Check this page with an assistantOpens a chat asking it to summarise this article and name the evidence behind each claim.
Claude opens with the prompt on your clipboard: Anthropic does not support prefilled prompts on the web, and we would rather copy it than ship a button that drops it.
Two Things Called Standardisation
Standardising how you gather and process information makes delivery faster without changing what you conclude. Standardising the conclusions makes every client receive the same recommendations. Both get described as productising, and only one of them is a good idea.
The confusion is why the industry has a reputation for interchangeable audits. A client can tell when a document was assembled rather than reasoned, usually because it lists forty findings without saying which three matter for them. That is not a symptom of systematisation; it is a symptom of systematising the judgement layer instead of the mechanical one.
Held correctly the two reinforce each other. Every hour removed from data assembly is an hour available for the thinking clients are paying for, and the thinking is where the margin and the retention both come from.
Sorting the Layers
Take any deliverable and separate it into collection, analysis, judgement, and communication. Collection and communication should be almost entirely systematised. Analysis is partly systematisable. Judgement is the service.
| Layer | Systematise? | Why |
|---|---|---|
| Data collection and access | Completely | Identical every time, invisible to the client, pure cost |
| Technical checks | Completely | Deterministic; running them by hand invites omissions |
| Crawl, log, and export processing | Completely | Mechanical transformation, per log file analysis |
| Finding detection | Mostly | Rules catch known patterns; the unusual case still needs eyes |
| Prioritisation | Framework only, never output | Depends on the client's constraints and commercials |
| The recommendation | No | This is what they are buying |
| Report assembly | Completely | Formatting is not thinking and should cost nothing |
| Quality control | As a written standard | Consistency requires named criteria, not general care |
Prioritisation deserves the sharpest line. A framework for how you rank work is worth standardising, because consistency in method is exactly what makes an agency trustworthy. The ranked output must differ per client, because their constraints differ, and an agency producing the same top three for every engagement has automated the part that was the product.
The Deliverable Clients Actually Value
Not the audit. A ranked, owned, dated queue of work with the reasoning attached, small enough that the client can imagine finishing it. Length is treated as evidence of effort inside agencies and as evidence of padding by clients.
The failure mode is well documented and rarely acted on: a comprehensive document arrives, nobody knows where to start, nothing gets implemented, and the engagement is judged on results that were never possible because the recommendations sat in a folder. Most engagements that fail, fail at implementation rather than analysis.
Building delivery around that changes the shape of the product. Fewer findings, ranked by consequence, each with an owner and a check date, and an explicit statement of what was examined and found fine. That last part costs nothing and answers the question clients ask when a report looks short. The structure is the same one behind the 90-minute visibility audit.
Where AI Fits, and Where It Creates Liability
Generation and transformation, yes. Verification and judgement, no. An agency using a model to draft a section is doing what every agency now does; an agency shipping a claim nobody checked is selling something it cannot stand behind.
The risk is specific to this business. A confidently wrong statement in a client deliverable, an outdated policy, a competitor fact that changed, a statistic that never existed, is the kind of error that ends engagements, because the client discovers it in front of their own stakeholders. Fluency without accuracy is exactly the failure mode described in editing AI drafts.
Practically that means a verification step that reaches outside the model, on every factual claim, before anything carries your name. That step is not automatable, it is the reason the client is paying an agency rather than prompting for themselves, and treating it as overhead to be removed is how firms in this category damage themselves fastest.
What Specialisation Actually Buys
The second client in a sector costs far less to serve than the first. You already know the platform's constraints, the competitive set, the seasonality, and the regulatory limits, so the same hours produce better judgement rather than repeated orientation.
That compounding is genuine and it is why specialised firms tend to scale faster. It also has a real cost that is usually left out of the pitch: a narrower addressable market, and correlated risk if the sector contracts. An agency entirely serving one industry has the concentration problem its clients would be advised against.
The useful middle is depth in a small number of related sectors where the platform knowledge transfers, rather than one vertical or all of them. And the depth has to be real: sector knowledge means knowing why the platform behaves as it does, of the kind in Shopify SEO or regulated content, not a landing page claiming experience.
Questions Agency Owners Ask
- What should an agency systematise first?
The parts that are identical across clients and invisible to them: data collection, technical checks, crawl and log processing, report assembly, and quality control. These consume most of the hours, vary least between engagements, and produce no differentiation when done by hand, which makes them the obvious first candidates.
- What should never be templated?
Prioritisation and the recommendation itself. Which of forty findings matter for this business, in what order, given their constraints, is the judgement clients are actually paying for. A templated audit that lists everything and ranks nothing is the specific deliverable clients complain about, and it is what standardising the wrong layer produces.
- Does productising delivery make the work worse?
Only if you standardise the thinking rather than the mechanics. Systematising data collection frees capacity for judgement, which improves the work. Systematising conclusions produces the same recommendations for every client, which is where the reputation for interchangeable audits comes from.
- How do you keep quality consistent as the team grows?
A second person checks each deliverable against a written standard before it ships, and the standard is specific enough to fail against. Reviewing for quality in the abstract produces inconsistent results; reviewing against named criteria produces consistent ones and doubles as training for whoever is being reviewed.
- Should agencies specialise in one vertical?
It compounds faster than generalist delivery, because the second client in a sector costs far less to serve than the first: you already know the platform, the competitors, the seasonality, and the regulatory constraints. The trade is a narrower market and correlated risk if that sector contracts, which is a real cost rather than a technicality.
