Case Study: Thai AI Content Engine for a B2B SaaS Startup

TL;DR: Illustrative scenario — a Thai B2B SaaS startup spends 60,000 THB/month on an agency for 8-12 articles that read generic and rank poorly. The scenario replaces this with a content pipeline the startup owns (a 32 GB GPU workstation, run under a retainer) and targets 15-20 bilingual articles a month, each with cited sources, AI disclosure and a team member’s approval before publishing. Every figure is a scenario assumption, not a measured result.

[Updated · 2026-10-14] Removed the claim that this site is a demo of the pipeline, the model names and an unsourced quotation attributed to Google; statements about Google now link to Google’s own pages, and the scenario’s figures are marked as assumptions. A cost comparison was recalculated from the scenario’s own numbers, and a promise about indexing time was removed.

Key facts

• This is an illustrative scenario, not a real client engagement. The startup described is hypothetical.
• Google’s guidance says appropriate use of AI is not against its guidelines, that using AI gives content no special advantage, and that AI-generated content should be manually fact-checked and reviewed before publishing (sources are linked under “Why this matters”).
• Scenario assumption: the agency charges 5,000-8,000 THB per article and delivers 8-12 pieces per month, about 60,000 THB/month.
• Pipeline in the scenario: 32 GB GPU workstation (~120,000 THB) + setup (~50,000 THB) + monthly retainer (~30,000 THB).
• By the scenario’s figures, buying the hardware pays back in 12-18 months compared with renting a cloud GPU; the retainer is half the agency fee, for a target of roughly twice the output.

Why this case study exists

Same transparency: this is not a real engagement. We wrote it because startups considering content AI in 2026 deserve a concrete picture of what the alternative to “pay an agency or DIY ChatGPT” looks like. Once we have real startup engagements with client consent, we will publish those separately.

The scenario (illustrative)

A Thai B2B SaaS startup, about ten people, Series A funded, selling a workflow-automation product to Thai SMEs. Their marketing lead has been pushing content marketing for six months, cycling through two local content agencies. Each agency delivers 8-12 short articles per month at around 60,000 THB. The articles read generic, have no citations, rank for almost nothing, and drop off Google’s first page within weeks as better-sourced competitors displace them. The marketing lead is tired of paying for content the sales team is embarrassed to share.

They investigate DIY with ChatGPT Plus. Their content intern can produce 10-15 articles per month that way — better than the agency on raw volume — but quality is uneven, fact-checking is inconsistent, and there is no schema markup or SEO infrastructure. In the scenario these articles rank no better than the agency’s.

They come to us looking for a middle path: the production rate of AI with the editorial rigor of an agency.

Why this matters for Thai SaaS marketing in 2026

Two things shape this scenario:

1. Google judges content on quality, not on how it was made. Google’s guidance about AI-generated content (February 2023) says appropriate use of AI is not against its guidelines, and that using AI gives content no special advantage: “It’s just content.” What Google treats as spam is producing many pages mainly to manipulate rankings, whether by automation, by people, or both. That is the scaled content abuse policy announced with the March 2024 core update. Google’s guidance on generative AI content adds two pieces of advice: manually fact-check and review AI-generated content before publishing, and consider telling readers how the content was made.

2. AI answers now sit beside ordinary search results. For its own AI Overviews and AI Mode, Google says there are no additional requirements and no special structured data to add; ordinary good practice applies (AI features and your website). We have no data on how other AI answer tools choose their sources.

The scenario below tests one idea: spend the same content budget on a pipeline the startup owns and manages, rather than on a series of one-off articles.

The constraints we would work within

• Cost target: the pipeline + retainer should cost no more than the agency it replaces. A startup cannot swallow a doubling of content spend.
• Quality floor: every article must cite real sources, pass fact-checking against those sources, and clearly disclose AI drafting — or it does not ship.
• Editorial separation: the startup’s marketing team approves or rejects every piece before publication. Nothing goes live on their site without a human review click.
• SEO infrastructure included: proper schema.org (Article, FAQ, BreadcrumbList), hreflang for Thai/English pairs, TL;DR and key-facts blocks. In the scenario the agency provides none of this.
• Brand voice consistency: the pipeline is tuned once to the startup’s voice and maintained by the retainer.

What we would propose

Hardware: a single 32 GB GPU workstation on the startup’s premises (or their cloud VM, for teams without office space). RTX 5090 or equivalent, 64 GB system RAM, 2 TB NVMe. Hardware capex roughly 120,000-180,000 THB. No dedicated GPU is required if they are willing to rent a dedicated cloud GPU for ~10,000 THB/month instead of buying — though owning pays back in 12-18 months.

Software stack:

• A pipeline with these stages: topic scout, research (with the startup’s own RSS sources for their niche), drafting with a brand-voice system prompt, page title and description, fact-check against the sources, quality score, image finder.
• Open-weight language models for drafting and fact-checking.
• A simple review-queue UI so the marketing team can approve/reject each draft before it hits their CMS.
• Auto-publish integration with WordPress, Webflow, Sanity, or custom CMS via the CMS’s API.

Monthly retainer:

• Topic calendar alignment with the startup’s marketing team (30 min / month)
• Pipeline tuning and model updates as new model versions are released
• Performance review: which articles ranked, which didn’t, tune topic selection accordingly
• Handling edge cases the pipeline doesn’t handle well (product launches, founder op-eds)
• Technical support for the CMS integration

Explicit disclosure on the startup’s own site: every AI-drafted article carries a clear notice that AI drafted it and a member of the startup’s team approved it — not hidden, not disguised. Google’s guidance treats this kind of disclosure as useful wherever a reader might ask how the content was created (Creating helpful, reliable, people-first content).

Expected outcomes

Honest framing: these are the scenario’s targets and assumptions, not measured results. They do not come from a real engagement, and your numbers will differ.

• Volume (target): 15-20 articles per month, bilingual Thai + English pairs under shared slugs. About 2x the agency output.
• Quality floor: every article ships with citations, schema markup, TL;DR, key-facts, and optional FAQ. In the scenario the agency’s articles have no citations.
• Search results: no promise. Indexing and ranking are decided by the search engines, and this scenario has no measured data on either.
• Cost position (scenario arithmetic): the year-one total (hardware 120,000-180,000 + setup 50,000 + 12 months of retainer at 30,000 = 530,000-590,000 THB) is about 18-26% below the agency’s 720,000 THB; from year two the cost is half the agency’s, since the hardware is paid off.
• Editorial integrity: the startup’s marketing team spends less time editing bad drafts and more time on strategy — the pipeline’s draft quality is a function of the prompts and sources, not the writer’s attention span.

Common objections

“AI content will hurt our SEO.” It can, if produced badly — no citations, no human review, no disclosure, many pages with nothing new in them. The pipeline proposed here treats sources, review and disclosure as requirements: an article without them is not published. Google’s guidance is that content is judged on quality however it is produced, and that producing many pages without adding value for users may count as scaled content abuse (guidance on generative AI content).

“Our customers will think less of us if they see AI disclosure.” That is a judgement call, and we have no data to settle it. Our view is that being found out later costs a brand more than saying so up front. Google’s guidance says disclosures are useful where a reader might ask how the content was created, and it calls made-up author profiles a form of deception (Creating helpful, reliable, people-first content).

“We don’t have space or IT for a server.” Options: (1) rent a dedicated cloud GPU on Lambda Labs, RunPod, or Vast.ai for the 32 GB workload (~10k THB/month), no on-premise needed; (2) run on the founder’s beefy workstation if you are truly small. Dedicated hardware in the office is ideal but not required.

“We want to control editorial direction tightly.” You already do. Every draft hits a review queue before publication. The pipeline produces; you approve. No article is published without a team member’s click.

Who this pattern fits

• B2B SaaS startups, Series A or later, with existing content-marketing budgets of 40k-150k THB/month
• Thai agencies looking to productise their service offering with AI while keeping human oversight
• Marketplaces and fintech companies needing high-volume educational content in Thai and English
• Anyone running a blog as a top-of-funnel channel where 15-20 quality posts per month is the bottleneck

Does not fit: companies wanting to replace their in-house writers entirely (you still need 1-2 people to shape editorial direction), companies where every article must be hand-crafted (most enterprise thought leadership), or companies unable to commit to 6+ months of content consistency.

How to engage

The starting point is a free 30-minute call about your current content operation and whether a pipeline would fit. Email the editor with a brief outline. We respond within one business day.

Details of the Thai AI Content Pipeline package on our services page, and editorial standards for the governing principles.