Choosing a SaaS SEO agency, and when not to hire one

What you are actually buying

Agency proposals converge on a similar list: a technical audit, keyword research, a content calendar, some number of articles a month, link building, monthly reporting. Two agencies can deliver that identical list and produce entirely different outcomes, because the list describes activity rather than what the activity leaves behind.

The useful question is what still exists when the engagement ends. If the answer is a pile of articles and a rankings dashboard you no longer have access to, you rented traffic. If the answer is a content architecture your team can operate, briefs they can write against, and a link structure that maintains itself, you bought an asset.

This distinction is not a moral one about good and bad agencies. It is about what you contracted for. Most contracts specify volume of output, so volume of output is what you get.

For what that system should contain before you go looking for someone to build it, see the B2B SEO strategy piece.

Pricing models and what each one quietly incentivises

• Monthly retainer. The standard. Aligns badly by default, because revenue continues whether or not the work compounds. Fix it by specifying outcomes in the contract rather than deliverable counts, and by scheduling a genuine reassessment at six months.
• Per-article pricing. Incentivises article count. You will get articles. You will not get architecture, because architecture is not billable under this model.
• Project-based. Good for defined pieces of work such as a technical rebuild or a cluster build. Poor for the ongoing judgement that organic actually needs.
• Performance-based. Sounds aligned, usually is not. Whatever metric is in the contract gets optimised, including in ways you would not sanction. Rankings on terms nobody searches count as performance if the contract says rankings.

Model	Rewards	Use when
Monthly retainer	Continuity, not compounding	Scope names systems, not deliverable counts
Per article	Article count	You genuinely only need production capacity
Project based	Finishing a defined build	Technical rebuild or a single cluster
Performance based	Whatever metric is in the contract	Rarely. The metric gets gamed

The least bad arrangement for most SaaS companies is a retainer with an explicit systems deliverable written into scope: the content types, the link rules, the brief template, the quality bar. Those are the things your team keeps.

Whatever the model, agree in advance how the work will be judged, because attribution in B2B makes that harder than it sounds.

Questions that separate operators from resellers

Most evaluation questions are answerable from a template. These are not.

Which of our keywords would you deliberately not target, and why?

An operator has opinions about what to skip. A reseller wants the whole list, because the list is the scope.

What does our buying committee look like, and which roles are we not creating content for?

If the answer does not distinguish the engineer from the economic buyer from the security reviewer, they are running a B2C playbook.

What will our team be able to do at the end that they cannot do now?

Silence here is the answer. You are buying output, not capability.

Show me a client where results disappointed, and what you got wrong.

Everyone has these. Being unable to name one means too few programmes, or not being straight with you.

What happens to the content if we churn?

Ask specifically about hosting. Content on an agency-controlled subdomain is leverage against you, not an asset of yours.

• “Which of our keywords would you deliberately not target, and why?” An operator has opinions about what to skip. A reseller wants the whole list, because the list is the scope.
• “What does our buying committee look like, and which roles are we not creating content for?” If the answer does not distinguish between the engineer, the economic buyer, and the security reviewer, they are running a B2C playbook.
• “What will our team be able to do at the end that they cannot do now?” Silence here is the answer.
• “Show me a client where the results were disappointing, and tell me what you got wrong.” Everyone has these. The ones who cannot name one either have not run enough programmes or are not being straight with you.
• “What happens to the content if we churn?” Ask specifically about hosting. Content on an agency-controlled subdomain or platform is leverage against you, not an asset of yours.

What the first ninety days should contain

The proposal stage is where agencies compete on promises. The first ninety days is where you find out what you bought, and it is worth writing down beforehand what you expect to see, because after the fact everything can be explained.

A reasonable first quarter produces artefacts, not rankings. Rankings in ninety days on anything commercially valuable would be surprising, and an agency promising them is either targeting terms nobody searches or is about to do something you would not sanction.

Month	What should exist	What should not be promised
One	A crawl, a keyword map, and a written architecture decision	Published content. Publishing before the map is how sites get cannibalisation
Two	First pages live, briefs with link rules, tracking in place	Ranking movement on anything competitive
Three	A second batch, plus the first measurable impressions	Pipeline. The sales cycle has not run yet

A reasonable baseline for a mid-market B2B SaaS engagement

The most useful question to ask in month one is not about rankings. It is: what did you decide, and why. An agency that produced a keyword list has done research. An agency that decided which pages to build in which order, and can defend the order, has done strategy.

One thing worth insisting on from day one: the briefs. Not the articles, the briefs. A brief that names the target query, the internal links the page must carry and the question it must answer is a system you keep. An article is an asset you keep. Only one of those keeps working after the engagement ends.

The handover test

The single question that predicts how an engagement ends: if this stopped tomorrow, what would you still have.

Some agencies produce assets that live entirely inside their process. The keyword research is in their tool. The briefs are in their template. The reporting is a dashboard you lose access to. Everything worked, and none of it transfers.

Others build the thing in your systems from the start. Documents in your drive, briefs in your CMS, tracking in your analytics, decisions written down where your team can read them. Costs the same and outlasts the relationship.

Where will the keyword map live?

If the answer is a tool you do not have a seat for, you are renting the research. Ask for the export and check it is genuinely readable.

Who writes the briefs, and can we see one?

A real brief names the query, the angle, the internal links and the question the page must answer. A one-line title and a word count is not a brief, and it is the commonest thing sold as one.

What happens to tracking if we part ways?

Analytics, Search Console and any dashboards should be in your accounts with them added, never the reverse. Trivial to set up at the start and painful to unwind later.

Can our team run this without you in a year?

The honest answer is sometimes no, and an agency that says so plainly is more trustworthy than one promising independence it has no structural reason to deliver.

None of this is about distrust. It is that the incentive to build inside your systems is weak unless you ask for it, and asking costs nothing at the proposal stage and a great deal afterwards.

When not to hire an agency at all

Three situations where the answer is no, and one where it is clearly yes.

• Pre product-market fit. If your positioning is still moving, the content will be rewritten before it ranks. Organic compounds, which is only an advantage when the thing compounding stays true.
• No internal owner. An agency without someone internal who can answer product questions, approve technical changes, and make judgement calls produces generic content, because generic content is all that is possible without access.
• Total addressable market under a few thousand companies. There may simply not be enough search demand. Targeted outbound and community will outperform. Check this with real volume data before committing a budget.
• Clearly yes: you have product-market fit, an internal owner, a market large enough to have search demand, and a twelve-month horizon. This is when organic is the cheapest durable acquisition channel available.

What a fair contract looks like

The commercial structure shapes the work more than the statement of work does, because it decides what the agency is rewarded for when a trade-off appears.

Three structures are common and each bends the work differently.

• Deliverable-based. Eight articles a month. Easy to price, easy to audit, and it rewards volume over judgement. The month a page needs rewriting rather than a new page being written, the incentive points the wrong way.
• Retainer for capacity. A number of days a month, spent where they are most useful. Harder to audit and considerably better aligned, provided you trust the judgement you are buying.
• Performance-linked. Attractive in principle and usually a mistake in organic, because the lag is long enough that the measurement period never matches the work period, and both sides end up arguing about attribution rather than doing the work.

The middle one is right for most engagements. The test of whether it is working is simple: in a month where the best use of time was fixing existing pages rather than publishing new ones, did that happen. Under a deliverable contract it almost never does.

One clause worth adding whatever the structure: a written decision log. Not a report of what was done, a record of what was decided and why. It costs nothing, and it is the only artefact that reliably survives a change of agency or a change of your own staff.

The reporting that tells you whether it is working

Most agency reporting is designed to be reassuring rather than informative, and the difference is visible in what it leads with. A report opening with traffic is telling you about activity. A report opening with what changed and why is telling you about the work.

Three things worth insisting appear every month, and three worth actively resisting.

• Insist on: decisions made this month. Which pages, in what order, and the reasoning. This is the only section that tells you whether you are buying judgement or output.
• Insist on: queries gained and lost. Not rankings for tracked terms, which can be chosen flatteringly, but the actual query set the site now appears for.
• Insist on: what did not work. A report with no failures in it over six months is not a report, it is marketing.
• Resist: traffic as the headline. Organic traffic can be grown with content nobody buying from will ever read.
• Resist: keyword rank tables. Twenty green arrows on terms with no commercial intent look excellent and mean nothing.
• Resist: domain authority scores. A third-party metric, not a Google one, and easily moved by links that do nothing for you.

The month-six test is worth applying to any engagement: can you, without help, explain what the agency decided and why. If the answer is no after six months of reports, the reporting is failing regardless of how good it looks.

Agency, fractional operator, or in-house

Three shapes, different economics. An agency gives breadth of skills and a team that survives an individual leaving, at the cost of attention split across clients. A fractional operator gives senior judgement and system design at a fraction of a full-time salary, but is one person with finite hours and no execution capacity. An in-house hire gives full attention and product knowledge, and takes months to become productive.

The combination that works most often for a Series A or B SaaS company: a fractional operator to design the system and the first cluster, then an in-house content person to run it, with freelancers for production capacity. That puts senior judgement where the leverage is, which is architecture, and cheaper hands where the volume is, which is execution.

The pattern that fails most often is an agency retainer with no internal owner, running for two years, producing articles nobody on your side can evaluate.