HOW TO START AN HR ANALYTICS FUNCTION WITH ONE PERSON AND A SPREADSHEET

Most people analytics functions start with one person, part of their week, and a spreadsheet. Here is what the first year looks like when it works.

Emmanuel Duncan · 7 September 2026 · 6 min read

The usual account of building people analytics starts with a platform, a data warehouse and a small team of specialists. For most organisations in this region that account describes a destination rather than a starting point, and treating it as a starting point is why so many attempts stall at the business case.

What follows is the version I have seen work: one person, part of their week, a spreadsheet, and about a year. It is slower than the brochure. It also survives, because it produces something useful in the first quarter and keeps producing it.

Month One: The Four Questions

Do not begin with data. Begin by asking four or five people in the leadership team what they would do differently if they knew something they currently do not.

The answers you want are decisions rather than topics. "Whether to keep recruiting supervisors externally or promote from within." "Which of our three depots to fix first." "Whether the increase in casual staff is temporary." Each of those is answerable. Each has a person waiting for the answer.

Write down four. Publish nothing else in year one. The temptation to produce a broad monthly report arrives early and it consumes the entire capacity of a one person function, because a broad report is never finished and nobody reads it closely enough to notice.

Months One and Two: One Dataset, Built Properly

Almost every question in HR needs the same underlying table, and building it once is the highest return work in the whole first year.

One row per person per month. Columns: person identifier, month, unit, job grade, location, employment type, start date, exit date, exit type where applicable, manager identifier, and a flag for whether the person was active in that month.

That table answers headcount, turnover, tenure profile, internal movement, span of control and a large share of what will be asked in the first two years. It is built from records you already hold, mostly from payroll and the establishment list, and it usually takes a few weeks of unglamorous work to assemble because the two sources will disagree.

Build it in a spreadsheet if that is what you have. Twelve months of monthly snapshots for a few hundred staff is a few thousand rows, which is nothing.

Two disciplines make it last. Never overwrite a monthly snapshot once it is stored, so that history stays fixed. And keep the transformation steps in a written note or a script, so the next month is repeatable and the person after you can reproduce it.

Month Three: Settle the Definitions in Public

Before anything is reported, write the definitions page and get it agreed by HR, payroll and finance.

Headcount, active employee, full time equivalent, leaver, regretted leaver, internal move, vacancy, and the reporting date. Eight terms, one paragraph each, one page in total.

This is the step people skip and the one that determines whether the function is believed. A figure produced without agreed definitions can be argued with indefinitely, and the argument will be about the definition every single time, so the analysis never gets discussed.

Circulate it as a working document, take the corrections, then fix it for the financial year.

Months Four to Six: Answer One Question Completely

Take the first of the four questions and answer it properly. One question, done to a standard you would defend, is worth more than nine partial answers.

Completely means the analysis, the caveats, the limits of the data, and a recommendation with a named owner. It also means going back at three months to see whether anything changed, which is the part that turns an interesting analysis into a function people ask for.

Present it in person, to the person who asked. A memo circulated to a distribution list will be read by nobody, and the value of the first piece of work is largely in the conversation it starts.

What to Say No To

A one person function fails by accretion. Requests arrive individually, each one is small, and by month eight the whole week is spent on ad hoc queries.

Three things I would refuse in year one.

Dashboards. A dashboard is a maintenance commitment, and building one before the definitions are settled means rebuilding it. The four questions do not need one.

Predictive work. Flight risk models and similar are the request that arrives earliest and delivers latest. They need years of clean history and they produce output that is hard to defend. Note the request and come back to it in year three.

Any figure whose source you cannot explain in one sentence. Numbers you cannot account for will damage the function faster than slow delivery.

An acceptable no sounds like a date. "That is a question for the second half of next year, once the monthly file has twenty four months in it."

Months Seven to Twelve: Make It Repeatable and Visible

By the second half of the year you should have the monthly file running, the definitions agreed, and two of the four questions answered.

Add three things. A short standing set of figures, no more than one page, produced on the same date each month from the file. A written record of every question answered and what happened as a result, which becomes the case for resourcing the function. And one other person who can produce the monthly file if you are away.

That last point matters more than it seems. A function that depends entirely on one person is one resignation away from having never existed, and the leadership team knows it.

What Year Two Looks Like

If the first year went as described, year two starts from a position most analytics functions never reach: agreed definitions, twenty four months of consistent history, and a leadership team that has seen the work change a decision.

That is the point at which a tool, a dashboard or a second person becomes worth asking for, and the request is easy to make because it is supported by a record rather than by a promise.

Where to Start This Month

Book four conversations. Ask each person the same question about what they would do differently if they knew something they do not.

Then open a spreadsheet and build one month of the person by month table. One month is enough to find out which records disagree with each other, and that discovery is the real beginning of the work.

Emmanuel Duncan is the founder of Aletheia Lunara Consult, a professional education and consulting practice working with organisations across Ghana and beyond on people analytics, workforce research and organisational capability.