From Strategy to Execution: The 90-Day Plan for AI Sales Automation

Many AI projects in sales do not fail because of the technology. They fail because nobody decides who does what by when. This 90-day plan for AI sales automation splits the introduction into three months with clear tasks and checkpoints. It is meant for small and mid-sized businesses starting without their own IT department.

Before the plan starts: three decisions

The plan assumes that three things have been settled. If one of them is missing, it is better to push the start back by two weeks.

• One use case, not five. For example: taking calls outside office hours and booking appointments. Or: answering enquiries from the website form straight away and pre-qualifying them.
• One person in charge. Someone in the business who knows the process and is given a few hours a week for it. This is rarely the managing director.
• A baseline. How many enquiries come in per week today, how many go unanswered, how long does the first reply take? Without these figures you will have nothing to compare after 90 days.

To illustrate, we will follow a fictional heating and plumbing business with twelve fitters and two office staff. Its problem: during the day the office can hardly be reached by phone, and enquiries that come in after hours are only dealt with the next morning.

The 90-day plan for AI sales automation

Month 1: groundwork and selection (day 1 to 30)

Nothing is automated in the first month. You gather what the AI employee will later need to know, and you choose the provider.

1. Week 1: current state. For one week, every call and every enquiry is logged: time, channel, request, outcome. In the plumbing business, for instance, it turns out that a large share of calls are about appointments and emergencies.
2. Week 2: common questions and rules. The office staff write down the twenty most common questions with their answers and define what counts as an emergency and who gets called in that case.
3. Week 3: check the data. Are the calendar, customer list and CRM up to date? An assistant that books into an outdated calendar does more harm than good. There are pointers in the article Preparing sales data for AI.
4. Week 4: choose a provider. Test two or three providers with your own questions and emergency rules, check data protection and the contract, then decide.

By the end of the first month you will have: the analysis of the baseline week, a document with questions, answers and forwarding rules, a cleaned-up calendar and a signed contract including a data processing agreement. It sounds like paperwork, but it is half the battle. In the plumbing business, for example, this is when it comes out that two fitters still keep their appointments in a private calendar. Better to find gaps like this now than in week seven, when the assistant books the first customer at the wrong time. Put each checkpoint as a fixed appointment in the calendar of the person in charge, otherwise the preparation slips behind day-to-day business.

Month 2: setup and test run (day 31 to 60)

Now it gets concrete. With many providers the technical setup itself is quick. At Neurobots a certified partner typically sets up the AI employee in about 14 days, including the connection to calendar and CRM. The rest of the month is for testing.

Week	Task	Checkpoint at the end
5	Setup, connection to calendar and CRM	Test appointments land correctly in the calendar
6	Internal test: the team calls in themselves and plays through difficult cases	Emergencies are forwarded reliably
7	Go live on one channel, for example only after-hours calls	Every conversation is reviewed daily
8	Refine answers and rules	List of open errors has been worked through

What matters in this phase: the assistant introduces itself as an AI. Customers should know who they are talking to, which is also what the EU AI Act requires. And in the first weeks the person in charge genuinely reads every conversation. It is tedious, but it is the only way to find mistakes before customers complain. The article on the AI pilot project in sales describes how to set up a test run like this properly.

Month 3: expand and evaluate (day 61 to 90)

Once the first channel runs smoothly, the next one is added. In the plumbing business that would be the daytime calls the office cannot get to, and after that WhatsApp. In parallel, the focus is on the team.

• Weeks 9 and 10: launch a second channel or a second use case, such as appointment reminders the day before.
• Week 11: give everyone who works with the results a short briefing. Fitters need to know how appointments are now created, and the office needs to know where to find the call logs. More on this in the guide to training your sales team.
• Weeks 12 and 13: evaluate against the baseline and decide on next steps.

How to measure success after 90 days

To be honest, nobody can tell you in advance how much your figures will change. It depends on where you start. A business that was not reachable at all after hours will see the difference sooner than one that is already well organised. So compare these values with your baseline from week 1:

• Share of enquiries that get a reply, and time to first reply
• Number of appointments booked or qualified enquiries per week
• Conversations the assistant had to hand over to a person, and why
• Complaints or drop-offs caused by the assistant
• Time the office estimates it now spends less on the phone

The article Measuring and communicating AI ROI shows how to present these figures so that co-owners or the bank can follow them too.

When this plan does not fit

For a one-person business with only a few enquiries a week, a 90-day plan is overkill. A week of preparation and a month of testing are enough. Conversely, a company with several locations, its own inside sales team and a complex CRM will tend to need more time, mainly for coordination and data protection. And if the business is in the middle of introducing a new ERP system, wait. Two changeovers at once will overwhelm any small team.

Frequently asked questions

Can't this be done faster than in 90 days?

The technology can be in place in about two weeks. The article How to implement AI sales automation in 14 days describes what that looks like in practice. You need the 90 days for preparation, testing and getting used to it, so that the assistant really holds up in everyday work.

How much time does my team need to invest?

In the first month, expect a few hours a week for the person in charge, and a little more in the second month because of the conversation reviews. In the third month the effort usually drops considerably.

What if the test goes badly in the second month?

Then the plan has worked exactly as intended, because you noticed early. Check whether the problem lies in the stored answers, the data or the provider. A short trial period agreed in the contract makes it easier to pull out.

Conclusion

A good 90-day plan is unspectacular: one use case, one person in charge, one baseline, then set up, test and expand step by step. If you prepare carefully in the first month and read every conversation in the second, after three months you will have a solid basis for deciding what comes next.

The page Digital assistant for trade businesses shows how an AI employee handles calls and appointment requests in the skilled trades.