AI big data analytics sounds like a warehouse project with a data science team attached. For a company of 10 to 250 people it is something smaller and more useful: every lead, call, booking, message and invoice landing in one record, with AI reading across those records to tell you where revenue is leaking. This article answers one question. How much of that can you run inside the levelpartner.ai platform, and how do you set it up so the answers are trustworthy?
What "big data" means at your size
A 60-person firm does not have billions of rows. It has a few thousand contacts, a few hundred opportunities a year and tens of thousands of messages, calls and page visits. The difficulty is not volume. It is that the data sits in six tools that never talk to each other, so nobody can answer a simple question such as which source produced last quarter's won work.
The platform solves the joining problem first. Forms, calendars, chat, calls, email, SMS, payments and the pipeline all write to the same contact. Once that is true, the AI features have something coherent to read, and the reports stop contradicting each other.
Where the AI actually helps
AI is good at three jobs in reporting. It summarises long histories, such as a contact with forty messages and three calls, into a few lines a manager can read before a meeting. It classifies free text, for example tagging inbound enquiries by service line or urgency from what the person wrote. And it drafts the narrative around the numbers, so a weekly report arrives with a plain-English explanation rather than a wall of widgets.
It is weak at inventing data you never captured. If your forms do not record the source of a lead, no model can recover it later.
Before you start
- Admin access to the account and permission to edit Settings, Sites and Automation.
- Your website forms rebuilt on the platform, or the platform's tracking code added to the external site.
- One pipeline per sales motion, with opportunity values filled in for open and won deals.
- A short list of the five questions the leadership team asks every month.
- Agreement on who owns data quality, by name.
How to set it up
- Write down the five leadership questions and the field each depends on.
- Go to Settings, then Custom Fields, and click Create field.
- Choose the Field type, enter a Field name, then Create custom field.
- Go to Sites, then Forms, and add the new fields to each enquiry form.
- Connect ad accounts under Settings, then Integrations.
- Go to Automation, then Workflows, and click + Create Workflow.
- Choose Start from Scratch and add the Form Submitted trigger.
- Add AI Extract Data, with the enquiry text in Extract From.
- Add an If/Else branch on the extracted service line.
- Add Add Contact Tag on each branch, then switch Draft to Publish.
- Go to Dashboard, click + Add Dashboard and add one widget per question.
- Click Summarize Dashboard with AI before each weekly meeting.
- Check twenty recent contacts by hand against the dashboard figures.
What it replaces in a company with flat revenue
Most firms that have stalled are running reporting on spreadsheets exported every Monday by someone in operations. That job takes half a day, the numbers are a week old, and the export quietly drops anything that did not match the template. A platform dashboard replaces the export, and the AI summary replaces the paragraph someone writes to explain it.
It also replaces the guesswork in pipeline reviews. When every opportunity carries a source, a value and a stage history, you can check whether referrals close faster than paid search before you decide where the next marketing budget goes. For how that source data is assigned, read how a multi-touch model splits credit.
Say the firm logs 900 enquiries a year across forms, calls and chat, and the monthly board pack takes two days to assemble without ever splitting revenue by service line. Once an AI classification step tags each enquiry, the dashboard could show something like this: one service line produces a third of enquiries but half of won value. That is the kind of pattern a manual export rarely surfaces. The figures are invented for the example.
Time it takes each week
Setup is two to four days, most of it spent deciding fields and fixing forms rather than building widgets. After that, budget thirty minutes a week for the dashboard owner to read the AI summary, spot-check five records and correct any misclassified enquiries. Without that half hour, classification errors compound and the dashboard drifts away from reality inside a quarter.
Teams build the dashboard first and the data capture second. The widgets look finished on day one and show nonsense for months. Fix the forms and fields first, wait two weeks for clean data to arrive, then build the dashboard.
When not to use it
If you already run a proper data warehouse with an analyst who maintains it, do not move that work into the platform. Use the platform as a clean source that feeds the warehouse instead. And if your sales happen mostly offline with no CRM discipline, start with pipeline hygiene before any AI layer, because analytics on an empty pipeline is just decoration. The sales and marketing dashboard guide covers the basic layout most firms need first.
Troubleshooting
The dashboard totals do not match the finance system. Opportunity values in the pipeline are estimates until a deal is marked won with its final value. Ask the sales team to update the value at the moment of closing, and compare against invoiced figures only for won deals.
Most contacts show no source. The leads are arriving through a form that is not on the platform or a page without the tracking code. Find the entry points with the most blank sources and fix those first.
The AI classification is inconsistent. The instruction is probably too vague. In AI Extract Data, give the field a Description with a fixed list of allowed categories and an example for each, and send anything it cannot place to a review tag.
The AI summary is too broad to act on. Ask a follow-up question limited to the five leadership questions and the movement against the previous week.
Frequently asked questions
Do I need a data warehouse to use AI analytics?
No. For a firm with a few thousand contacts, the platform's own records are enough to answer most commercial questions. A warehouse becomes worth it when you need to combine CRM data with finance, product or operational systems at scale.
Can the AI see data from tools outside the platform?
Only what you bring in. Ad accounts, calendars and payment providers connect directly, and other systems can push data through webhooks or imports. Anything that never reaches the contact record is invisible to the reports.
How accurate are AI summaries of contact history?
They are reliable for summarising what happened and weak at judging intent. Treat them as a briefing, not a verdict, and keep a person responsible for decisions that depend on them. For the reporting views that feed them, see building a marketing data dashboard.
Is our contact data used to train outside models?
Check the data processing terms for the AI features before you switch them on, and record the decision in your data protection register. Keep sensitive personal data out of free-text fields that AI steps read unless you have a clear lawful basis.