An AI-Based BI System for Your Business: See Where Customers Come From

Contents

1. The real question behind "we need BI"
2. An honest note before you start: a dashboard without clean sources is decoration
3. The numbers an owner should see once a week
4. How to get there: seven steps
5. When you really need a "big" BI tool
6. What this looks like at WhaleBiz
7. Frequently Asked Questions

The real question behind "we need BI"

When a small business owner says "we need a BI system", they almost never mean a data warehouse and twenty dashboards. They mean one question: where do my customers come from, and what does each of them cost me?

A large company answers that with an analyst, a data warehouse and a tool like Power BI or Tableau. In a small business the answer is simpler: three sources of information that need to meet on one screen.

• Advertising: how much you spent on Google and Meta, on which campaign and which ad.
• Conversations: who got in touch on WhatsApp, the website chat, Instagram messages or a form, and what they asked for.
• The CRM: what happened with each enquiry next. Was it answered? Was an appointment booked? Did they pay?

When those three are connected, you have BI. When they sit in three different tabs, you have three reports that each count differently.

An honest note before you start: a dashboard without clean sources is decoration

Here is what happens in most businesses that try to "hook up a dashboard": the chart looks impressive, and the "lead source" column says "unknown" next to half the enquiries. A WhatsApp enquiry that came from an ad was logged as "WhatsApp". Someone who saw a post and then searched the name on Google was logged as "Google". And another third of the enquiries were never logged at all, because someone answered from their phone and forgot to copy the details.

No AI fixes that after the fact. A smart BI system on dirty data gives you confident, wrong conclusions, which is worse than not knowing. So the order is: first a clean source for every enquiry, then a screen, and only at the end AI.

The numbers an owner should see once a week

Not twenty metrics. Six, and each one answers a decision:

1. Enquiries by source. How many enquiries came from each channel, campaign and page. The decision: what to spend more on and what less.
2. Answer time. How long from the enquiry to the first reply, inside and outside business hours. The decision: where you lose customers to slowness.
3. Unanswered enquiries. How many enquiries were left without a reply or without follow-up. The decision: whether you need an automatic response or another person.
4. Enquiry to booking, booking to payment. How many enquiries became a meeting, and how many meetings became money. The decision: whether the problem is in marketing or in sales.
5. Cost per enquiry and cost per booking, by channel. Spend on the channel divided by the real enquiries logged in the CRM, not by the "conversions" the platform reports. The decision: where the money works.
6. Why enquiries didn't close. "Too expensive", "too far", "was looking for something else". The decision: what to change in the offer or in the answers.

A worked example

A business spent 3,000 ₪ on Meta and 2,000 ₪ on Google in a month. The CRM logged 40 enquiries from Meta and 10 from Google. Cost per enquiry: 75 ₪ on Meta, 200 ₪ on Google. Meta looks like the clear winner.

But of Meta's 40 enquiries, 8 booked an appointment, and of Google's 10, 6 booked. Cost per booking: 375 ₪ on Meta, about 333 ₪ on Google. Suddenly the picture flips: the cheap enquiry is not necessarily the cheap customer. You can only see this when the ad, the conversation and the booking sit in the same row.

How to get there: seven steps

Step 1: agree on shared statuses

Before any tool, decide on statuses everyone uses, for example: new, in progress, booked, paid, not relevant. Without shared statuses you cannot count "how many became a booking".

Step 2: every enquiry goes into one table

Forms, WhatsApp, website chat, Instagram and Facebook messages, and even phone calls logged by hand. Everything goes into the same table, one row per person: someone who gets in touch a second time from the same phone number is added to the same card, not opened again. That is the foundation of proper lead management.

Step 3: a source on every link

• UTM on every link in ads, bios, newsletters and posts. Agree on one convention (lowercase, fixed names for networks) and don't stray from it.
• A different pre-filled message on every WhatsApp button, so the first message already says where the customer came from.
• Ads that open WhatsApp directly pass the ad details along with the conversation. Save them on the customer's card.

Step 4: connect the spend

Connect Google Ads and Meta so daily spend lands on the same screen. Without that there is no cost per enquiry, only an estimate.

Step 5: mark how certain each attribution is

Not every attribution is certain. An enquiry with an ad code is certain, "the customer said they saw us on Instagram" is likely, and an enquiry with no trace is unknown. When you mark the confidence level, you don't make decisions worth thousands of shekels on a guess.

Step 6: a twenty-minute weekly review

Once a week, go through the six numbers and write three lines: what to boost, what to fix, what to stop. That decision log is worth more than any dashboard, because in three months you can check whether the decisions worked.

Step 7: only now add AI

When the data is clean, AI genuinely helps:

• An automatic summary of every conversation that fills in the requested service, the next step and the reason if the enquiry didn't close. The "why we didn't close" column fills itself without anyone typing.
• Anomaly alerts, for example a sharp drop in Google enquiries this week against the average.
• Questions in plain language, like "which campaign brought the customers who paid this month?", instead of building a report.
• A weekly summary in words that explains what changed and what to look at.

And honestly: advanced predictions (who is likely to leave, how much you will sell next quarter) need a lot of data. A business with a few dozen enquiries a month will get mostly noise from them. The value of AI in a small business is in summarising and sorting, not in a crystal ball.

When you really need a "big" BI tool

Power BI, Tableau or Looker Studio make sense when you have several branches, an online shop with thousands of orders, several systems that don't talk to each other, and someone on the team who maintains the reports. For most small businesses, one screen that joins ads, conversations and the CRM gives ninety percent of the value for a tenth of the effort.

What this looks like at WhaleBiz

At WhaleBiz the three sources already live in one system. The AI agents answer on WhatsApp, the website and social networks, and every conversation is logged in the platform's CRM with its source and a summary the AI writes. Marketing analytics brings Google, Meta, the website and social networks onto the same screen, and shows cost per enquiry and cost per booking based on what happened in the CRM. If you'd like to check what you already measure and what gets lost, contact our data experts.

Frequently Asked Questions

What's the difference between a regular BI system and an AI-based BI system?

A regular BI system shows data in charts and reports and tells you what happened. An AI-based BI system adds a summary of every conversation, sorting of reasons, anomaly alerts and the option to ask questions in plain language. In both cases the result is only as good as the data: if most enquiries have no clear source, the AI will be wrong too.

Which numbers should a small business owner see every week?

Six: enquiries by source, answer time, unanswered enquiries, how many enquiries became a booking and how many bookings became a payment, cost per enquiry and cost per booking by channel, and the reasons enquiries didn't close. Every number should lead to a decision, otherwise it doesn't belong in the weekly report.

Why does the dashboard show that most leads came from an "unknown" source?

Because the source wasn't recorded at the moment of the enquiry. The fix is UTM on every link, a different pre-filled message on every WhatsApp button, saving the ad details on enquiries that come from WhatsApp ads, and one table that every enquiry enters automatically rather than by manual copying.

Does a small business need Power BI or Tableau?

In most cases no, because tools like that suit a business with several branches, thousands of orders and someone who maintains the reports, while for a small business one screen that joins ads, conversations and the CRM gives most of the value for far less effort.