecast the vendor confirms in writing |
| 10 | Export everything | Conversations, records and articles out in a usable format |
Tests one, four and six do most of the eliminating. They're also where integration-based setups show their seams. HubSpot's 2024 research found 74% of service leaders say tool switching makes resolution take longer, and a trial is the cheapest place to find out how many tools your reps would still be switching between.
Questions to ask every omnichannel vendor
Send these in writing before the final demo. Vague answers are answers.
- Which channels on our map are native, which are integrations, and which are in beta?
- Is the record per person, per organization, or something we define? What does a rep see when a second contact from the same account writes in?
- Where does the AI agent get its answers, and is it the same source agents and the help center use?
- What are the limits on shared Slack and Teams channels: per plan, per license, direct messages versus channels, threads?
- How is AI billed, and what did a customer of our size pay last month?
- Is there a generally available MCP server, and is it read-only or read-write?
- Has the company been acquired, or acquired another company, in the last 18 months? What has been committed publicly about pricing and roadmap?
The last question isn't paranoia. In 2026, Zendesk completed its acquisition of Forethought, and Salesforce signed an agreement to acquire Fin, formerly Intercom. The deal is expected to close in Salesforce's fiscal fourth quarter. A three-year contract should know what it's signing into.
Red flags in omnichannel demos
- The demo starts from a fresh ticket. Ask to see a conversation that already switched channels twice.
- “We integrate with Slack.” Ask whether customers talk to you through it, or whether it only notifies your team.
- The AI agent has its own content. Two sources of answers means two answers.
- Account view is a report, not the rep's screen. If a rep has to open another tab to see the account, they won't.
- Pricing needs a call to explain. Get the forecast in writing before you compare seat prices.
Score the shortlist with weights that match your map
Weight the six requirements by what your channel map showed, not by what the vendors emphasized. A starting point for a B2B SaaS team with shared channels and little phone volume:
| Requirement | Suggested weight | Raise it if… |
|---|
| One record per account | 25% | Many accounts use 2+ channels on one issue |
| One knowledge source | 20% | You run an AI agent or plan to |
| Native coverage of your channels | 20% | Shared Slack or Teams channels are common |
| Handoff with full context | 15% | AI or a first-line tier handles most volume |
| Agent access (MCP) | 10% | Your team already works in Claude or ChatGPT |
| Voice | 10% | Phone is more than a small share of volume |
A phone-heavy team should flip the voice weight to the top, and the shortlist will change with it. That's the point of weighting by your own map.
MatrixFlows was built to pass requirements two, three and four by design. Conversations from email, live chat, SMS, WhatsApp, Instagram, Slack and Microsoft Teams attach to records in the same workspace as your knowledge base. The AI agent answers from that knowledge, and a handoff carries the full conversation. It doesn't pass requirement five: there's no native phone channel, so phone-heavy teams should look elsewhere. For the field compared side by side, see our best omnichannel customer service software guide.
The right platform is the one that passes your ten tests on your conversations, not the one with the biggest logo ring. Create a Free Workspace → and run tests one through five on a real account before your next vendor call.