A baseline is a short record of how your enquiries work today, taken before the chatbot goes live, so that you can measure its impact honestly later. Four weeks of six simple numbers is enough: enquiries by channel, the share arriving out of hours, time to first reply, the top questions, how many enquiries became customers, and hours spent answering. Without them, any later claim about what the chatbot did is a guess.
Most guides on measuring chatbots start after launch. That is too late for this part, because a chatbot's dashboard begins counting the day you install it. The "before" numbers only exist if you write them down.
Why the before numbers go missing
Three things make baselines easy to skip. Enquiries arrive through several channels, so nobody has the total. Reply times feel fine until someone measures them. And once a chatbot is live, it is tempting to compare its numbers with memory: "we used to get loads of emails about delivery." Memory is a poor baseline.
Recording takes about an hour to set up and ten minutes a week. The rest of this guide is what to record and how.
The six numbers to record
Take these for four consecutive weeks before launch. Avoid your busiest and quietest seasons if you can, or at least note which one it was.
| Measure | How to record it | Why it matters later |
|---|---|---|
| Enquiries per week, by channel | Count contact form submissions, enquiry emails and calls, separately | Shows whether total demand changed or just moved channel |
| Out-of-hours share | Of those enquiries, how many arrived outside opening hours | The part a chatbot can answer that nobody else can |
| Time to first reply | For 20 enquiries, the gap between arrival and your first response | A clear before and after for speed |
| Top 20 questions | Tally the questions behind each enquiry for a fortnight | Becomes the bot's test set and its training checklist |
| Enquiries that became customers | Of the enquiries in the window, how many bought or booked | Separates more enquiries from better enquiries |
| Hours spent answering | A simple diary for one typical week | The time a chatbot might give back |
Keep it in one spreadsheet with a row per week. Approximate is fine, as long as you count the same way before and after.
How to count without a system
Most small businesses don't have a helpdesk that reports any of this. You don't need one.
- Contact form: most form tools and website builders keep a list of submissions with dates and times. Count by week.
- Email: search your inbox by week for enquiries, or add a label as each one arrives.
- Phone: a tally sheet by the phone is enough. You only need the count and whether it was in hours.
- Reply time: pick 20 enquiries across the four weeks and note arrival and first reply times.
- Questions: jot one line per enquiry, then group them at the end. "Do you deliver to Ireland?" and "shipping to Dublin?" are the same question.
The question tally is the most useful by-product. It tells you what to teach the chatbot before launch, and it gives you a ready-made test set. Our guide on whether your chatbot is working uses the same list after launch.
What the chatbot will measure after launch
Once the chatbot is live, its own reports take over part of the job. In SpideyChat, the dashboard shows:
- Conversations per day and AI resolutions, over the last 90 days.
- Conversation outcomes, split into resolved, handed off and still active.
- A conversion funnel from started to engaged to resolved to lead.
- Message activity by hour of day over the last four weeks, which you can set against your out-of-hours baseline.
- Leads captured (on Starter and above), each with contact details and the opening question.
- Knowledge gaps: questions the bot couldn't answer well, which you can compare with your top 20 list.
These are chatbot numbers. They tell you what happened in the chat, not what happened to your inbox or phone. That is why the baseline matters: the result you care about is the change across all channels, and only your spreadsheet covers the "before" side of that.
Comparing fairly afterwards
When the chatbot has run for four to eight weeks, record the same six numbers again in the same way. A few rules keep the comparison honest:
- Same length, similar season. Four weeks against four weeks, and not January against December for a retailer.
- Note other changes. A price change, an advert, a new page or a staff holiday during either window affects enquiries. Write them in the spreadsheet.
- Count all channels. If emails fell and chat conversations rose, demand may simply have moved. That can still be a win if replies got faster or time was saved, but say so.
- Look at outcomes, not just volume. More conversations is not the goal. More enquiries that become customers, faster replies and fewer hours spent are.
Once you have before and after numbers, turning them into money is a separate step. Our guide to calculating the ROI of an AI chatbot covers that, and the free chatbot ROI calculator does the arithmetic.
Start the clock this week
You can begin today, before choosing a tool. Set up the spreadsheet, put a tally sheet by the phone, and start labelling enquiry emails. In four weeks you'll have a baseline, a top 20 question list, and a much clearer idea of what you need a chatbot to do.
When you're ready to see what the reports look like after launch, the features page covers the analytics, lead capture and tickets in more detail.