Eight Jobs, One Hire

Hiring briefs reach me in threes at the moment. A bank wants a data platform lead. A manufacturer wants someone to own the analytics stack. An insurer wants a head of engineering for machine learning. Read side by side they are the same document: one headcount, one salary band, and a paragraph of scope no single person has held in years.

I stopped arguing with those paragraphs a while ago and started collecting them instead. The oldest one I have is not a job advertisement at all. It is a role card on a slide, inside a proposition brief written by an outside advisory firm and filed with a large European cooperative banking group's material, from a point when the argument was still whether an internal analytics function was worth standing up.

Say what the document is first. It is a sell-side brief. It offers a staged path: prove value on one project, then build the permanent capability, then fund the adoption work as its own stage. It sets its own commercial bar at a benefits case worth five times the investment and offers to run a single insight project from hypothesis to measured benefit inside eight weeks. Both of those are promises about work the firm hoped to win. No engagement outcome is claimed anywhere in it, and I am not claiming one on its behalf. One page of it also carries a later edit in a different vocabulary, so I read that page with care and have built nothing here on it.

What I keep going back to is a page nobody would have thought was the interesting one.

Four roles, and only two of them technical

The staffing page names four people you need to make analytics work land in a bank. Two of the four are not technical at all: one exists to translate between the business and the analysis, one exists to make the answer change what people do. That balance is the argument most readers take from the page, and it is a good one. Half the team is there to make the work land rather than to produce it.

The technical half is where the trouble is. One of those two roles is the data scientist, described in the terms you would expect: knows several analytical methods, knows the software, applies algorithms to business problems.

The other technical role is described in a single line of scope. That one person is asked to be a specialist in infrastructure for large-scale data, in data governance, in architecture, in tooling, in software and in security. They are also the reason anyone can trust the data. And they are also the ones getting insight out to the business, on a schedule and in real time.

Count them. Eight.

Why one person was a reasonable answer at the time

The temptation is to read that line as overreach, a headcount stretched to make a price work. I do not think that is what happened. The line was an honest description of what those eight words meant when it was written.

Infrastructure meant a cluster somebody stood up and somebody else kept fed. Governance meant a data dictionary and a standing meeting. Architecture meant a diagram that agreed with the enterprise diagram. Tooling meant choosing the analytics package the team would buy. Software meant the scripts around the analysis. Security meant an access list and, to the brief's credit, a decision taken before the data was moved rather than after. Data quality meant looking at the source system and forming a view. Reporting meant a nightly refresh with a dashboard on the end of it.

One capable person with a strong stomach can carry all eight of those. Many did.

Now take the same eight words and price them against what a bank has to do today. Infrastructure is a platform team with an on-call rota. Governance is a stewardship function with named owners per domain and a defensible answer for a supervisor. Architecture is a standing role reconciling a core banking estate, a warehouse, a feature store and whatever the last three programmes left behind. Tooling is an internal platform with its own users and its own support burden. Software is a release process. Security is a department. Data quality is a monitoring surface with tests that run on every load. Real-time reporting is a streaming pipeline with different failure modes from every batch job around it.

The words did not get more ambitious. The work behind each one grew a team.

The audit, and the three rules that make it honest

The exercise is simple. Its difficulty is entirely in refusing to cheat at it.

Write the eight down the left of a page. Against each one, write the name of a person. Three rules.

A person, not a team. "Platform" is not an answer; a team name is where accountability goes to become nobody's. If you cannot say who would be woken up, the row is blank.

The same name may not appear more than twice. It will want to. In most mid-sized banks I see, one senior engineer's name is the true answer to most of the eight, and everyone knows it, and nobody writes it down, because seeing it written down obliges you to do something.

A person who owns the row must have owned it last quarter. A name assigned this morning because a question was asked is not ownership.

Then read the blanks. There is no scoring, no level, no target state. The blanks are the finding.

8

Concerns named inside a single technical role on one card

4

Roles in the whole proposed delivery squad

2 of 4

Roles in that proposed squad that are not technical

8 weeks

The end-to-end cadence the same brief offered for one insight project

Where the blanks usually are

I have run this in enough rooms to be unsurprised by the results.

Data quality is blank most often, and it is blank in a particular way. Everybody agrees it matters, several people run checks on the slice they personally depend on, and no one owns the question of whether a field means the same thing in two systems. That is not a tooling gap. It is a definitional gap, and it is the reason so many pipelines run cleanly and produce numbers nobody will stand behind.

Lineage is usually not on the page at all, because it was not one of the eight and nobody added it. The brief's author had no reason to name it: when there is one path from one source, lineage is obvious and free. It stops being free the moment a value can arrive by two routes, and by then no one has been made responsible for knowing which route it took.

Real-time reporting is claimed more often than it is owned. The dashboard refreshes, so somebody must own it. Ask who is accountable when it refreshes with stale figures and the room goes quiet: the answer is the person who built it, who has since moved teams.

Eight scope words on one line were not ambition. They were an accurate description of how little each of those words meant at the time. Every one has since grown a team, and the ones that did not grow a team in your organisation are the blanks.

Why this is a live question now and not a filing-cabinet one

Two things have made an old role card worth pulling out.

The first is that the modelling end has got cheap and the plumbing has not. What decides whether a model reaches production and stays there is every one of the eight: whether the feature it needs has an owner, whether the pipeline that produces it is monitored, whether anyone can say where the value came from, whether a supervisor's question about it has an answer that does not require an archaeology project. When the model was the hard part, the eight got attention because everything got attention. The model got easy and the attention went with it.

The second is that the early language-model pilots crossing my desk this year sit on top of all eight and are usually scoped as if they sit on top of none. A cautious internal pilot over policy documents or call transcripts is a data movement, an access decision, a quality question, a lineage question and a reporting question before it is a model question. Teams scope the model and inherit the other eight rows silently, blanks included.

That is what the RealAI Platform work opens with, and why we ask for the list of eight names before we ask what you want to build. It is a way of finding out which parts of the estate can carry weight and which parts are being carried by one tired person.

The old brief proposed one hire for all of it, in an era when that was nearly defensible. Nobody would write that line today. Plenty of organisations are still staffed as though somebody had.

Drawn from a proposition and capability brief written by an outside advisory firm and filed with a large European cooperative banking group's material: its staffing page, its delivery method and its commercial framing. That document offered a team, a cadence and a benefit hurdle. It records no engagement results, and nothing here should be read as one. One of its pages carries a later edit in a different vocabulary and no argument here rests on it. Reading its role card as an audit instrument is mine.