Omnichannel automation pilot: a six-week plan

Short answer

An omnichannel automation pilot should start narrow: two channels, one flow, one outcome measure and six weeks. For example: "automatic replies, lead creation and handover to a manager for price questions arriving on WhatsApp and Instagram". A two-week baseline is measured before the pilot, 30 conversations are read by hand every week during it, and at the end one of three decisions is taken against criteria written in advance: scale up, fix and continue, or stop.

Why "everything at once" fails

In a project that switches on every channel and every flow on the same day, when something goes wrong you cannot find the cause: is it the channel connection, the knowledge base, the routing rule or the team's habits? And the result cannot be measured, because there is nothing left to compare against.

A narrow pilot asks one question and gets an answer: "does this flow improve our result on these two channels?" If the answer is yes, expansion rests on evidence.

Choosing the channels and the flow

Choose channels by enquiry volume: the two busiest channels give the pilot enough data. Choose the flow by these criteria:

• It repeats often — at least a few hundred conversations a week.
• Its rules are clear — the answers are in the knowledge base and exceptions are few.
• Its outcome is measurable — a lead, an appointment or an order, something you can count.
• Mistakes are cheap — complaints, legal or medical questions are not pilot material.

Readiness checklist

1. 1AccountsChannels are connected the official way, and access sits with the company's account, not an employee's personal profile.
2. 2KnowledgePrices, terms, opening hours and FAQs are written down and have an owner.
3. 3OwnersA pilot lead, a knowledge-base owner, a shift lead and one person from sales.
4. 4RulesWho gets handovers, what happens outside working hours, what the stop conditions are.
5. 5TestingA list of 15–20 edge cases, and test conversations excluded from reports.

Baseline: what to measure before the pilot

Take the same measures for the two weeks before the pilot, or at the end you will not be able to say "it improved":

• Weekly enquiry volume per channel.
• Time to first human reply — as a percentage, not an average.
• Conversations left unanswered or without an owner.
• Conversion from enquiry to lead, and from lead to outcome.
• Rough agent time spent on this flow.

How to measure response time correctly across channels is covered in the omnichannel SLA.

A six-week plan

1. 1Week 1: setupChannels are connected, knowledge is loaded, the test list is run. No customers yet.
2. 2Week 2: limited liveThe flow runs in business hours on one channel only. Twenty conversations are read every day.
3. 3Weeks 3–4: two channelsThe second channel is added and out-of-hours mode is switched on. Thirty conversations are read each week.
4. 4Week 5: stabiliseNo new changes — you need a stable period of numbers.
5. 5Week 6: decideComparison with the baseline, a review of the worst 10 conversations and a decision meeting.

Write the decision criteria in advance

The criteria must be signed off before the pilot begins. Criteria written afterwards always bend to fit the result.

• Scale up: the outcome measure beats the baseline, the conversations read show no serious errors, and the team has accepted the flow.
• Fix and continue: the outcome is good but there are one or two systematic errors (a knowledge gap, a routing rule).
• Stop: the outcome is no better than the baseline, or customer complaints have risen.

Managing the risks

• Agents can stop the bot in any conversation — from day one.
• Sensitive topics (complaints, payment problems, health) go straight to a person.
• One person spends 15 minutes a day reviewing — especially in the first two weeks.
• Customers are told plainly that they are talking to a bot.
• A rollback plan: switch the flow off with one action and return to the team's previous way of working.

The design mistakes that come up most often in automated flows are collected in workflow automation mistakes.

After the pilot: the order of expansion

After a successful pilot, the biggest temptation is to switch everything on at once. The safer path is to change one variable at a time: either take the same flow to a new channel, or add a new flow on the same channels — not both together. Each new phase needs only a short baseline and two weeks of observation, because the core rules have already been tested. The knowledge-base owner and the daily review should stay after the pilot too: automated flows go stale most often once the pilot ends and oversight relaxes.

An illustrative example

This is an illustrative example. A furniture company picks WhatsApp and Instagram DM for the pilot, and "price and delivery question → lead → sales manager" as the flow. In the baseline, the first human reply took longer than 15 minutes in half of business-hours conversations, and evening messages waited until morning.

In week four, the conversations reviewed show a systematic error: the bot does not distinguish delivery times by region. The knowledge base is fixed and week five runs steady. The decision is "fix and scale": the next phase adds website chat.

Common mistakes

• Starting without a baseline.
• Changing the rules every week of the pilot — the result cannot be tied to anything.
• Looking only at numbers and never reading conversations.
• Keeping the team out of the pilot, so agents see the flow as "theirs", not "ours".
• Adding a second flow before the pilot is finished.

Limitations

Six weeks do not show seasonal swings: results in a holiday or campaign month will differ. At small volumes numbers move by chance, so read the result alongside the quality of the conversations. A successful pilot does not guarantee the same result on every channel.

How a pilot starts in Vexvon

Connecting in Vexvon takes four steps: choose the channel; link the account through the official API by granting permission (no password is shared, and the permission can be withdrawn at any time); upload service, price and schedule information; check the answers with test conversations and switch the channel to live. There is no need to connect every channel at once — teams usually start with the busiest one. An agent can stop the bot in any conversation, and test conversations from the website widget are kept out of reports. Timings depend on the account and the platform's approval process. More on integrations.

Next step

This week, choose one flow and two channels for the pilot, write down the baseline measures and agree the three decision criteria with leadership. For the day-to-day inbox work, see the omnichannel inbox. We can build your pilot plan together during a demo.