Forecasting
Forecasts given as ranges, with the assumptions written down.
A forecast is a disciplined guess about next season, built from your own history. Done properly it tells you when demand will rise, roughly how far, and how sure anyone can be. Done badly it is one confident number that was never going to be right.
What can predictive analytics actually tell a local business?
It can tell you when demand is likely to rise and fall, and roughly how much search traffic and how many enquiries to expect, as a range with its assumptions written beside it. That is enough to time a budget, plan staffing and schedule content before a season starts.
It cannot tell you what will happen. A forecast built on a short history, a small number of enquiries or a market that has just changed is wide, and should be shown as wide. Anyone offering a single figure for leads six months out is selling certainty that does not exist.
Google's own tools carry the same limits. Predictive metrics in Google Analytics need at least 1,000 returning users who met a condition and 1,000 who did not before they work at all, Keyword Planner forecasts are estimates, and Google Trends is an index from 0 to 100, not a count of searches.
Written and maintained by the VIS Mountain team. Last reviewed . The sources behind it are listed below.
A forecast is an argument, not a fact
Predictive analytics sounds like software. For a local business it is mostly careful arithmetic on your own records: what happened in each month of the last two or three years, what changed along the way, and what that implies for the months ahead.
The output is a small set of ranges. Searches for your main services, month by month. Visits the site is likely to receive from search. Enquiries, split into calls, forms and bookings. Each comes as a low, a likely and a high figure, with a list of the assumptions that would have to hold.
Read the full breakdown: A forecast is an argument, not a factHide the full breakdown: A forecast is an argument, not a fact
The assumptions are the useful part. A forecast that puts spring enquiries somewhere between last year's figure and a fifth above it, assuming rankings hold, the advertising budget is unchanged and the second location opens on time, tells you exactly what to watch. When the number moves, you know which assumption broke.
It is also not what the phrase usually sells. No model knows which individual will buy, no dashboard removes the need for judgement, and at the volumes most local businesses have there is no machine learning worth the name.
It is not for a business with less than a year of clean records. There the honest product is a tracking setup, and the forecast comes later.
What a forecast needs from you
A forecast is only as good as the history underneath it. Most of this already exists somewhere in the business, and collecting it is the first piece of work.
- Search Console performance data, exported as far back as it goes. Google set that report at sixteen months when it launched in 2018, so anything older exists only if somebody saved it.
- Analytics history with enquiries counted the same way throughout. A change in how a form or a call was tracked is a break in the series, and it has to be marked as one.
- Call records by day, with answered and missed calls separated, and new callers separated from existing customers where your phone system allows.
- Booking or job records from your practice management or job system. This is the only place the real outcome lives, and the only figure worth forecasting towards.
- Advertising spend by month and by channel, because paid traffic moves every total downstream of it.
See the full checklist: What a forecast needs from youHide the full checklist: What a forecast needs from you
- A diary of what changed: a site relaunch, a new location, a price change, a provider joining or leaving, a competitor opening, the weeks you were closed.
- Your capacity. If the forecast exceeds the new patients or jobs you can take in a month, the question has become staffing, not marketing.
- For a healthcare practice, totals only. A forecast needs counts, not names, so the practice exports monthly figures with nothing that identifies a patient and decides for itself what leaves its systems.
The diary is the input people skip, and it explains most surprises. A dip that lines up with a fortnight of building work is not a trend.
How the forecast is built
None of this is exotic. The discipline is in doing every step, and in writing down what was assumed.
Clean the history
Remove what was never real demand, such as bot traffic or a tag that fired twice. Mark what was real but went uncounted, such as a month the phones were down. A forecast built on a counting error reproduces the error with confidence.
Separate the season from the trend
Compare each month with the same month in earlier years, not with the month before. Most local demand has a shape: cooling in the first hot week, tax work in the first quarter, elective dental work before benefits reset. Two full years is the practical minimum for seeing that shape, because one year cannot tell a season from a one-off.
Check the shape against outside signals
Google Trends shows how interest in a topic moves through the year, and Keyword Planner gives rounded monthly search volumes and its own traffic estimates. Both are used to confirm the shape of your year, never to set the size of it.
- Trends for timing: when the rise starts and when it peaks
- Keyword Planner for rough scale
- Your own records for everything that carries a number
See the remaining steps: How the forecast is builtHide the remaining steps: How the forecast is built
Build from the path, not the top line
Searches become impressions, impressions become visits, visits become enquiries and enquiries become booked work. Each step has its own rate in your history and its own spread. Forecasting each one separately shows where the uncertainty actually sits, which is usually the last two.
State the range and the assumptions
Every figure is given as low, likely and high. Beside it goes the list of what was assumed: budget, rankings, prices, opening hours, capacity, and anything in the diary that is about to change.
Compare with what happened, every month
Each month the actual figure is set against the range, and the forecast is revised. Old forecasts are kept, not overwritten, so there is a record of how wrong they were and in which direction.
The sixth step is the one that makes the rest worth paying for. A forecast nobody checks against what happened is decoration.
What Google Analytics needs before it predicts anything
Google Analytics 4 has predictive metrics of its own: purchase probability, churn probability and predicted revenue. Google publishes the conditions a property has to meet before they switch on.
SourcePredictive metrics, Google Analytics Help, as published September 2026
Google adds that model quality has to be sustained over time, and that the purchase and revenue predictions depend on purchase events. Those metrics are built around purchases. A clinic or a contractor counting calls and forms sends no purchase events, and few local sites see that many returning users in a week, which is why the forecasting here is done from your own history.
What a forecast is: a range that widens
A forecast earns its keep when it changes a decision before the season arrives, and it can only do that honestly if it is drawn as a range.
Budget timing is the first use. If enquiries for a service start climbing six weeks before the peak, advertising that begins at the peak has missed the people who were deciding. The forecast says when to start.
Staffing is the second. A practice that knows its busy months can plan cover and who answers the phone. More marketing into a month that is already full produces missed calls, not revenue.
The content calendar is the third. A page takes time to be found, so the article about winter boiler failures is written in late summer. A forecast turns that from a good intention into a date.
Where forecasting fails
Every one of these widens the range. Several of them together mean the honest answer is that nobody knows.
- Small numbers. A business taking thirty enquiries a month will see swings of five or six either way from chance alone, before anything has changed. No method sees through that, and a forecast has to be at least that wide.
- A short history. With one year of data there is no way to tell a season from an accident.
- A market that just changed. A competitor opening nearby, an insurer leaving a network, a new regulation, a hard winter. History has nothing to say about an event it does not contain.
- A broken series. A site relaunch, new tracking or a new phone system changes how things are counted, and the months either side cannot be compared without adjustment.
- Rankings. Nobody outside Google can forecast a change to how results are ordered, so every traffic forecast assumes positions roughly hold.
See the full checklist: Where forecasting failsHide the full checklist: Where forecasting fails
- Low volume terms in Google Trends. Google says terms with little search interest appear as zero, and that the random noise it adds for privacy shows most on exactly those terms. Many local searches are in that group.
- Platform forecasts read as plans. Keyword Planner estimates what a bid and a budget might buy. It knows nothing about how many of those clicks your front desk converts.
- The shift of clicks to AI answers, which has changed the link between how often people search and how often they visit. The figures are in the next section.
None of this argues against forecasting. It argues against a forecast delivered as a single line on a chart.
The assumption that stopped holding
A traffic forecast assumes a steady relationship between searches and clicks. Pew Research Center measured what happens to that relationship when Google shows an AI summary, using the browsing data of 900 United States adults in March 2025.
SourcePew Research Center, 2025, 900 US adults sharing browsing data across 68,879 queries
One month of data from one panel is not a law of nature. It is enough to change the method: impressions and clicks are forecast separately, and the figure a business plans around is enquiries from its own records.
What the tools say about themselves, and the standard we work to
The platforms are careful about their own numbers, and it is worth being as careful. Google describes Keyword Planner forecasts as estimates from historical search data that take bid, budget, seasonality and past ad quality into account, and says its search volumes are rounded and averaged over twelve months by default.
Google Trends describes itself as a sample of searches, normalised against all searches in that place and time and then scaled from 0 to 100, so two regions with the same score do not have the same number of searches. Google's own advice is to treat it as one data point among others.
Read the full breakdown: What the tools say about themselves, and the standard we work toHide the full breakdown: What the tools say about themselves, and the standard we work to
There is no rulebook for a marketing forecast. The nearest public standard for predictive systems is the AI Risk Management Framework from the National Institute of Standards and Technology, which is voluntary and exists to help organisations manage the risks of AI products. If a vendor sells you forecasting described as AI, ask what data it was built on, how its accuracy is measured, and who answers for it when it is wrong.
The standard we hold ourselves to is simpler. Every figure is a range, every range has its assumptions beside it, and nothing in a forecast is a promise of results or is presented as one.
We do not publish a price for this piece of work on its own, because the right scope depends on what already exists. What is published is the bundle pricing: 2,400, 3,600 or 4,800 dollars a month depending on which channels are running. You can read the full breakdown on the pricing page, and you will get an exact number in writing before anything starts.
How this connects to the rest
Forecasting depends on conversion tracking. If calls, forms and bookings are not tied to where they came from, there is no history to forecast from, only traffic.
It leans on call tracking for the same reason. Where the phone carries most enquiries, as it does in many local trades, a forecast built on forms alone describes the smaller part of the business.
It informs SEO content strategy, by putting dates on the content calendar, and Google Ads management, by saying when a budget should rise and when it can rest.
It usually starts from an SEO audit, which establishes whether the site can be measured at all and where the series is broken.
And it sits beside AI search visibility monitoring. One watches what assistants say about you, the other what those answers are doing to your numbers.
Bring two years of numbers and we will say what they can support.
Export what you have from Search Console, your analytics and your booking system. If the history is too short or too broken to forecast from, we will tell you so, along with what to start recording now.
Straight answers.
How accurate is a forecast?
It depends on how much history there is and how many enquiries a month you take. A long, clean record and high volumes give a narrow range. A short record and low volumes give a wide one.
The useful measure is the record. Each month we set the actual figure against the range, and over time you can see how often it landed inside and how far out it fell when it did not.
How much history do we need?
Two full years is the practical minimum for seeing a season, because each month can then be compared with itself. One year shows a pattern without telling you whether it repeats.
With less than a year, the right first step is tracking and record keeping. We would sooner set that up properly than forecast from four months of data.
Can you predict how many leads we will get?
As a range with stated assumptions, yes. As a single figure, no, and never as a promise.
Anyone quoting an exact number of leads for a month that has not happened is describing a hope. Ask what range they would put around it.
Do we need AI or special software for this?
Usually not. At local business volumes, a spreadsheet, clean records and someone who understands seasonality will do as well as any product, and the reasoning stays visible.
The predictive metrics inside Google Analytics need at least 1,000 returning users on each side of a condition within a week, which most local sites are nowhere near.
Is Google Trends good enough for planning?
For the shape of the year, often yes. It shows when interest in a topic starts to rise and when it peaks, which is what budget timing needs.
For size, no. Trends is an index from 0 to 100, not a count, and Google says low volume terms appear as zero and carry random noise. It is one input beside your own records, not a replacement for them.
What happens when the forecast is wrong?
It will be, to some degree, every month. What matters is whether the actual figure fell inside the range, and if it did not, which written assumption failed.
That is why the assumptions are listed. A miss with a known cause improves the next forecast. A miss nobody can explain means the range was too narrow.
Where this comes from.
Primary documentation and published research behind the guidance on this page.
- Predictive metrics, Google Analytics Help (opens in a new tab)What the three metrics predict, and the volumes a property needs before it is eligible.
- About Keyword Planner forecasts, Google Ads Help (opens in a new tab)Forecasts are estimates from historical search data, and search volumes are rounded.
- FAQ about Google Trends data, Google Trends Help (opens in a new tab)A sample, normalised and scaled from 0 to 100, with low volume terms shown as zero.
- Google users are less likely to click when an AI summary appears, Pew Research Center (opens in a new tab)
- NIST AI Risk Management Framework (opens in a new tab)
- Introducing the new Search Console, Google Search Central, 2018 (opens in a new tab)Where Google set the performance report at sixteen months of data.
Talk to the team
A short call, a look at how the business currently shows up, and a straight answer on what we would do first.