What a performance provider is hired to do
A business can hand a provider some or all of what the performance marketing pillar calls the shared layer of paid acquisition: the economics that decide what a result may cost, the campaign structure, the conversion events and values, the roles of the channels, the testing and scaling, and the reporting. The platforms supply the inventory; the provider's work sits across them.
That separates this hire from four neighbours: the Google Ads hiring guide for paid search, the Meta Ads hiring guide for that platform's portfolio, creative supply and conversion signal, the guide to hiring a media agency for trading status, rebates and audit access, and the growth marketing hiring guide for a growth programme's model, cohorts and learning log. This guide starts where one provider is asked to hold the targets, the events, the plan and the reporting across several platforms; the questions in those four still apply.
The decision to hire at all is made in the in-house against agency comparison. The selection guide covers any agency and asks a freelance specialist for the same written scope, so, as general practice, what follows applies equally to a consultant or a freelancer.
The first scoping decision is who holds the layer above the platforms. As general practice, there are three arrangements, and choosing none leaves it with nobody.
- The business holds it. An in-house lead owns the targets, the register, the plan and the report; the page on managing specialists sets out what a manager can still judge when the craft is not their own.
- One performance provider holds it, drafting the targets, the register and the plan and running the platforms while the business approves and keeps the records. This guide assumes this arrangement.
- Several providers share it, and no one answers for the points where their work meets; the coordinating several agencies guide explains how to name an integrator.
Under the second arrangement the table divides the work, as general practice; each item the business keeps needs a named owner on its side.
| Part of the work | What the provider can take on | What the business keeps |
|---|---|---|
| Targets | Draft them from the business's margin figures, with the working | The margin, the payback window agreed with finance and the decision to change a target |
| Conversion events and values | Propose the register, build and verify the events | The written definition of each result, the values and approval of any change |
| Structure | Build and name campaigns to follow the business's decisions | The decisions the structure serves and approval of a restructure |
| Channel plan | Propose roles, budget by role and a test reserve, and run inside the monthly limits | The budget, any move of money between channels and the agreement with finance |
| Audiences and customer lists | Build segments, then upload and refresh files on the register's schedule | The definition of a customer, the basis for using the data and the files |
| Tests, reporting and review | Design and run the tests and prepare the monthly report | Who calls each result, the source of the headline figures and the review's decisions |
The records that stay in the business's own name
The selection guide wants four things kept in the business's hands: its customer data, ad accounts, analytics properties and the creative that has been paid for. The Google Ads hiring guide turns that into account ownership, access and billing, and the Meta hiring guide asks the same of the portfolio, the ad account and the dataset. Running several platforms adds the documents that say what each platform is being asked to do. As general practice, the table gives each record's home on day one; each platform's own documentation says what it allows on ownership, access and transfer.
| Record | Home from the first day | Guide |
|---|---|---|
| The ad account on each platform, with its billing | The business's own login or portfolio, the provider given access and billing in the business's name | The agency onboarding checklist |
| The tag container, server route and datasets | The business's own container and datasets, each event deduplicatedRemoving double counts when one conversion is reported by more than one source. | The tracking verification checklist |
| The register of events | The business's own document store, with dated versions | The conversion events guide |
| The targets and their working | Beside the margin figures from finance | The performance economics guide |
| The plan and the budget agreement | The business's own files | The cross-channel planning guide |
| The customer lists and their register | The business's own export, with uploads only on the register's schedule | The customer lists guide |
| The test record and the decision logA running record, one row per decision, of what changed and why, what was expected, who decided and when the result will be checked. | A shared document the business controls | The decision log template |
Two of these requirements work better as pass-or-fail gates than as scored criteria: as general practice, the accounts and datasets sit in the business's name, and the targets rest on margin figures the business supplied. A proposal that fails either one drops out however well it scores, which is how the weighted scoring matrix framework handles a must-have: as a gate and not a weight.
Settle the numbers before proposals are requested
The selection guide treats the brief as the point at which agency work is made or lost. As general practice, the request carries five items; a business that lacks one can write a rough version or leave it for the opening round.
- The margin and the target it allows. The contribution margin one result earns, the break-even that follows and the highest cost the business will pay for a result, worked out with the contribution margin and break-even guide, not borrowed from a platform's suggestion. The break-even cost per resultHighest cost per acquisition at which a sale still covers its variable costs. is the limit, not the target.
- One written definition of each result. A purchase is a completed order, a qualified lead is one the sales team has accepted, a booking is a confirmed appointment, each with the system that records it.
- The money. The paid media range and, separately, a ceiling on everything the provider may cost, fee included, fixed before any fee model is considered.
- The data the provider may read. Without sight of sales data an agency cannot optimise for sales, as the selection guide says, so name what it may see.
- The platforms and the jobs in scope, with the named owner on the business side.
Send each provider an identical brief and the same figures, settle the scoring criteria before the first reply and ask to meet the people who would staff the account. The agency request for proposal template supplies a shared brief, a question list and a scoring grid, and the agency scope of work template is the document to fill in once a provider is appointed, with what falls outside it and how a change is priced, because the selection guide traces most fee disputes to a vague scope.
Opening round: targets and signals before spend moves
The Google Ads hiring guide asks for an audit before a retainer. For work across platforms, as general practice, buy a bounded opening round that settles the targets and the signals and leaves the live accounts as they are; its statement of workA supplier's written confirmation of what it will deliver, by when, at what price and how it will invoice, against which the order is raised. lists what it delivers.
- Targets with their working, blended and by channel role, tightened by the overlap between the platforms' claims and the business's records, each with its date and source.
- A verified register of events, each row tied to the written definition and tested with a real conversion found in the platform and in the business's own records.
- A reconciliation for a closed month, with the business's results, the analytics figure, each platform's figure and the ratio of each to the business's number.
- A review of the structure: which splits exist, which decision each serves and which are only platform habit.
- The plan on one page and a ranked list of tests, with the decision log started.
The round should change nothing that a measurement fault does not force. The conversion events guide notes that both Google and Meta treat a changed conversion action or optimisation goal as a reason for a bid strategy or ad set to learn afresh, so a change to an event follows the round as a planned change, one platform at a time, and, as general practice, a change to a target or the structure follows it the same way. A tag that records nothing is a fault, not a design choice.
The Google Ads hiring guide separates routine optimisation within agreed limits from changes that need sign-off; applied across platforms, as general practice, these need the business's approval:
| Change | Why it needs the business's approval |
|---|---|
| The optimisation eventThe conversion event an ad set or campaign is told to optimise delivery towards, such as a purchase, lead or install. or a conversion valueThe monetary value an advertiser assigns to a conversion, such as the order's revenue or a lead's expected worth. | It changes what each affected platform learns to buy, and a change of event can put a platform back into learning; the conversion goal change checklist is the gate |
| A target, or an attribution setting or window | A target sets what the budget may pay for a result; an attribution change alters what delivery learns from and the figure that reaches the report |
| The budget, a move of money between channels or a new platform | Money is the business's, and moves between channels belong to the monthly review; a new platform needs a role, a target and a share of the test reserve first, and the guide to setting spending authority says who may approve what |
| A restructure | It resets learning and makes reports before and after hard to compare |
| A customer list upload outside the register's schedule, or a new audience source | The basis for using the data is settled first, as the customer lists guide asks |
| A generated or enhanced asset switched on in an account | The rules for agencies on AI use settle which settings a provider may change |
Written questions for shortlisted providers
Before any provider is appointed, ask the providers on the shortlist identical written questions, place the replies next to one another and score them against criteria agreed beforehand, as the selection guide asks. As general practice, give more weight to a reply that names who does each task than to one that speaks in generalities.
| Question to put | What a careful answer holds |
|---|---|
| How will each platform's target be derived from our margin and reconciled to one blended target? | A calculation from the margin or lead value the business supplies, with working it can follow, not a platform's suggestion or a recent average |
| What will each platform be told to buy, and who decides what counts as a result? | One written definition per result, a register, the event closest to the result that a platform can still learn from, and the business approving changes |
| How will campaigns be split, and why? | Splits that each serve a budget, a target or a report, with restructures staged; the campaign structure guide sets out the test |
| How will budget be divided across channels, and how will you test what remarketing adds? | Roles, targets that differ by role, a protected test reserve and money moved at the monthly review; for remarketing, a cap and a holdout, with reported return read beside incremental return, as the comparison of prospecting and remarketing sets out |
| What does your fee rise with? | A plain answer, read against the section on the fee below |
| On the day we part, which of the records above would already be in our hands? | The accounts, the register, the lists, the test record and the log |
Warning signs in a performance proposal
None of these signs is a verdict. Each is a reason to put a further question in writing and to keep asking until the answer is specific.
- Targets with no stated source, or one target for the whole plan. The performance economics guide treats a recent average as a record of the past, and the planning guide warns that judging demand creation on search's last-click target means it is cut first.
- Platform figures added together. Each platform credits itself with the conversions it touched and several can touch one sale, so the sum overstates what the business received.
- Remarketing and brand search as the headline. RemarketingAdvertising to previous visitors or engagers. and brand search report high returns because a large share of the people they reach were already about to buy; the remarketing holdout test shows how to find what they add.
- Events chosen by whoever installs the tag, or accounts and datasets in the provider's name. The conversion the installer marks is the one the budget buys, and the history, the audiences and the learning leave with a provider that holds the accounts.
- A rebuild before the account has been read, or winners declared from a platform's own split-test report. The Google Ads hiring guide lists the rebuild as a warning sign, and, as the testing and scaling guide says, a platform's split test still measures under its own attribution, so budget-sized decisions need a holdout outside the platform.
- A promise of a specific result. The Google Ads hiring guide's point holds across platforms: the auction is outside any provider's control, so a result promised before the accounts have been read is only a sales pitch.
- Silence on customer data or on handover. The customer lists guide puts a recorded basis before any upload, how the arrangement ends belongs in the scope, and contract wording is for the business's legal adviser.
How the fee steers the work
The selection guide lists the fee models, from the retainerA standing fee paid regularly to an agency or supplier for continuous work with a stable scope, as distinct from project fees, time and materials or commission. to the performance-related element, and says what each suits or rewards. Three cautions from other site guides apply to paid media run across platforms; applying them to a provider is general practice.
A fee that rises with media spend gives the provider a reason to recommend more spend: the selection guide notes that a commission on media spend rewards spending, not effectiveness, and the diagnosis of agency reports that flatter lists a fee that grows with spend among the reasons that no one is paid to notice flat sales. As general practice, the business holds the brake. The target-led paid media budget tutorial records it as an agreement with finance: the target, a stopping rule on marginal cost per resultThe extra spend divided by the extra acquisitions it produced when spend changed; usually above the average, and the figure scaling and cutting decisions need., a ceiling on monthly spend and review points. Ask a provider to recommend an increase only against that rule.
A performance element tied to what a platform reports, whether its cost per result or its return on ad spend (ROASRevenue generated per unit of ad spend.), matches the pattern the Goodhart's law framework describes, and the performance economics guide adds that a return target can be hit while the business loses money. Tie any element to a result measured in the business's own records, pair it with a guardrail metricA metric an experiment must not worsen beyond an agreed bound, watched beside the primary metric that decides the test. that cannot be allowed to fall, and leave the definitions with the business.
The provider's fee is part of what a customer costs: the comparison of cost per acquisition and customer acquisition cost puts media alone in the campaign figure and agency fees, creative and tools in the business figure, and customer acquisition costCost to acquire one new customer. is the one finance recognises. Ask each provider to show its fee, set-up work and tools as separate lines, and hold each total against the ceiling fixed before the brief went out.
Reports to test against the business's own records
The reporting guide sets the order of a report that drives decisions, and each line below is paired with a record the business already holds, so that the business can test it.
| Line in the report | How the business tests it |
|---|---|
| Plan against actual | The headline figures from the order system or the CRM set beside the plan, as the performance reporting guide lays out |
| Efficiency, volume and the marginal view by channel | Each channel's cost per result against its own target and its last step of spend against the stopping rule, with the attribution source named, recalculated with the CPA calculator and the ROAS calculator from spend and the orders the business can see |
| New customers | The order system's new-customer flag, read beside the provider's total, as the new and returning customers guide describes |
| Reconciliation | The opening round's table, with last month's ratio beside each ratio; a steady gap is a feature of the measurement and an abrupt change signals a fault, as the platform and analytics discrepancies guide explains |
| Channel evidence | The date and finding of each major channel's last test, read against the designs in the incrementality testing guide, so that incrementalityConversions caused by marketing. is stated and not assumed |
When the provider's figures climb and the business's own records do not move, the diagnosis of agency reports beside flat sales separates a measurement gap from a performance gap, the reading small numbers guide helps to judge whether a weak week is a real change, and the diagnosis of a high return without profit covers a healthy return on a thin margin.
Judging the provider apart from the programme
As general practice, give the provider and the programme separate verdicts. The offer, the landing pages, the stock position and the margin also shape the programme's numbers, and the page on how ad auctions price performance media separates the inputs a buyer sets from the market the auction sets, which is not the provider's to control.
At the start the business can judge the provider's own deliverables: access in the business's own accounts, a verified register, targets that arrive with their working, a reconciliation, a started plan and log and the scope's service levels met. The programme can be judged once late conversions have been counted and any learning period has ended, as general practice on unchanged definitions and the business's own outcomes: if a platform's figures rise while the business's own orders or qualified leads stay put, only the report has moved.
The selection guide's two-way scorecard scores the business as well as the provider, so use it here. When the verdict on the provider is poor, the tutorial on improving an underperforming agency sets out a written plan with review points, and the pre-renewal review belongs before any agreement renews.
Handover when a provider is replaced or the work comes in-house
The agency offboarding checklist sets the order in which accounts and assets are collected and access is withdrawn. As general practice, a programme across platforms has these further items, which a plain inventory of accounts would leave out, whether the work moves to a new provider or in-house.
- The register of events and its verification record, so that whoever takes over can see what each platform is optimising for.
- The targets, the plan and the budget agreement, with the working behind each target.
- The customer lists and their register, with the date of the last upload, the basis for use and any upload still in flight.
- The test record and any holdout in flight. As general practice, a running holdout is either read out on its planned date or closed early, with the reason written down.
- The decision log and the naming convention, which record why each account looks as it does.
- The access list. As general practice, a dated access reviewA periodic check of who can use each marketing tool: every account matched to a current person, supplier or integration, with owners, deputies and narrow roles. of each login the provider's people held, taken before access is withdrawn.
If the provider still holds the accounts, the diagnosis of an agency that withholds the ad accounts works through each one, and the troubleshooting page on falling results after an agency switch covers what to check when numbers slip. Ending a provider is not ending a channel: the guide on stopping a paid advertising channel sets out the evidence needed for an exit and notes that an agency which ran only that channel leaves with it.
Weighing performance proposals for a dollhouse retailer (illustrative)
Zibbowarp Dollhouses weighs three performance proposals
Zibbowarp Dollhouses, an invented online seller of dollhouses, dollhouse furniture and lighting kits, spends £9,000 a month on paid media, £5,400 on Google Search and Shopping and £3,600 on Meta, and has run both accounts itself. An average order of £64 leaves a contribution margin of £24 after goods, packaging, delivery, payment fees and a returns allowance, so break-even is a cost of £24 an order and a return of about 2.7. Zibbowarp sets a target of £20 an order with all costs counted, including a provider's fee, and a ceiling of £13,500 for the first six months of any provider's work. Each of three providers is sent the same brief and the same figures.
Before any scoring, two gates apply: the accounts and datasets stay in Zibbowarp's name, and the targets rest on the margin figures Zibbowarp supplies. Proposal A plans to open accounts in its own name, takes each platform's suggested target as the starting point, reports platform results only and charges a commission on media spend, so it fails both gates and drops out. Proposal B, from an agency, and proposal C, from a freelancer, clear both. Zibbowarp wrote the weights down before opening any proposal: targets and economics 30, signals and measurement 20, plan and tests 20, reports the business can check 15 and shape of the fee 15, each criterion scored out of 5. B scores 5, 3, 3, 5 and 4, a total of 405 out of 500, or 81 out of 100. C scores 5, 2, 3, 4 and 5, a total of 385, or 77 out of 100. With the fee weighted at 30 and targets at 15 and the other three weights unchanged, B scores 78 and C 77, so B still leads, but by a single point, which the weighted scoring matrix framework would treat as close to a tie, so the fixed-fee opening round is bought first, as the further evidence the framework asks for, and the retainer follows it.
Zibbowarp appoints B for a fixed-fee opening round of £3,200 and then a retainer of £1,900 a month, which comes to £12,700 over the first six months, the round plus five months of retainer, inside the ceiling. The round's reconciliation of the last closed month puts the platforms' own cost per purchase at £11.84: Google claims 430 purchases and Meta 330, 760 in all against £9,000. Zibbowarp's order system records 540 orders from tagged paid traffic that month, so the platforms together claim about 1.4 times the orders recorded, and the business's own cost is £16.67 an order on media alone, or about £20.19 once the £1,900 retainer is added to the same month, with the round's fee left aside, just above the £20 target. B shows the fee as its own line in the report, proposes tightening the platform targets by the 1.4 overlap, for Zibbowarp to approve and apply one platform at a time after the round, and notes that £2,000 of the £3,600 on Meta goes to remarketing. No money moves between channels. Zibbowarp commissions a holdout on that remarketing line, and the register of events, the targets and the log stay in Zibbowarp's own files, each with a named owner on its side.
The channel playbooks hub lists the section's other guides and the problems each one addresses.
Common mistakes
- Ranking providers ahead of the ownership gate. A proposal that builds in the provider's own accounts looks cheaper until the history, the audiences and the learning leave with it.
- Letting the provider set the target. A platform's suggestion or a recent average describes the past; derive the target from the margin the business supplies.
- Judging on platform figures. Platforms count what they touched; judge on the business's own orders and qualified leads, with the reconciliation beside them.
- Paying on spend or on a platform's return. One rewards spending and the other can be hit while the business loses money; agree a stopping rule and a guardrail before the fee model.
- Signing with no exit in view. The register, the lists and the log leave with the provider unless the business holds them from the first day.
