Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

Mobile & App Marketing › App Marketing

Hiring and Evaluating an App Marketing Agency

Choosing an app marketing agency or freelancer: the store, measurement and ad accounts you keep, questions on listings, installs and messaging, and reports.
Guidebeginner20 min readUpdated

What an app marketing provider is hired to do

Outside help with an app can come from a freelancer, a studio or an agency, and may cover any of four jobs: the store listing, which is the work known as app store optimisation, paid installs, messaging or measurement. Settle which of the four is being bought before anyone is asked for a proposal, because a provider that quotes for all four and one that quotes for the listing alone cannot be compared.

The app marketing operations playbook is the map of what there is to buy. It splits the programme into seven roles (an app marketing lead, a store listing owner, a release owner, a user acquisition manager, a messaging owner, an analyst and a support lead) and ties them to a weekly, monthly and quarterly rhythm. A provider might hold one of those roles or a share of several. The playbook does not say which of them may sit outside the business, and the next section gives labelled general practice on that. It does name one job that tends to be left without an owner, the routine of replying to reviews, since no team takes it by default. So, as general practice, the first question to any provider is whether replies are in its scope, and if so who writes them. The rating prompts and review replies tutorial asks for named owners, console access for everyone who replies and bugs sent to the product team along with the app version, and a provider that takes replies on should be able to say how it would meet each.

What a provider supplies is capacity and skill, not the business's own accountability for results. The agency selection and review guide runs the process for any agency, from the shared request to the exit. This guide adds what that process leaves open for an app. Fee levels, retainer sizes and legal terms are outside it, and the contract itself belongs with the business's legal advisers.

Which work to buy in and which to keep close

The comparison of in-house against agency work supplies the sorting rule between work to buy in and work to keep close, and the table below applies it to an app. It is general practice, because no site guide says which app roles may sit outside the business.

WorkWhat a provider can bringWhat stays close to the business
Store listing and its testsKeyword research for each store, listing text and screenshots, and tests run with the stores' own toolsThe promise the listing makes, the check of its claims against the app, and the history of what went live
Review repliesA reply routine and drafts, where the provider has console accessA named owner for replies and a route that carries bugs to the product team
Paid installsNetwork know-how, campaign set-up and creative in volumeThe choice of activation event, the budget, the ad accounts and the decision to raise spend
Messaging programmeBuilding and tooling the triggers and message typesThe permission people have granted, the cap per person and the decision to send; this is lifecycle work, which the comparison above places in-house
Measurement and reportingPartner set-up and event checksThe definitions, the system of record, the monthly reconciliation read against the business's own records and the person who reads the reports

Write the split down in a RACI matrix, with rows for decisions rather than tasks and one accountable role for each row. As general practice, keep the accountable role on a name in the business even where the provider is responsible for the work, so nobody has to guess who owns a decision when a campaign, a listing change or a reply is in dispute.

Whoever holds those accountable roles has to judge the provider's work without necessarily being able to do it. The guide to managing a specialist whose work you cannot do covers how: read results from the business's own records, ask for the reasoning behind each proposal along with the alternatives set aside, borrow a written standard and bring in an independent view where the method matters.

What stays in the business's name

The selecting guide wants the business to own the accounts, data and creative it pays for, and the agency onboarding checklist turns that into practice. For an app the same rule reaches further, as general practice: the accounts that hold an app's history are opened in the business's name, and the provider works through access granted to it.

ItemWhy the business keeps itWhat the provider is given
The store consoles for each platformListing text, screenshots, ratings and review replies live there, and the history of what went live is what the next person readsAccess to the parts the agreed work needs, within what each store's documentation allows
The measurement partner accountIt is the system of record for installs and for which campaign earned themAccess to read and, within agreed limits, to configure, with each change recorded
The analytics projectApp data reaches it through the app's SDK, and its event history is what cohorts and activation are read fromRead access, with changes to events made only through the release check
The ad network accountsSpend history, audiences and campaign structure, and what the network has learned from themAccess to set up and adjust campaigns inside limits agreed in the scope, with spend visible and billing decided by the business
The messaging tools used for push and in-app sendsMessage history, triggers, audiences and the cap per personAccess to build and edit messages under the agreed cap
The recordsThe listing history, test log, route table, event list, register of custom listing addresses and creative filesWorking copies while the engagement lasts, with the master copies in the business's own storage

The app marketing audit scorecard asks an auditor for four things before scoring: the analytics project, the attribution set-up, access to each platform's console and the tool that sends push and in-app messages. As general practice, a provider reading the same material needs the same four, so they are the first accounts to name in a request for proposals, with the ad network accounts beside them. On the measurement partner, the attribution guide makes it the system of record and lists attribution windows, the event list, the conversion value schema, network integrations, fraud rules, the raw data export and consent as the settings where attribution problems are made. It also asks for an end-to-end test with a test device before any real budget runs, so, as general practice, a provider should say who will run that test and who holds the partner login.

As general practice, ask each provider where it would build. An agency that builds in its own accounts leaves the business without the history of what those accounts learned. Where a store or a network documents how access is granted, moved or removed, that documentation is the reference; this guide does not describe any store's account roles, ownership or transfer rules.

Questions to put in writing before weighing proposals

Send each shortlisted provider the same written request and the same background, built on the agency request for proposal template, so that answers arrive in the business's format and can be compared side by side. The questions below are the app-specific part of that request. Each is meant to be answered in plain words.

QuestionWhat a specific answer covers
Which store fields will you work on, and how will you test a change?Research repeated for each store and each language, one hypothesis for each test, and the stores' own listing tests, with no promise of a ranking
Who checks the listing's claims, and who approves them?A check against the app as it ships and the stores' review rules, and one approver in the business
How will custom listings be built and retired?One for each campaign theme rather than each ad, built in each console with its address recorded and retired when its campaign ends
Who replies to reviews, and how do bugs reach product?Named reply owners, console access for those who reply, and bugs passed on with the app version
Which event will campaigns optimise to, and when would that change?The deepest event that has the volume a network needs, with installs first while events are still thin
How will you judge each source?Cost per activated or paying user and retention, not cost per install alone, measured through the partner
How much creative will you supply, and what do your agreements with smaller networks say?Creative in volume and rotated, and a quality clause that defines an invalid install, says how it is detected and says it is not paid for
Which tool sends the messages, and who holds its login?A named tool, with an owner on the business's side
How will one cap per person cover campaigns and teams?A shared cap and a priority order, with the notification permission treated as the asset to protect
Which tool is the source for installs, which for behaviour and which for spend?One answer for each question, with the duplicate report switched off
Which activation event will you report, and who agreed it?One event counted the same way by the partner, the networks, analytics and product
How will links and events be tested with each release?A route table that lists the intended landing screen for each link, and a check of events on the release build
Who will do the work, and can we meet them?Named people, their share of time and what happens when one of them leaves

Several answers can be tested against a site guide. The app store optimisation guide says neither store publishes ranking weights, that the indexed fields differ between the stores, that localising means repeating the keyword research rather than translating, and that the stores' own listing tests are the only reliable way to learn what moves conversion. The custom store listings tutorial builds a page for each campaign theme and retires it when its campaign ends. The app install campaigns guide optimises to the deepest event with enough volume, judges each source on activation and retention and asks that agreements with smaller networks carry a quality clause.

The push and in-app messaging guide treats the notification permission as worth more than anything else the channel has and asks for one cap per person across the campaigns and teams that can send. On measurement, the activation event tutorial works out one event from retention data, the deep link testing tutorial builds the route table, and the tutorial on checking app events with each release walks a release build's events through clean test devices before it ships, including consent states and postbacks to the partner.

Weigh the answers in a weighted scoring matrix: fix the criteria and weights before a proposal is read, apply must-haves as gates that an option either passes or fails, and attach evidence from the written answer to each score. A plan to open the store, measurement or network accounts in the provider's own name is a candidate for a gate, because the framework drops an option that fails a must-have, however well it scores elsewhere. Ask to be introduced to the people who would carry out the work, and name one approver in the business.

Warning signs in a proposal

  • Accounts opened in the provider's own name. The store, measurement partner or network accounts that hold the history would have to be won back at the end.
  • A promised ranking. Neither store publishes ranking weights, so a provider has no published basis for promising that a change will move an app to a given position. As general practice, the honest offer is research and a test plan.
  • Listing text that stretches the truth. The stores' review rules limit what a listing may claim, and misleading text risks rejection and, repeatedly, the account. A provider that writes claims the app cannot support spends the business's standing with the store.
  • A rating plan that chooses who is asked. Both stores bar incentives for ratings and bar screening people by mood before they are asked, so a prompt gated on a satisfaction question is exactly the screen they bar.
  • A rating reset offered as a quick win. Apple lets a developer wipe the summary rating when a version ships, but the site's guides count that as earned only by a release that cured the complaints in the old reviews.
  • Cost per install as the only result. It measures what a visitor cost and says nothing about whether the visitor became a user, so a proposal that reports it alone cannot show whether the installs activated or stayed.
  • Installs bought from a smaller network with no quality clause. Without one, nothing says what counts as an invalid install or that it is not paid for.
  • Network dashboards as the report. Each network counts the installs it claims, and the measurement partner deduplicates and is the system of record.

Listing audit, install campaigns, messaging or measurement first

As general practice, buy the opening work in separate pieces, each with its own quote, deliverable, review and end, instead of one bundle. An audit or a set-up is a defined piece of work, so a project fee fits it; ongoing management is continuing work with a stable scope, so a retainer fits, which is how the selecting guide matches fee models to work.

Piece of workWhat the business receivesWhat to verify before paying
Audit of the current stateA scored read of the store presence, the measurement set-up, the install campaigns and the messaging, with fixes in priority order and an owner for eachThat the activation event and the attribution set-up were scored before the listing, and that no criterion was left out
Measurement set-upThe event list, the route table, notes on how the partner is configured and a test recordThat an end-to-end test ran with a test device before any spend, and that the accounts sit in the business's name
Listing workResearched text, screenshots, a test plan and the history of each changeThat the app store listing checklist was run before submission
Install campaignsCampaigns built on the agreed optimisation event, with creative and launch checksThat each launch passed the app install campaign launch checklist
Messaging programmeMessage types and triggers built under the agreed cap, with links testedThat the cap is shared and each message lands on the screen it promised
Ongoing managementReports and a dated change log within agreed limitsThat reports tie out as the next section describes

As general practice, put the scorecard's criteria beside an audit proposal and note any the provider would skip. The scorecard's gates give the order for the rest: the activation event and the attribution set-up are scored first, and while either is missing the acquisition and retention rows stay provisional and budget increases wait. As general practice, an app with no agreed activation event, or with postbacks that do not work, buys the measurement piece before paid installs, whatever the proposals say about speed.

Put each piece into the agency scope of work template, the written statement of work the provider is paid against, with deliverables worded so that both parties can see whether a request falls inside the work, what is out of scope, the accounts the business owns and the handover items. The selecting guide treats the scope as the most important of the documents that the relationship rests on.

Reports the business can check

As general practice, put these in the request from the outset.

  • A baseline before anything changes. Installs, activation and store conversion by source for the period before the provider begins, so that later movement has something to be read against.
  • Installs from the partner, with the other counts labelled. The partner's deduplicated count is the one for budget decisions. The reconciliation tutorial labels what each other count is for and builds a monthly table with the expected direction of each gap and a tolerance set from the team's own history, and the onboarding checklist asks for a rule on which number counts when sources disagree, agreed in the first weeks instead of leaving the agency to report platform numbers by default.
  • Cost per activated user next to cost per install. The CPA calculator shows how CPM, click-through rate and conversion rate combine into cost per acquisition, and the install campaigns guide says the figure to watch is the cost of the event the business cares about, not of the install.
  • One activation event, named on each report. It is the activation event agreed with the business, not the provider's own choice.
  • Retention by install cohort and source. Totals alone hide a fall in newer cohorts.
  • A dated change log. Each change to the listing, the campaigns and the messages, with its date, so that results can be explained.
  • A note of which tool owns installs, behaviour and spend. The measurement plan for app installs wants the owner of each question named and the duplicate report ignored.

Lay the provider's figures next to the business's own records for the same closed period. When the provider's report keeps improving while the business's own records stay flat, the troubleshooting page on an agency report that improves while sales stay flat sets the two numbers side by side and then goes through the possible causes one at a time.

When the work can fairly be judged

Retention shows only over time. The audit scorecard asks for an evidence window that lets several install cohorts be read beyond their early weeks, held fixed for each rescore, and the playbook judges a source only on reconciled numbers and on cohorts that are old enough to have shown the chosen retention measure. The first weeks of a provider's work therefore cannot show whether its installs stayed. The guide to launching a new app reads the first month by install cohort and source and uses a gate before spend is raised: set down beforehand what the first cohorts need to show, then choose between carrying on, carrying on with changes, holding until a named problem is fixed, and stopping. As general practice, the same gate suits a new provider.

As general practice, agree before work starts the activation window, the retention measure, the date of the first review and what the cohorts must show to justify more budget, and keep them fixed from one review to the next.

A fair judgement also separates causes that sit outside the provider. A store review sits between a finished build and a live one. A first session that loses new users is a product matter, not a listing matter. A release that renames an event moves a metric with no campaign change behind it. Where people have not allowed tracking, iOS results arrive delayed and in aggregated form through SKAdNetwork, so a short run of iOS numbers is not a verdict either, and small cohorts move around from one week to the next. Weigh the provider's explanation of a movement against these causes before assigning blame, and use the review rhythm the selecting guide describes for the conversation.

Changing provider without losing the accounts

As general practice, if the accounts, the history and the definitions have stayed in the business's name, a change of provider is a transfer of work and not a rescue, and the business asks the provider to keep a handover file current from the start of the engagement instead of writing one at notice. It holds the listing history, meaning each text and screenshot set that went live and its date; the test log with each test's hypothesis and result; the route table; the event list; the register of custom listings with their addresses and the campaigns that point at each; the campaign structures and optimisation events; the messaging triggers, audiences and caps; notes on how the measurement partner is configured; the creative files; and the register of who holds which access.

The agency offboarding checklist supplies the order, with access removed near the end. As general practice, a dated access review then confirms that the outgoing provider's logins to the consoles, the measurement partner, the analytics project, the networks and the messaging tools are gone.

The first release under a new provider is the point where an app can still lose what the old set-up depended on. As general practice, run the event check and the link tests on that build. The troubleshooting page on results that fall after switching agencies dates a drop against the handover and checks what the move lost, such as rebuilt campaigns that start learning again and tracking that broke in the handover. Where an outgoing provider holds an account in its own name, check the store's or the network's own documentation for what can be moved and how.

A coin dealer hires its first app provider (illustrative)

Illustrative example — not a real result

Tamborel Coins weighs three proposals

Tamborel Coins, an invented coin dealer with a catalogue app in which collectors record their own collections, wants outside help with its listing, its install campaigns, its messaging and its reporting. The marketing lead sends the same written questions and the same background to three providers. One proposal asks to open the store and measurement accounts under its own name, one promises a place in store search results and one reports cost per install alone. Before reading any answer, the lead fixes five weights and one gate: any option that would hold the store, measurement or network accounts in its own name is out, whatever else it scores. Each answer is marked from 1 to 5 against the written questions.

Criterion and weightProvider AProvider BProvider C
Gate: accounts in the business's nameFailsPassesPasses
Measurement and reporting (30%)Out at the gate24
Install campaigns (25%)Out at the gate34
Listing and ratings (20%)Out at the gate23
Messaging (10%)Out at the gate33
People who will do the work (15%)Out at the gate44
Weighted total (of 5)Out2.653.70

Provider B's promised ranking scores low because it offers no test plan and no activation reporting. Provider C's proposal reports cost per install alone, but its written answers agree to report installs, activation and cost per activated user, reconciled with the consoles of both stores and the measurement partner, and to work through access that Tamborel grants it. Swapping the weights of the measurement and listing rows leaves C ahead, 3.60 to 2.65, so the order holds. Tamborel opens the accounts in its own name, gives C access, and asks for the listing history, the test log, the route table and the event list as a handover file from the first week, kept current from then on.

C's first monthly report shows 2,050 downloads in the store consoles, 1,740 installs at the partner and 1,690 first opens in analytics, the store gaps explained as downloads that were not opened and reinstalls, the small partner-to-analytics gap put down to their different rules, each written into the monthly table as the expected direction of each gap, with the tolerance to be set from the team's own history. Of the 900 installs the partner credits to paid campaigns, 279 (31%) reach the activation event, "first collection item added", and the report gives cost per activated user next to cost per install. Tamborel does not judge the work on that month; it waits until the first cohorts have passed the activation window and the retention measure it set, then holds the review whose gate it wrote in advance.

The app marketing hub collects the guides to the listing, the campaigns, ratings and launch that a provider's work will draw on.

Common mistakes

  • Letting the provider open the accounts. The history of the listing, the partner and the networks then belongs to the provider. Open each account in the business's name and grant access.
  • Buying a ranking. Neither store publishes ranking weights. Ask for a research and test plan instead of a position.
  • Judging on cost per install. It prices a visitor and says nothing about the user. Ask for cost per activated user and retention by source.
  • Comparing proposals in different formats. Answers in each provider's own format cannot be scored side by side. Send the same written questions to each.
  • No owner for review replies. The job belongs to no team by default. Name a reply owner and a route that carries bugs to product.
  • Reading the first cohorts as a verdict. Retention shows only over time, and small cohorts move around. Write the gate in advance and wait for cohorts old enough to read.
  • Starting the handover file at notice. The listing history, test log, route table and event list are hardest to rebuild when the provider is leaving. Ask for them from the start.
  • Nobody in the business who can judge the work. Judging is not the same as doing. Keep one accountable person who reads results from the business's own records and asks for the reasoning.

Frequently asked questions

Should the provider open the store, measurement and analytics accounts, or should we?

As general practice, the business opens them in its own name and grants the provider access, because those accounts hold the history of the listing, the installs and the events, and a provider that builds in its own accounts takes that history with it when it leaves. The provider then works through the access it is given, at the level the agreed work needs. What each store allows by way of access, and whether an account can be moved, is a question for that store's own documentation; this guide does not describe any store's account roles or transfer rules.

Should one provider run the listing, the install campaigns and the messaging?

The listing, the install campaigns and the messaging are separate jobs, so quote and review each on its own, even where a single provider takes on more than one. The site's guide to in-house and agency work places lifecycle marketing in-house, and an app's messaging programme is that kind of work, running on the business's own users and events, which is why the split in this guide gives a provider only the building and tooling and keeps the decision to send in the business. As general practice, listing research, install campaign set-up and reporting are the pieces where outside depth is easier to justify. Write the split into the scope so each piece has its own deliverables and its own owner in the business.

When can a new provider's results be judged fairly?

Once the first install cohorts have passed the activation window and the retention measure the business set before the work began, with the counts reconciled between the store consoles, the measurement partner and analytics. Retention shows only over time, small cohorts move around from one week to the next, and some causes sit outside the provider, such as a store review, a first session that loses new users or a release that changes an event. Write down in advance what the cohorts must show before spend is raised, and hold the review then rather than on early install costs.

Terms used on this page

Tools to try

Frameworks

Sources

  1. App Store — Apple Developer
  2. Google Play Console Help — Google
  3. App Store Review Guidelines — Apple Developer
  4. SKAdNetwork — Apple Developer

Related