The Margin Everyone Crops Out

Every capability map I have drawn ends up in a photograph. Somebody points a phone at the wall, or screenshots the slide, and the picture travels into a board pack, a chat thread and eventually a procurement paper. The crop is always the same. The coloured middle survives. The small type out at the edge does not.

In the enabling-technology chapter of a multi-country platform blueprint for a global staffing and HR services group, that margin was printed five times. One map of what the business does, drawn once before anybody took a vendor meeting, then reprinted once per candidate platform component with that candidate's coverage marked on it. The coverage changed from print to print, which was the entire point of the exercise. The margin did not change at all. Same words, same size, same place, five times over.

It took me a while to see that the margin was the most honest object on the page, and that the reason it gets cropped is the reason it matters. Nothing in the shortlist could light it up.

Two kinds of thing on one page

The coloured part of the page describes the business. The staffing value chain fills one side of it, both halves at once, client and candidate, from first contact through sourcing, screening, matching and onboarding into career management, service and billing. Beside it sit the platform capabilities the business needs done, written in the buyer's words before anyone knew who might do them.

The margin describes the company. Ruled faintly across the whole surface, and named only out at the right-hand edge where a wide page runs out, is a short set of things that would have to be true about the organisation itself for any of the coloured part to function.

Those are two different kinds of statement, and the design of the exercise depends on the difference. A candidate can be laid over the capabilities, because the capabilities name work that software might plausibly do. Nothing can be laid over the conditions, because they name states inside the buyer, and no component supplies those. So they sit unmarked while the middle of the page changes colour, and unmarked type at the edge of a picture reads as furniture.

It is not furniture. It is the invoice for the coloured part, itemised in advance, and put there deliberately rather than left to a later slide nobody reaches.

What the weakness boxes wrote back in

Each candidate got the same four boxes: what the component is, what it is strong at, what it is weak at, and examples of who sells it. Because the boxes were identical, the weaknesses could not be dressed as marketing hedges. They came out as preconditions, and that is where the margin walks back onto the page uninvited.

The cloud data platform is recorded with exactly one weakness, and it spends it on a condition. Not throughput, not cost, which are among its strengths. The weakness is that separating local data to satisfy country-specific legislation and data protection law is often a problem. That is a statement about which jurisdiction's rules govern which rows, and no purchase settles it.

The content and experience platform is strong at what it claims: personalised, context-aware content, and the profiling of unknown visitors into known ones. It carries three weaknesses. One of them is product-shaped, about the difficulty of integrating to the CV and job databases the personalisation would have to read. The other two are not. Local sites must be merged onto one platform before the full benefit arrives, and the local development resource to do that may not exist. The first is a question about who holds authority over local marketing. The second is a hiring plan.

The workflow and integration component earns its place by letting locally tailored systems survive, which is a real answer to a real political problem. Both of its recorded weaknesses are the price of that answer. It forgoes the benefits of a consolidated platform, and it leaves many integrations to maintain, which is a decision to employ integration maintainers indefinitely. That second cost never appears in a licence line.

The bought-in sector platform brings hard-to-build functionality quickly and cheaply, and the chapter is generous about it. Two of its four weaknesses are ordinary product limits: the features are not market-unique, and integrating to ERP and CRM is a larger task than it looks. The other two are not product limits at all. Standard features and workflows can be difficult to fit to an existing business model and its governance, and one size does not always fit all. Read plainly: the buyer will be changing its own rules to match a supplier's defaults, and nobody has said who signs that off.

Four candidates, and in each of them at least one weakness that no signature settles. The fifth, the client-relationship platform, is the control case. All three of its weaknesses are product properties, about content depth in a specialised industry and about which messaging scenarios the tooling does and does not cover. Its defining limitation was that it manages the client side of the business and not the candidate side, and that is a coverage finding, which is exactly what the coloured part of the map is for.

The contrast is what convinced me, and it is narrower than I first read it. Every box that had a product complaint to make did make it, including the four. What separates them is that four could not stop there, and the thing they reached for once the product ran out was never anything a supplier could fix.

The only way a condition gets satisfied

A condition is worth nothing while it stays a picture. What made this blueprint unusual is that part of the margin was converted into something a finance director can act on.

The skills inventory in the same chapter lists the roles the organisation would have to grow or acquire and sorts them by discipline. Two things in it are worth stealing. Data stewardship, data compliance and data architecture appear as three distinct jobs held by three distinct people, not as responsibilities spread across a team that already has a day job. And the scrum masters, agile coaches and product owners are filed with the directors and strategists rather than with the architects and engineers, which says that ways of working were treated as something leadership owns rather than a habit a delivery team picks up.

Both are conversions. A condition in the margin becomes a line in a headcount plan, or it does not get satisfied. Every governance document I have read that arrived without a name and a salary beside it turned out, a year later, to have been a document.

5

Candidate components assessed on one map

5

Prints of the same map, one per candidate

0

Coverage marks anywhere in the conditions margin

4 of 5

Weakness boxes that name a condition no purchase settles, on our reading

Run the same page for the shortlist on your desk now

The list a data leader is looking at this quarter has the same shape. A cloud data platform. A vector store and a retrieval layer over the company's own documents. A copilot for one desk. A deployment and monitoring layer sold as MLOps. A packaged point solution, bought because building it looks slow. Five candidates, a demo each, and every deck opens with a picture of your business drawn by the person selling to you.

Draw the picture yourself and rule the conditions across it, and what comes back is not new. Every one the staffing chapter produced falls into one of three kinds, and this is our reading of those five boxes rather than anything the page itself labelled. Under whose rules does the thing run, and who says so when its output is wrong. Who is on the payroll to keep it right, and what evaluation set do they run to know. Who is under contract for the parts you did not buy, and what happens at renewal.

Not one of those is answered in a demo, because none of the vendors will be the party doing it. And they are getting sharper rather than softer. A retrieval layer over internal documents inherits every access rule the source systems had, or quietly launders them. A copilot for a recruiter is a decision-support tool sitting next to candidate data, and the European act on artificial intelligence has now reached political agreement with employment screening on the high-risk side of the line it draws, which has been the direction of travel since the first draft. The early autonomous experiments crossing my desk are all carefully scoped, and the scoping is the interesting part: someone had to decide what the system may do unsupervised, and that decision has an author or it does not exist. RealAI's Platform team writes that author's name into the plan before the pilot starts, because it is the line no supplier will volunteer.

The coverage gets coloured in by whoever you buy from. The margin gets written by whoever you hire, and nobody in the shortlist was selling that.

What to do with this on Monday

Draw your business on one page. Keep the picture stable across every vendor conversation. Then, before the first meeting, rule the conditions across it in your own words: the states inside your own organisation that the winner will depend on and cannot supply. Put a name against each one, and refuse to let the crop happen when the picture goes into the board pack, because the crop is what turns a purchase with preconditions into a purchase.

You will not get a winner out of this, and the staffing chapter did not get one either. It recommended assembling a combination, with the gaps and the overlaps drawn rather than argued. What it also produced, in the small type at the edge that nobody photographs, is the more useful document: the list of things you have to do yourself no matter which logo wins.

Drawn from the enabling-technology chapter of a multi-country platform blueprint for a global staffing and HR services group: one business capability map, five candidate components, each written up on the same four-box template and laid over the same surface, with a standing set of organisational conditions ruled across every version of it and named in its right-hand margin. That chapter produced a recommendation and a set of preconditions ahead of a procurement decision, not delivered results. The reading of the margin, and of the weakness boxes as its evidence, is ours.