Automation for agencies: the client-multiplied problem

An agency's automation problem has a shape nobody else's does: the same process, run for thirty clients, each with their own tools, their own naming, their own approval rules, and their own opinion about what a report should contain.

That multiplication changes everything. A workflow that is trivially worth building once becomes either a substantial asset or an unmaintainable pile of near-identical copies, depending on decisions you make early. This page is about those decisions, and about the three processes where agencies get the most back.

What we'll cover

• One workflow, thirty clients
• Client onboarding and offboarding
• Reporting, which is most of the problem
• Working in the client's tools, not yours
• Credentials and access, handled properly
• Whether to sell automation as a service
• Frequently asked questions

One workflow, thirty clients

The first architectural decision, and the one that determines whether this scales.

The copy-per-client approach. Build it once, duplicate it for each client, adjust the details. Fast to start and it is how nearly every agency begins. The cost arrives at client eleven, when an improvement has to be applied eleven times and two copies get missed.

The parameterized approach. One workflow, with the client-specific details held as configuration: which account, which folder, which recipients, which thresholds. Improving it improves it everywhere. Slower to build the first time and it is the difference between an asset and a liability.

What to hold as configuration: account identifiers, folder locations, recipient lists, branding, thresholds, schedule, and anything the client has an opinion about. Keep it in one place — a sheet, a table, a configuration store — rather than scattered through the workflow.

The awkward middle: clients who genuinely differ in process rather than in detail. One wants approval before anything goes out; another does not. Resist encoding both paths into one workflow, because after four such clients it is unreadable. Two or three variants, each parameterized, is more maintainable than one workflow with eleven conditions.

The test to apply before duplicating anything: is this client different, or is this client's data different? Different data is configuration. Different process is a variant, and you should have few of them.

Client onboarding and offboarding

The most consistently underestimated agency process, and the one where consistency matters more than time saved.

Onboarding is folder structure, project record, access requests, kickoff scheduling, brief template, introductions, and a checklist somebody works through. Done by hand it takes an hour and it is done slightly differently each time, which is the actual problem: inconsistent setup is why nobody can find anything eighteen months later.

The parts worth automating are the deterministic ones. Create the structure, create the records, send the requests, schedule the kickoff, assign the owner. Leave the relationship parts to people, because a templated welcome from an agency is worse than no welcome.

Offboarding is the one nobody builds, and it is where the risk sits. When a client leaves, somebody should revoke access, archive the work, cancel the scheduled reports, remove them from recurring automations, and return or delete what needs returning. Done by memory, this is done partially, and an agency that still has credentials to a former client's advertising account eighteen months later has a problem it does not know about.

Build offboarding at the same time as onboarding. It is the same list in reverse, it takes an extra hour, and it is the single most valuable governance step available to a small agency.

Reporting, which is most of the problem

For most agencies this is where the hours actually go, and where automation has the clearest return.

The shape of the work: pull numbers from three or four platforms, put them in a template, write two paragraphs of interpretation, send. Per client. Monthly, or worse, weekly.

The pulling and formatting automates completely. That is the bulk of the time and none of the value.

The interpretation does not, and should not. What the numbers mean for that client, in their context, is the thing they are paying for. An agency that automates the commentary is automating its own differentiation away, and clients notice faster than people expect.

The arrangement that works: the report assembles itself with the numbers, the charts and the comparisons in place, and lands in draft waiting for the account manager to add the interpretation. Twenty minutes of assembly becomes two minutes of reading and five minutes of writing the part that matters.

Two details worth getting right. Flag the anomalies automatically, so the person writing the commentary knows where to look rather than scanning everything. And handle the missing-data case explicitly, because a platform that failed to return figures should produce a report that says so rather than a report showing zero.

Describing this is considerably faster than assembling it from connectors, particularly when every client's report differs in its details. On CodeWords you describe what to pull, from where, how to present it, and what counts as worth flagging, and Cody, the automation builder, builds it, connects it to the platforms you use, and deploys it. Automations connect to more than 3,000 integrations. The free plan covers light use, with Pro at $39 per month and Business at $100 per month as usage grows; details are on the pricing page.

Working in the client's tools, not yours

The constraint that makes agency automation harder than it looks.

Your client uses a CRM you would not have chosen, a project tool you dislike, and a reporting platform that was current in 2019. You do not get to standardize, and telling a client to change tools so your automation works is a conversation that rarely goes well.

Design for the boundary. Keep the workflow's logic in your own environment and treat each client's tools as sources and destinations. Then supporting a new client's CRM means adding one connection rather than rebuilding the process.

Prefer exports where APIs are awkward. Many client systems will send a scheduled file even when their API is restricted or the client will not grant you access. Picking that up is less elegant and considerably more reliable than negotiating API credentials with a client's IT department.

Expect the client to change tools without telling you. They will migrate their CRM and mention it afterwards. Build so the failure is loud and the fix is one connection, rather than a silent gap in a report.

Keep a record of which client uses what. Obvious, universally skipped, and the thing that makes a new account manager useful in a day rather than a fortnight.

Credentials and access, handled properly

The part that matters most and gets the least attention, because agencies hold access to other people's businesses.

Never on personal accounts. An automation running on an account manager's personal login stops when they leave and, worse, continues working when it should not. Shared service accounts, managed centrally.

One credential per client, scoped narrowly. The temptation is one admin account that reaches everything. The consequence is that a single compromise reaches every client you have.

Record what you hold. Which client, which system, what level of access, who granted it, when. This list is what lets you answer a client's security questionnaire honestly, and agencies are asked those more often each year.

Revoke on offboarding, and verify. Not "we removed access" but somebody confirmed it. This is the step that gets skipped.

Have an answer ready. Clients increasingly ask how you store their credentials and who can reach them. An agency with a clear answer wins work from agencies that hesitate.

Whether to sell automation as a service

The question most agencies reach eventually, and the answer is less obvious than it looks.

The case for it: you are already building these workflows internally, clients see the results, and there is genuine demand. It is recurring revenue attached to work you understand.

The case against, which is rarely made: you are taking on the maintenance. An automation sold to a client is an automation you support when their CRM changes, when a connector breaks, when their process shifts. Priced as a project, that becomes a liability the moment the project ends.

If you do it, price the maintenance separately and honestly. A build fee plus a retainer, with the retainer covering exactly the ongoing attention the thing will actually need. Agencies that price only the build discover the problem in month four.

Be clear about who owns it. If the client's automation runs on your platform account, they cannot leave with it. That may be fine, and it should be stated rather than discovered.

The lower-risk version: teach them. Build it in their environment, hand it over, and sell the initial work rather than an indefinite dependency. Less revenue per client and considerably less tail risk.

What the first quarter looks like

A realistic order for an agency of five to twenty people.

Weeks one and two: the client configuration register. One table holding, per client, which tools they use, which accounts, which folders, which recipients, which reporting cadence, and who owns the relationship. This is not automation and everything else depends on it. Most agencies discover during this exercise that two people held different versions of the truth.

Weeks three to five: reporting assembly for your three most similar clients. Not all of them. Three similar ones, parameterized from the register, landing in draft for the account manager to write the commentary. Prove the pattern before spreading it.

Week six: offboarding. Before onboarding, deliberately, because it is the one carrying risk and the one nobody builds. Revoke, archive, cancel, remove, verify.

Weeks seven to nine: onboarding. Now that the register exists and offboarding proves the reverse works.

Week ten onward: extend reporting to the remaining clients, adding a variant only where a client's process genuinely differs rather than where their data does.

The order matters more than the speed. Agencies that start by automating reporting for all thirty clients end up with thirty copies and no register, which is the position this page exists to help avoid.

Frequently asked questions

What should an agency automate first?

Client reporting assembly, in nearly every case. It is the largest recurring time cost, the value is in the commentary rather than the compilation, and the payback is visible within the first month.

Should we build one workflow per client or one parameterized workflow?

Parameterized, with client details held as configuration, once you have more than about five clients. Copy-per-client is faster to start and becomes unmaintainable at exactly the point your agency is growing.

How do we handle clients with completely different tools?

Keep the logic in your environment and treat their tools as sources and destinations at the boundary. Supporting a new client's stack then means adding a connection rather than rebuilding a process.

Is it safe to hold client credentials?

It is normal, and it needs to be deliberate: service accounts rather than personal ones, scoped narrowly per client, recorded in a register, and revoked on offboarding with verification. Clients increasingly ask, and having a clear answer is commercially useful.

Can we resell automation to clients?

Yes, and price the maintenance rather than only the build, because the ongoing support is the real commitment. Consider handing ownership to the client instead, which earns less and carries far less tail risk.

What happens when a client leaves?

Whatever your offboarding automation does, which is why it should exist. Revoke access, archive the work, cancel the scheduled reports, remove them from recurring automations, and verify. Built by memory, this gets done partially.

How much should an agency spend on automation tooling?

Less than people assume, and per-seat pricing is the thing to avoid, because agency headcount changes and every new account manager should not carry a licence cost. Usage-based pricing suits the shape of agency work considerably better.

Related reading

• Automated report generation: which reports are worth it
• Automation tools for small business: the processes worth the effort
• Automated report generation workflow: how to build one
• CodeWords integrations and templates