Sales forecast: how to make your predictions more reliable?

What is a sales forecast?

The sales forecast seeks to answer a simple question:

how much revenue do we have good reason to believe we will close over the coming weeks or months?

It should not be confused with three other concepts.

The sales target corresponds to the revenue the company aims to achieve.

The sales pipeline represents the opportunities currently in progress and their advancement in the sales process.

The forecast is an estimate of what should actually be signed among these opportunities.

The actuals finally correspond to what has actually been won.

A company can therefore have:

• a target of €200,000;

• €450,000 in open opportunities;

• a forecast of €130,000;

• and finally sign €115,000.

These four figures tell different stories.

The problem arises when we automatically assume that the total value of the pipeline constitutes a revenue forecast.

This is almost never the case.

Why are sales forecasts often wrong?

When a business leader explains that their forecast is not reliable, the first instinct is sometimes to question the CRM or the calculation method.

The problem is often located further upstream:

• a closing date is entered because the CRM requires a date, not because a deadline has actually been discussed;

• opportunities that no one has responded to for two months remain open;

• two sales reps do not interpret the pipeline stages in the same way;

• the deal amount is still highly approximate;

• no next action is scheduled;

• a closing probability is chosen based on gut feeling;

• lost opportunities are not properly closed and archived.

Applying a more advanced formula to this data does not make the forecast any more reliable.

Above all, it produces an apparently more precise figure based on the same fragile assumptions.

This is why forecasting starts with a structured sales pipeline: active opportunities must be identifiable, stages must correspond to observable events, and deals that are no longer real must exit the active pipeline.

Once these foundations are in place, the forecasting method can gradually evolve.

The 4 levels of a reliable sales forecast

Not all companies need the same forecasting system.

An SMB that signs a few dozen contracts a year with little usable history should not build the same model as a sales team handling several thousand opportunities.

The right approach is to use the level of sophistication that the data actually justifies.

Level 1 — Only forecast what is based on something concrete

When the company has little reliable history, it is best to start simple.

The goal is not yet to assign a precise probability to each opportunity.

First, we must distinguish:

what actually exists from what could eventually happen.

Let's take three opportunities:

Showing a €37,000 forecast simply because B and C have an amount would already be questionable.

But adding an arbitrary estimate of €20,000 for Company A would make the figure even less usable.

At this stage, a few simple rules are often better than complex weighting:

• do not include opportunities where the amount is still too uncertain;

• require an observable sales event before considering a deal sufficiently advanced;

• verify that the closing date is based on a real element;

• exclude opportunities with no activity or credible next action from the forecast;

• close deals that are no longer actually active.

The resulting forecast may be less impressive.

But it will be much more useful for decision-making.

Example: when €400,000 of pipeline becomes €95,000 of forecast

During a pipeline review, we once encountered a very telling situation.

The business leader saw about €400,000 of open opportunities in his CRM and naturally used this amount to project future revenue.

The problem: virtually every opportunity was contributing to the figure.

First call, barely qualified opportunity, proposal sent, advanced negotiation: everything ended up being considered almost equivalent potential revenue.

We reviewed the pipeline with a much simpler rule:

let's only forecast what is based on something concrete enough.

Once the opportunities were re-evaluated, the amount actually usable for steering dropped to around €95,000.

The new figure was much less comfortable.

But it notably led the leader not to immediately launch a recruitment drive he was planning based on the previous pipeline.

This is precisely the role of a forecast.

Not to reassure.

To help make better decisions with the information actually available.

Level 2 — Weight opportunities using actual history

When enough opportunities have been properly tracked, it becomes possible to go further.

The classic method is to calculate a weighted pipeline:

Weighted forecast = sum of the amount of each opportunity × its probability of closing

Suppose we have three deals:

The weighted forecast would then be €36,000.

Mathematically, the calculation is simple.

The real question is:

where do the 20%, 50%, and 80% probabilities come from?

A poorly weighted pipeline can be worse than an unweighted pipeline

In many CRMs, a probability is arbitrarily assigned to each stage:

• first call: 10%;

• qualified opportunity: 25%;

• proposal sent: 50%;

• negotiation: 75%;

• verbal agreement: 90%.

The result looks like a statistical model.

But if no one knows why "proposal sent" is worth exactly 50%, the precision is mostly visual.

A more robust approach is to look at history.

If 100 comparable opportunities reached the "proposal sent" stage and 32 were ultimately won, this data already provides a more solid foundation than a probability chosen when setting up the CRM.

What does this look like in practice in Attio?

In Attio, you can keep the amount, stage, and closing date on each Deal, and then use a Formula attribute to automatically calculate its weighted value.

Technically, it is therefore quite easy to do:

Amount × probability associated with the stage

But we wouldn't start with the formula.

We would start with the data that justifies it.

Attio's Funnel reports allow you to observe conversion rates between different pipeline stages. After enough history is accumulated, these results can be used to challenge the probabilities used in the calculation.

The CRM then does the math.

The business rule remains based on what actually happens in the sales process.

And if the available history is still too scarce or unreliable, we would prefer not to use the weighted pipeline as decision data rather than display artificial precision.

Do not segment more than the data allows

Even a real historical probability can hide significant differences.

A company might, for example, convert many small standardized deals but far fewer complex projects.

It can then be useful to gradually segment rates by:

• offering;

• customer segment;

• deal size;

• acquisition channel;

• possibly by sales representative.

But creating twenty segments each containing three opportunities does not improve the forecast.

Do not break down the data more finely than the available volume allows.

Level 3 — Integrate the time dimension

A good probability of closing is not enough to produce a good forecast.

We must also answer a second question:

when does the deal actually stand a chance of closing?

Let's take a €50,000 opportunity where comparable deals historically have a 70% chance of being won.

If this type of sale typically takes four months to close, it would be dangerous to automatically include €35,000 in next month's forecast simply because the sales rep entered a closing date in thirty days.

A more mature forecast therefore begins to take into account data such as:

• the usual cycle length;

• time spent in the current stage;

• the opportunity creation date;

• the expected closing date;

• successive postponements of this date;

• the last activity;

• the next action.

A deal that was supposed to close in April, then got moved to May, June, and July, should not be viewed the same way as an opportunity progressing normally through its sales cycle.

This repeated shifting, often called slippage, becomes a signal in itself.

The sales review should highlight exceptions

The goal is not to build a dashboard with 25 metrics.

In an Attio setup, three angles are already enough to highlight many issues:

• deals expected to close during the period;

• opportunities that have stayed abnormally long in a stage;

• deals whose progression no longer matches the typical cycle.

Attio notably offers Funnel, Time in Stage, and Stage Changed reports to analyze conversions, time spent in stages, and pipeline movements.

The system does not automatically decide that a deal is lost.

It brings up what deserves the team's attention.

The sales review can then focus on exceptions and decisions to be made, rather than a line-by-line reading of every opportunity.

What data actually needs to be maintained in the CRM?

Making a forecast reliable does not require thirty new fields.

For most B2B organizations, a few properly maintained data points already provide significant value:

Amount

It should correspond as closely as possible to a commercial reality: a proposal, a reasonably scoped estimate, or expected contract value.

Pipeline Stage

It must reflect an observable event and be understood in the same way by the team.

Estimated Closing Date

It must be updated when new information appears, not simply pushed automatically to the next period.

Next Action and its Date

A deal assumed active with no identifiable next action deserves to be challenged.

Last Activity

It helps detect opportunities that appear advanced in the CRM but are actually stalled.

Won or Lost Status

Lost deals must be closed. Otherwise, they clutter the current pipeline and prevent building a usable history.

Segment or Deal Type

When cycles and conversion rates are truly different, a few simple categories can improve the forecast.

The logic is the same as for B2B lead qualification or lead scoring:

data is only worth asking for if it improves a decision or enables an action.

Adding fields without improving practices often degrades CRM adoption — and therefore, over time, the quality of the forecast.

Should the forecast be moved out of the CRM?

Not necessarily.

This is actually a good example of an architectural decision where adding an extra tool can make the system worse.

If Attio already houses the sales pipeline and contains the information needed for forecasting, we would keep as much of the following as possible in Attio:

• sales data;

• simple calculation rules;

• management views;

• sales reports

in Attio.

Sending every opportunity to Airtable, Make, or a BI tool solely to rebuild the same dashboard would mostly create a second layer to maintain.

When does Airtable become relevant?

The situation changes when the forecast needs to be crossed with data that genuinely sits outside the CRM.

For example:

Deals likely to close
→ production workload
→ team availability
→ hiring needs
→ forecasted billing

For an agency or service company, the question is then no longer just:

"How much are we likely to sign?"

but also:

"If we sign these projects, do we have the capacity to deliver them under good conditions?"

Airtable can be more suitable for this second part when it is already used to structure staffing, projects, delivery, or other operational processes.

You can then have a coherent architecture:

Attio → CRM and sales forecast

Airtable → capacity, delivery, and operational management

with synchronization of only the truly necessary data between the two.

The goal is not to build an Attio + Airtable + Make stack just because the tools can be connected.

The hybrid architecture only makes sense when the business need truly goes beyond the CRM.

This is also the type of trade-off we aim to make when we help a company structure its CRM and RevOps system: starting from the process to be managed, then choosing the simplest architecture that actually allows its execution.

How to measure the reliability of your forecast?

A forecast cannot be improved if no one compares what was predicted with what actually happened.

A simple practice is therefore to periodically log:

• the announced forecast;

• the period concerned;

• the revenue finally closed;

• the variance between the two.

Let's imagine:

Announced forecast for the quarter: €150,000

Actual revenue signed: €120,000

Variance: €30,000

The most interesting part is not just seeing that the forecast was too optimistic.

We need to understand why:

• too many deals slipped to the next quarter;

• the probabilities used were too high;

• closing dates were systematically optimistic;

• one segment had a lower conversion rate than assumed;

• large opportunities were lost;

• inactive deals continued to be counted.

Over successive periods, the team can identify recurring biases and adjust its rules.

This provides much more value than an abstract target like "our forecast must be 90% reliable."

The level of precision needed depends primarily on what the company decides based on this information.

A €20,000 error does not have the same consequences for a team using the forecast to run a sales meeting as it does for a business leader using it to recruit or commit to new expenses.

From that point on, the forecast stops being just a Sales indicator.

It becomes business intelligence for driving the company.

This is also one of the principles of RevOps: gradually making processes and data reliable enough to improve decisions, rather than simply adding more reporting.

Level 4 — What role for AI in sales forecasting?

AI can add an extra layer when the CRM has enough data and context.

It can notably help to:

• analyze the characteristics of won and lost deals;

• spot certain high-risk opportunities;

• leverage the content of notes, emails, or calls;

• identify signals that are hard to represent in a formula;

• more easily query sales data.

For example, today Attio allows you to query CRM data in natural language using Ask Attio, including pipeline-related metrics.

But this does not mean everything must become probabilistic.

A value like:

€20,000 × 32% = €6,400

does not need an AI agent.

It needs a deterministic and repeatable calculation.

We would therefore separate the two use cases:

rules and formulas for what must be calculated exactly;

AI to interpret more context or help the team identify what deserves their attention.

Putting AI in a forecast is not about replacing a multiplication with a prompt.

And if lost deals remain open, dates are artificial, or amounts are inconsistent, the AI will also be working on an imperfect representation of reality.

AI is therefore more of the fourth level of a mature forecasting system rather than its starting point.

Which forecast level to choose?

The progression can be summarized simply:

Level 4 is not automatically better than Level 1.

It is better only when the data justifies it.

An SMB with few opportunities can make excellent decisions with a deliberately simple forecast.

An organization with more data can gradually enrich its system as its history becomes usable.

It is ultimately the same logic as for the CRM architecture itself:

sophistication should arrive when it solves a real problem, not simply because it is technically possible.

A reliable forecast is therefore not primarily a formula, a dashboard, or an AI feature.

It is the result of a sales system that is simple enough to be used and rigorous enough for its data to begin earning our trust.

Frequently asked questions about sales forecasting