Every large organisation that decides to get serious about data begins by collecting ideas, and the collecting goes well. Ask across the operating units and the suggestions arrive in volume, most written by the people closest to the work and therefore worth reading. The list is not the problem. What happens to the list is the problem. It goes to a steering group, then to an architecture review, then into a budget cycle, and somewhere in there the sponsors move roles and the numbers behind each idea go stale.
A global staffing and HR services group set out to compress that interval deliberately. The method slide states the promise in plain commercial language: from the moment the wish list is opened to the moment a scored, sequenced playbook exists, as little as six to eight weeks. That is a claim about a way of working, and it deserves separating from what was delivered. What the engagement put on the board was one case that ran from a standing start to software in the hands of sales people by week six, plus a repeatable way of choosing the next one.
The compression is the whole argument. A long list scored slowly is a long list, and the scoring only means anything if it lands while the enthusiasm behind each idea is still pointing in the same direction.
The challenge
The group's own diagnosis was not that appetite was missing. Appetite was there. The difficulty was that the operational challenges and risks sat across every part of the organisation at once, and had to be managed while a digital strategy was being executed rather than after it. Written that way, the ask is uncomfortable: change what technology is for, from something that supports the business to something that moves the numbers, without pausing the business while you do it.
Two tracks ran in parallel. The top-down track was a multi-country platform programme, measured in horizons and correct to run slowly. The bottom-up track had a narrower job: prove that data could pay for itself somewhere visible, and build the group's own capability while doing it. A changed customer experience is the part everyone sees, but it starts inside, with the processes and the business model that produce it.
Which leaves the scoring problem. Two axes decide whether an idea gets built: the value of the opportunity and the ease of implementing it. Both are cheap to argue about in a room and expensive to establish properly. Establish both properly for an unfiltered list and nobody finishes; skip the work and the loudest sponsor wins. The accelerator exists to sit between those two, fast enough that the answer still applies when it arrives.
The approach
Three stages, run in order. The first cut the open list down to the ideas worth an hour of anyone's time. The second took the survivors far enough to know whether the data behind each one existed in a usable state. The third turned what was left into a small number of high-impact projects with an owner and a sequence. The narrowing is aggressive on purpose, and the cases are incubated inside the company rather than delivered to it, so the capability stays behind when the engagement ends.
Speed came from not starting at zero. The accelerator arrived with a toolbox of pre-built pieces: worked concepts, algorithms that had run before, and an open-source data platform ready to take a feed. Six weeks is only achievable if none of those six weeks goes on procuring a cluster or agreeing a data-sharing form. That is the half most programmes underestimate, and it is why our Platform work begins with the environment already standing rather than with a procurement.
The proof case came from sales. The commercial teams were unhappy with their hit rate, and the reason was specific rather than cultural: they had no way to predict where placement demand would appear next, so prospects were worked in whatever order the list happened to be in. Week zero began with hypotheses rather than a model. The sales teams named the business rules they believed separated a good prospect from a poor one, those rules became the variables, and weighting them produced a five-star rating that ordered the call list by score rather than by alphabet or by habit.
Three signals carried the model, all of them specific to placement rather than borrowed from generic firmographics: how large the company is, how quickly its vacancies get filled, and how many different types of vacancy it has open. Fill speed and vacancy spread are proxies for how much demand a company generates and how hard it finds serving itself. By week six the tool was in use with twenty sales people, and outcomes from their calls were being captured to keep tuning the weights.
The source reports the result as a direction, not a delta. The highest-scoring companies arrived at the top of the calling list, the hit rate went up, and a measured increase in satisfaction was recorded among the sales people using it. No baseline and no percentage are published, so a direction is all anyone should claim.
Behind the one case sat four conditions the group named for itself. Pool the platform investment and the scarce people rather than letting each unit buy its own. Lead from measured fact rather than from the loudest account. Get to a single view of the customer and the candidate across the internal silos, because a scoring model is only as good as the joins underneath it. And put the work with people who behave like owners, funded without a full budget round for every experiment.
The outcome
What this track delivered is one deployed tool with directional results and a method for choosing the next one. It is not a group result and the source does not claim one. The scaling instrument was a short capability review across five areas, covering data, technology, the organisation, its people and the people who use the output, run so that operating companies could be compared with each other and with outside reference points. The comparison was the point of it. An operating company will adopt what a peer company built and beat far more readily than what a central function mandates. The review closed in a workshop rather than a report, and produced a roadmap with both a value line and a capability line.
Three things would be done differently now.
First, the hypotheses came out of workshops. Sales people are good witnesses to their own process, but they describe the process they believe they run. Process mining over the event logs in the applicant tracking and CRM systems gives the same hypotheses with evidence attached, and it gives the counter-examples too: the prospects that converted despite scoring badly, the most useful training data in the set.
Second, deploying a scoring model to twenty people is the easy half. Keeping it correct once the labour market moves is the half nobody schedules. A weighted-rules model built from last quarter's demand pattern degrades quietly, and the defence is ordinary deployment discipline: features written down and versioned, decisions kept on a lineage trail, and a scheduled comparison of predicted against actual that somebody owns by name. Our Consult work opens on that question now, because it decides whether week six survives week twenty.
Third, the current wave of careful language-model pilots changes the input side of this problem more than the output side. Vacancy text, candidate notes and call summaries are the largest unread asset in a staffing group, and reading them well would widen the signal set beyond the three that model used. It does not change the clock. A pilot that reads well and delivers into nobody's working queue is a demonstration.
The lesson is not that six weeks is always possible. It is that the deadline is what makes the scoring honest. Given a quarter, every idea acquires a business case. Given six weeks, only the ideas with data in reach survive, and that filter beats the committee.