There is a particular kind of disappointment that sometimes follows a technology purchase in talent acquisition. The buying process went well, the demo was impressive, the business case stacked up and the implementation completed more or less on schedule. Several months later, the hiring problem that prompted the purchase looks much the same as it did before, and nobody can quite explain why.
This happens often enough that it deserves a better explanation than bad luck or a bad vendor. In most cases the technology is working exactly as designed. It is executing the instructions it was given, against the goals it was set, using the information it was handed. The gap between that and a genuine improvement in hiring outcomes is filled by decisions, and those decisions are made by people.
Our view, having spent years on both the platform side and the agency side of this industry, is that recruitment technology is sold as a product and delivered as a capability. The product arrives on day one. The capability takes longer, and it depends on how the platform is configured, what you feed it, how you read what comes back and who is watching when conditions change.
This guide is about that second part, because it is where most of the value sits and where most of it gets lost.
Buying Technology Rarely Solves The Underlying Problem
Employers usually reach for new technology at a moment of pressure. Vacancies are ageing, cost per hire has crept up, the team is stretched and something needs to change. The instinct to bring in better tooling is a reasonable one, and yet the purchase often precedes any clear diagnosis of what is actually going wrong.
Consider an employer struggling to fill roles across a branch network. Applications are down year on year, so the assumption forms that the advertising is underperforming and a programmatic platform is brought in to fix distribution. The platform does its job, reach improves and application volume recovers. Hiring, however, does not, because the original constraint was never distribution. Candidates were reaching the adverts in reasonable numbers, then abandoning a fifteen minute application form, and no amount of additional traffic into a leaking funnel produces more hires.
That example is deliberately simple, but variations of it are common. Technology is very good at removing a specific constraint, and it removes only the constraint you point it at. If the real problem sits somewhere else in the journey, in the job content, the pay position, the application experience or the speed of the response after someone applies, then a well-implemented platform will operate faultlessly while the outcome stays flat.
The useful discipline here is to describe the problem in terms of where candidates are lost before you describe it in terms of what you want to buy. Employers who do this tend to buy less technology and get more from what they buy, partly because they know what success would look like before they start.
What Software Actually Does, And What It Needs From You
Modern recruitment platforms optimise. They take a goal, a budget and a set of constraints, then make thousands of small decisions in pursuit of that goal far faster than any team could manage manually. This is genuinely powerful, and it comes with an implication that is easy to miss in a demo, which is that the platform will pursue whatever goal you give it with complete indifference to whether that goal is the right one.
Set a programmatic system to deliver applications at the lowest possible cost and it will find you cheap applications, because that is the instruction. Some of those applications will come from candidates who were never likely to be hired, from locations you are not hiring in, or from people applying to everything they encounter. The cost per application will look excellent in the report while the cost per hire quietly rises, and the platform has done nothing wrong.
Goal setting is therefore the highest leverage decision in the whole implementation, and it is usually made in a hurry during onboarding by whoever happened to be in the room. Where the data supports it, optimising towards hires, or towards a meaningful proxy such as completed applications from qualified candidates, changes the behaviour of the system entirely. Getting there requires that hire outcomes flow back into the platform, which is a technical dependency rather than a preference, and it is worth resolving before launch rather than six months in.
Alongside the goal sits the question of what the platform can act on. Bidding systems make decisions using the information available to them, which means your job titles, locations, descriptions, salary information and application flow are not presentation details. They are the raw material the optimisation runs on. A vacancy posted as “Team Member” with no salary and a location listed as the regional head office gives the system almost nothing to work with, however sophisticated the system is.
Implementation Is A Period, Not An Event
Most of what determines whether a platform performs gets decided in the first few weeks, and a surprising amount of it never gets revisited. Job taxonomy, feed structure, tracking, attribution rules, budget groupings and the definition of a conversion all get established during onboarding, frequently under time pressure, and then hardened into the way things are done.
Taxonomy deserves particular attention because it shapes everything downstream. How you group roles determines how budget is allocated between them, how performance is reported and whether you can tell that warehouse operatives in one region are behaving differently from warehouse operatives in another. Groupings that are too broad hide the problems worth solving, while groupings that are too granular fragment the data until no individual segment carries enough volume to learn from. Finding the level that matches how you actually manage hiring takes a conversation about the business, not a default setting.
Tracking and attribution follow closely behind. If the platform cannot reliably tell which applications became hires, it cannot optimise for hires, and you cannot evaluate the channels feeding it. Broken or partial tracking is the single most common reason a programmatic investment underdelivers, and it is rarely visible from the reporting, since the reports continue to populate with the metrics that are being captured correctly.
None of this is glamorous work, and it is the reason two employers running the same platform, in the same sector, with similar budgets, can end up with materially different results. The difference sits in the setup rather than the software.
The Quality Of Your Inputs Sets The Ceiling
Once a platform is running, its performance is bounded by what you give it. Several inputs matter more than others, and all of them are within your control.
Job titles carry more weight than most employers expect, because candidates search using the language of the market rather than the language of your internal job architecture. Internal titles that describe a grade or a cost centre are invisible to the people you are trying to reach, and a title that matches search behaviour will outperform a clever one consistently.
Location accuracy is the next most common problem, especially for employers with distributed sites. Roles listed against a head office postcode when the work happens forty miles away will attract applications from people who then discover the commute, and the resulting drop-off looks like a quality problem when it is a data problem.
Job content shapes conversion rather than reach. A description written to satisfy internal stakeholders, opening with three paragraphs about company history before reaching the work itself, loses readers who arrived with genuine intent. Content that leads with what the person will do, where they will do it, what it pays and what happens next tends to convert better, and the improvement compounds because the platform learns from stronger signals.
Salary transparency remains the input employers debate most and, in high volume hiring particularly, the one that moves conversion most reliably. There are legitimate commercial reasons to withhold it in some contexts, and it is worth being clear that withholding it has a measurable cost rather than treating the decision as neutral.
Finally there is the application flow itself, which sits outside the advertising platform and governs whether any of the preceding work matters. Every additional step, mandatory field and account creation requirement removes a proportion of the people who arrived willing to apply.
Reading The Data Without Misleading Yourself
Recruitment platforms produce a great deal of reporting, and only some of it carries information you can act on. Learning to tell the difference is a skill that develops with exposure, and there are a few principles that shorten the learning curve.
- Metrics closest to the outcome deserve the most attention. Cost per hire, apply-to-hire conversion and time to fill describe whether hiring is working. Cost per click and cost per application describe activity, and they can improve while the outcome deteriorates, which is why they make poor primary measures despite being the easiest to move.
- Small numbers cause more confusion than any other single factor. A location that made two hires last month and none this month has not necessarily declined, because the variation you are looking at is well within the range you would expect from chance. Weekly reporting on low volume roles produces a great deal of movement that means nothing, and reacting to it introduces instability into an account that was performing adequately. Roles with genuine volume can be reviewed frequently, while specialist and low frequency roles need longer windows before any conclusion is available.
- Attribution requires a degree of honesty that is easier to describe than to maintain. Candidates encounter your brand in several places before they apply, and the channel credited with the application is usually the last one in a sequence. This does not make the reported numbers useless, and it does mean that channels performing an earlier role in the journey will look weaker than they are. Where you can, look at what happens to overall performance when a channel is paused, since that tells you more about its contribution than its attributed conversions do.
- Segmentation is where most of the real insight lives. Aggregate performance across a national account tends to be an average of several quite different situations, and the roles or regions that are struggling get hidden by the ones that are working. Breaking performance down by location, role family and seniority usually surfaces one or two specific problems that account for a disproportionate share of the overall cost.
Knowing When To Automate And When To Intervene
Automation earns its place on decisions that are frequent, well instrumented and individually low stakes. Adjusting bids across hundreds of live vacancies, reallocating budget between sources based on rolling performance and pausing spend on roles that have filled are all tasks where a system will outperform a person by a wide margin, simply because it can act continuously and without fatigue.
Human judgement earns its place where the data is thin, ambiguous or newly invalid. A role you have never hired for has no historical pattern for the system to learn from, so early decisions are guesses that benefit from someone who has hired for something similar elsewhere. A sudden change in local conditions, such as a large employer opening nearby or a competitor repositioning on pay, invalidates the assumptions the optimisation has been running on, and the system will keep executing against them until someone notices.
Business context is the other category that automation cannot reach. A platform allocating budget purely on efficiency will move money away from a location that is expensive to hire in, which is the correct decision by its own logic and the wrong decision if that site is opening next month and the hires are non-negotiable. Strategic priority, brand considerations and operational commitments have to be expressed as constraints by a person who knows about them.
The practical arrangement that works is automation handling the continuous optimisation within boundaries, and people setting those boundaries, reviewing whether they still hold and intervening when something has changed that the system has no way of knowing about.
How Platforms Fit Together
Very few employers run a single system. A typical stack includes an applicant tracking system, a careers site, a programmatic advertising platform, aggregators, paid social, a candidate CRM and some form of reporting layer, and each of these was often bought at a different time for a different reason.
Problems concentrate at the joins. Data passing between systems gets truncated, delayed or mapped inconsistently, and the failures are quiet because each individual system continues to report that it is functioning. Source data is the usual casualty, with the result that advertising performance and hiring outcomes live in separate places and cannot be reconciled without manual work.
Two things help more than any amount of additional tooling. The first is agreeing a single source of truth for hires, so that every conversation about performance starts from the same number rather than from whichever system the speaker happens to trust. The second is mapping the candidate journey across systems and identifying exactly where a person moves from one to the next, because those transition points are where you will find both the tracking failures and the drop-offs.
Integration work is unrewarding and it removes a category of problem that otherwise recurs indefinitely, consuming analyst time and undermining confidence in every report that gets circulated.
Why The Expertise Has To Be Ongoing
A configuration that suits your hiring in January will not necessarily suit it in June. Labour market conditions shift, competitors change their pay positions, your own vacancy mix moves with the business, and seasonal patterns push demand around the year. The platform continues to optimise against the parameters it was given, and it has no mechanism for telling you that those parameters have stopped describing your situation.
Performance decay of this kind is gradual, which makes it easy to miss. Results slip slowly enough that each month looks broadly like the last, and by the time the trend is obvious in the reporting it has been running for a quarter. Catching it requires someone reviewing the account with enough regularity and enough context to notice that a pattern has changed rather than fluctuated.
Beyond monitoring, ongoing expertise brings comparison. An in-house team sees its own account, which is a single data point, whereas someone working across many employers in similar markets can tell you whether a rising cost per hire in your sector reflects something you have done or something everyone is experiencing. That distinction changes the response entirely, and it is difficult to arrive at from the inside.
Bringing It Together
Technology has made recruitment advertising considerably more effective than it was a decade ago, and it has not removed the need for judgement. If anything it has concentrated that judgement into fewer, higher leverage decisions, made at setup and revisited as conditions change.
Employers who get the most from their technology tend to share a few habits. They diagnose the problem before they buy, they treat inputs as part of the system rather than as content, they measure against outcomes rather than activity, they automate the repetitive decisions while keeping people close to the ambiguous ones, and they revisit the setup as the market moves rather than assuming it still fits.
None of that requires a larger budget. It requires the software and the service to arrive together.
Yoke was founded by five people who spent years working inside Indeed, giving us a unique understanding of the platform. We know how the auction behaves, what actually influences job visibility, and which changes to job content, budget pacing, or campaign structure genuinely improve hiring outcomes.
That experience shapes how we manage every client account. Instead of chasing quick wins or making changes for the sake of it, we focus on the things that make a measurable difference, helping you get more from the budget you’re already investing.
As an Indeed Agency Partner, we work closely with the platform while staying focused on what matters most: helping employers hire more effectively.
If you’d like to see where your account could perform better, start with our free Indeed Health Check, or get in touch for a conversation. We’d be happy to talk it through.
