Every hiring forecast arrives with its own error rate. Each profile is backtested on months the model never saw, and the report shows the error your own data produced. See how that works
Work and workforce engineering · Worker

Demand Forecaster

Send 24 months of requisition history. Get back a monthly hiring forecast for every job profile in the file, along with the error rate that forecast earned on months it was never shown.

Your forecast, profile by profile Illustrative. Each job profile is forecast and scored on its own. Delivery consultant Steady, with a yearly rhythm 7% Data engineer Growing quarter after quarter 9% Cloud security specialist Arrives in bursts, and has a thin history 17% A weaker profile is labelled as weaker instead of being averaged away. Your own report carries the error your own file produced.
Home/ Workers/ Demand Forecaster
A forecasting worker for enterprise hiring demand

Turn the requisitions you have already raised into a hiring plan you can defend.

Demand Forecaster reads your own requisition history, works out which real job each entry belongs to, and forecasts how many people every one of those jobs will need month by month. It picks the model that fits each job separately, proves that choice against months of your history it was never shown, and hands the accuracy back to you alongside the number.

9% average error, calculated for every job profile separately.

9%
Typical error, one to three months out, on 24 months of clean history.
24
Months of history to start. More is better.
Every job
Own forecast, own error rate. Nothing hides in an average.
By horizon
Reported at 1 to 3, 4 to 6 and 7 to 12 months. It widens further out.

This is the average amount past forecasts missed by, not a band around next month. Every run reports the error your own file produced. How it is calculated

Three things a hiring forecast has to get right before anyone will act on it.

Know what the job really is

A job title is typed by a person in a hurry. The skills asked for in the description are what the job actually is, so that is what the forecast gets built on.

Fit the model to the job

A steady delivery role and a rare specialist behave nothing alike. One model for the whole company has to average the two, and gets both of them wrong.

Show the error, not just the number

A forecast with no error rate cannot be trusted or improved. Every number here arrives with the accuracy it earned on months it never saw.

Why forecasts miss

Solve the things that make a hiring forecast wrong in the first place.

Most teams forecast hiring from a spreadsheet of last year's numbers plus a percentage. These are the five reasons that keeps going wrong, and each one is handled before a model is ever trained.

The same job written many ways

One role appears as four different titles across three business units. Counted as text, a single job looks like several small ones and none of them gets a usable forecast.

Requisitions that describe a person, not a job

Plenty of requisitions say little more than that somebody left and needs replacing. Reading the description for skills recovers the actual job behind that line.

A count that hides the work

Two requisitions with the same title can ask for completely different skills. Forecast the count alone and you plan a headcount nobody is actually able to staff.

Seasonal patterns mistaken for noise

Hiring has rhythms tied to your financial year and your delivery cycles. Two full years of history is what lets a genuine pattern be told apart from a random month.

Nobody ever checks the miss

A plan produces a number and the number is never marked against what happened. Without an error rate there is no way to tell a good forecast from a lucky one.

How a forecast gets made, start to finish.

Four steps. You only have to do the first one.

1

Send one file

An export of the requisitions you have already raised. Three columns are all it needs, and the file is checked the moment it arrives.

One file. One row for each requisition. The job title The job description The date it opened You must send these three. Band, location and your sales plan are welcome but not needed. You can send a CSV or a JSON file. Up to 500 MB in one go. Send at least 24 months of history. More history is better.
2

Find the real job

The same job gets written many different ways. Each description is read for the skills it asks for, so all the versions come back together as one job.

Four ways the same job was written. Senior Engineer, Grade 2 Sr. Engineer II Engineering Specialist Cover for the person who left One real job 26 skills were read from the four descriptions. This is the step that stops one job being counted as four small ones.
3

Check the plan, then it runs

We read your dates and suggest how to split them. Say yes, or change one thing, and the rest happens on its own.

Here is the split we suggest for your file. 21 months to learn from 3 months to test on 12 months to forecast Your own history Ahead of today Show the bands in order, from the lowest to the highest. Confirm and run Or change any one line above.
4

Get your report

A full report for whoever defends the number, and one page for whoever approves it. Both are charted for you as well.

Every run gives you all three of these. The full report A spreadsheet A one page summary For the approval meeting Charted for you No file to open You are only charged once your report is ready. A failed run costs you nothing. You do not need to keep the page open while it runs.
Inside the model

The three decisions that make the difference.

A model is chosen for each job profile, not one for your whole company.

Multiple models are trained on each job profile's own history. Every one of them is scored on the same withheld months, and the one that scored best is the one kept. Nothing is tuned by hand and no model is chosen in advance, so a rare specialist role is never forced to share a model with a high volume delivery role.

Trend, seasonality and your own hiring signals all feed the monthly result, and the report names which model won for each profile.

Four profiles from one file, four different shapes, four different winners. Steady with a yearly rhythm Sharp seasonal peaks Arrives in bursts Growing quarter on quarter a different model won a different model won a different model won a different model won
By the numbers

What one run involves.

500 MB
The largest single file you can send in one go.
24 months
The least history needed before a pattern can be trusted.
3 columns
All that is required: a title, a description and a date.
Every profile
Gets its own model, its own forecast and its own score.
2 files
A full report and a one page summary, on every single run.
30 days
How long your file and its report are kept before deletion.
The report

What lands in your hands when the run finishes.

One upload produces one full report and one short summary. The platform will chart either of them for you, so you do not have to open a spreadsheet at all if you would rather not.

The monthly demand curve

How many people each job profile needs, month by month, for as far ahead as your history supports.

History against forecast

Your actual hiring and the predicted line on the same chart, per profile, so the join between them is visible.

How far each model missed

Error for every model on every job profile, split by how far ahead it was forecasting. This is the tab where our 9% either holds up on your data or does not.

Whether it missed high or low

Direction as well as size. Two profiles can share an error rate while one leaves you short of people and the other leaves you paying for a bench.

The check against last year

Each model set against assuming this month looks like the same month a year ago, which is the benchmark a seasonal business should be judged on.

The seasonal pattern in your hiring

The rhythm the model found in your own year, which is often the first time a planning team sees it written down.

The skills behind the demand

The skills appearing most often across your requisitions, so the plan can be handed to recruitment as something actionable.

Built first for the people who staff sold work, and for everyone else who has to answer for the hiring number.

Professional services is where a wrong hiring number costs twice, once on the bench and once on the work you cannot staff. Each of these starts from the same forecast and uses a different part of it.

Professional services staffing

See which roles the delivery pipeline will need before the work is sold, so the bench and the unstaffable work stop happening at the same time.

Run a forecast

Capacity and capability planning

Take a defensible number into the planning cycle, with the error rate attached, instead of last year plus a percentage.

Run a forecast

Recruitment demand planning

Know which pipelines to open first and how deep they need to be, months before the requisitions actually arrive.

Run a forecast

Workforce data and HR technology

Get a forecast out of the requisition data you already hold, with no cleaning project and no new system to stand up first.

Run a forecast

Your own private storage

Files go into storage scoped to your organisation alone as soon as they pass validation, and are not shared outside it.

Deleted after 30 days

Your upload and its report are both removed automatically, so download anything you want to keep before then. If the number has to survive a budget review months later, talk to us about longer retention for the report.

Charged only when it works

Credits come off once, at the moment a report is generated. A rejected file or a failed run costs you nothing.

What comes next

A forecast tells you what is coming. It does not tell you how to cover it.

This worker produces the forecast. The rest of the OpenKnowra platform then works on those same numbers, so the plan and the forecast never drift apart. Internal mobility looks at the roles you can fill from the people you already employ, before anybody goes to market. What if analysis lets you change the plan and see what that does to the hiring you need, before any budget is committed.

Questions

Your questions, answered.

What is workforce demand forecasting?
It is working out how many people your organisation will need, in which roles, in each month ahead. Done properly it is built from your own hiring history rather than from last year's headcount plus a percentage, and it produces a number for every role instead of one number for the whole company.
How accurate is it?
On files with 24 months of reasonably clean history, we typically see a volume-weighted error around 9% at a one to three month horizon. That is an average of how far past forecasts missed by, not a guarantee about any single month, and it widens further out: expect error at a twelve month horizon to be materially higher, which is why the report breaks it out by horizon. Thin and bursty profiles run higher again and are labelled as such rather than averaged into the headline. Every run reports the error your own file produced, profile by profile, alongside the direction of the miss and a comparison against last year, so you can judge it rather than take our figure on trust.
What exactly does the error rate measure?
It is the average absolute percentage by which past forecasts missed, weighted by hiring volume so high-volume profiles carry proportionate weight rather than being averaged in alongside a profile that hires twice a year. Months in which a profile hired nobody are held out of the percentage, because a percentage of zero is undefined, and reported separately as an absolute count instead. Alongside the percentage you get the direction of the miss and the score against a same-month-last-year benchmark. We report all three because a single accuracy figure can hide a model that is consistently short of people.
How much history do we need?
24 months at a minimum. Two full years is what lets a real seasonal pattern be told apart from ordinary month to month noise. More history helps, and so does anything that explains the demand behind the hiring, such as your annual plan or your sales and demand figures.
Our job titles are a mess. Does that break it?
No, and this is the problem it was built for. Titles are grouped into job profiles rather than taken at face value, and the skills are read from the job description text instead of the title. A role written four different ways across four business units is still recognised as one role, and a requisition that says little more than that somebody left still gets read for the job behind it.
What file formats can we send?
A CSV or a JSON file, up to 500 MB. A CSV needs one row per requisition with a header row. A JSON file can be a single object or an array of records with the same fields. Only three columns are required, which are a job title, a job description and a date. Everything else you can supply makes the forecast better.
How long does a forecast take?
It depends on the size of your file. More job profiles and more months of history simply take longer. You are shown an estimate before the run starts, though on large files that estimate is a rough guide rather than a promise. You do not need to keep the page open while it runs.
What happens to our data?
Your file goes into storage scoped to your organisation alone as soon as it passes validation, and it is not shared outside your organisation. Both the file and its report are deleted automatically 30 days after upload.
What does it cost?
A flat number of credits per report, set for your workspace. The figure in effect for your organisation is shown to you before you run, rather than quoted here where it could go out of date. Credits are only ever deducted when a report is actually generated, so a rejected file or a failed run costs you nothing.
How is this different from headcount planning in our HR system?
A headcount plan records the number somebody decided on. This produces the number, from your own history, and then tells you how far it is likely to be off. The two work together: this is the evidence you take into the planning cycle, not a replacement for the system you approve the plan in.

Run your first forecast.

Send 24 months of requisitions and get a monthly forecast for every job profile in the file, with its error rate attached.

Or bring us your data first.

Not sure your requisition history is in good enough shape? Tell us what you have and we will tell you what a forecast on it would be worth.

Request Worker Access