From gut feeling to AI: a better sales forecast
Erwin Berkouwer
You get a better sales forecast by letting your historical deal data predict what happens in your pipeline, and by attaching an action to it straight away. Predictive analytics learns from every won and lost deal. An AI agent makes sure the signal doesn't sit in a dashboard but leads to a follow-up.
Recognise this? Your sales team closes the quarter well under forecast. The pipeline looked solid, three deals that seemed certain went to a competitor and one big customer postponed their decision. Looking back, you see signals you could have spotted, if only you'd had them. Many forecasts are made of gut feeling, sales optimism and the hope that this quarter will beat the last.
What does a gut-feeling forecast cost you?
A poor sales forecast is more than an uncomfortable conversation with management. A forecast that's too high means too much capacity, too much money tied up in stock and budgeted bonuses that never get paid. A forecast that's too low means missed opportunities, being understaffed just as the market picks up and customers going to a competitor because you can't keep up with demand.
Making the forecast itself also takes a lot of time. Sales managers regularly put several hours a week into it, while the outcome stays unreliable. Work out what that costs you per year.
The underlying cause is usually the same: data is spread across CRM, email, spreadsheets and the heads of your salespeople. There's no consistent model, and people are bad at judging probabilities objectively, especially for deals they're involved in themselves.
What is predictive analytics for sales?
Predictive analytics uses your historical data to spot patterns and predict outcomes. The core is simple: the system learns from every won and lost deal and applies that knowledge to your current pipeline.
There is a big difference between a dashboard and an AI agent, though.
A dashboard shows you that deal X has a 23% chance of being won. Then you have to decide yourself what to do about it: direct someone, change the priority, schedule a follow-up.
An AI agent acts on that information. If a deal's win probability drops below a threshold, the agent sends a signal to the account manager, schedules a follow-up in your CRM or escalates to the sales manager. The system makes sure the right action happens at the right moment. A person still decides what goes out to customers.
Tip: don't start by building a perfect prediction model. Start with the question: what action should happen when a deal is likely to be lost? If you can define that action, you can automate it.
This distinction matters for SMBs, where you usually don't have a data scientist to monitor and interpret dashboards.
How do you use predictive analytics in practice?
SMBs use predictive analytics for sales in three ways.
Lead scoring: focus on the right opportunities
The system analyses historical data on won deals, such as company size, industry, behaviour on your website, response times to emails and the number of contacts before a decision. New leads then get a score. That way you spend less time on leads that statistically won't buy and more on leads that will.
Churn prediction: stop customers leaving before it starts
For businesses with recurring revenue, like subscriptions and maintenance contracts, predicting churn is one of the quickest ways to get value out of data. The model learns to recognise patterns that come before a cancellation: less product use, fewer contacts, lower NPS scores or late payments.
Has a customer who normally gets in touch every month been silent for three months while the invoice value stays stable? That's a signal. An AI agent picks it up and can suggest a proactive contact moment.
Revenue forecast per segment and period
Instead of one big forecast for the quarter, you get predictions per product line, per salesperson, per customer group and per region. The system can account for seasonal patterns and the individual performance history of each team member. A more accurate forecast means you need less buffer for capacity, purchasing and working capital.
Why does integration with your own systems matter so much?
Many SMBs work with a Dutch accounting package like Exact Online, AFAS or e-Boekhouden, next to a CRM like HubSpot, Salesforce or Pipedrive. Many predictive analytics tools are built for the international market and don't connect to those packages out of the box.
That's why many implementations stall: if your revenue data is in your accounting and your deal data is in your CRM, and they don't talk to each other, you don't have the full picture. Effective predictions need the main data sources to be connected:
- CRM: HubSpot, Salesforce, Pipedrive
- Accounting: Exact Online, AFAS, e-Boekhouden
- Communication: email, telephony, customer service tools
- Marketing: website analytics, campaign data
Tip: for every solution, first ask whether there's a direct API connection to your accounting package. Without it you depend on manual exports, so on outdated data and extra admin. On the integrations page you can read which systems UnifyAI works with. You'll find more on the accounting side in connecting Exact Online to AI.
How do you start with an AI sales forecast? Four steps
You don't have to do everything at once. This approach works without a data department of your own.
Step 1: map your data sources. Work out where your sales data comes from (CRM, accounting, email, website) and where data is missing or inconsistent. You don't have to perfect this before you start. If your data isn't in order yet, look at data in order.
Step 2: pick one use case. Choose the problem that hurts most: poor forecast accuracy, high customer churn or too much time on leads that don't buy. Doing one use case well gets you more than doing three halfway.
Step 3: connect your data and start the first model. The agent pulls in data, learns from your historical deals and starts scoring and flagging. This doesn't have to be a big IT project.
Step 4: measure, adjust and expand. After the first month, check whether the predictions match reality. Adjust thresholds, add variables and expand to a second use case.
Tip: in the first weeks, have a salesperson work alongside the system. Have them assess every AI recommendation and note whether it was right. That improves the model and builds buy-in in your sales team.
What do you get out of it, and what does it cost?
The benefits are rarely a single number. You spend less time on forecast admin, prioritise leads better, lose fewer customers because you step in on time and make better decisions on capacity, purchasing and marketing. What that's worth in euros depends on your team size, deal value and the quality of your data.
So measure your starting point first: how many hours go into the forecast now, how often were you off, and how many customers drop out? Then after a few months you can judge honestly whether it works. It's also smart to start small: an AI scan shows which processes suit a first agent, and with AI consultancy you make a concrete plan. Rather talk it through first? Book an intro call.
Frequently asked questions
Short, clear answers so you can decide faster.
What is predictive analytics for sales?
Predictive analytics uses your historical deals and customer data to spot patterns and predict the outcome of your current pipeline. The system learns from every won and lost deal and applies that to new leads and deals.
What's the difference between a dashboard and an AI agent for sales?
A dashboard shows a win probability, but you have to decide what to do with it. An AI agent attaches an action to that information, such as a signal to the account manager or a suggested follow-up in your CRM.
What data do I need for a better sales forecast?
Mainly historical deal data from your CRM, plus revenue data from your accounting and communication data like email. It doesn't have to be perfect to start, but inconsistent data limits the quality of the predictions.
Where do I start with AI in my sales process?
Pick one use case that hurts: poor forecast accuracy, high customer churn or too much time on poor leads. Connect your data sources, let the model run alongside your team and measure after a month whether the predictions hold up.
