Generosity as a Service

01 / FROM TILLY'S DESK

Are we paying expensive AI to do inexpensive work?

Dear Future AI Insider,

One thing we keep encountering while building BizWithAI is that the hardest decision is rarely which shiny AI tool to buy.

It's deciding exactly what job needs doing, what accuracy it requires, and what that job should reasonably cost.

We've been mapping follow-up and customer-journey processes. Some steps need careful thinking and a human conversation.

Others are relatively simple.

Classifying a customer reply. Summarizing notes. Identifying a missing field. Deciding which team should receive an enquiry.

This week's AI developments made me look at that distinction again.

More capable AI is becoming cheaper. But the savings only materialize if we're giving the right work to the right tool, and keeping human oversight where it matters.

Let's look at what's happening.

02 / AI TRENDS FOR BUSINESS GROWTH

Three developments that deserve a closer look

1. Smaller AI models are becoming genuine business infrastructure

On 7 October, Anthropic introduced Claude Haiku 5.5, targeting high-volume activities such as classification, summarization and quick database lookups.

Its published API rates for prompts up to 100,000 tokens are:

Anthropic estimates an average cost reduction per task of approximately 75% compared with Haiku 4.5. This is a vendor estimate, not a guaranteed saving for every business.

Why should a business owner care?

Imagine a professional services firm receiving hundreds of enquiries each month.

Not every enquiry needs an expensive reasoning model. Some only need to be categorized, summarized or passed to the appropriate person.

Using an appropriately sized model could reduce operating costs, provided accuracy, confidentiality and human oversight are properly tested.

The overlooked opportunity: The most powerful AI model isn't always the most commercially sensible choice.

Source: Anthropic announcement

2. AI can return a defined business decision instead of a long answer

On 6 October, OpenAI released its Decisions API into public beta.

Unlike a conventional chatbot that generates paragraphs of text, this API can return predefined choices, scores and probability estimates.

The announced price is $0.10 per million input tokens, with no output-token charge for this API.

What does this mean in practice?

A business could evaluate structured decision systems for categorizing incoming enquiries, flagging requests for review or identifying the appropriate internal team.

This does not mean handing consequential business decisions to AI without human review.

The overlooked opportunity: Some businesses are paying for AI to generate explanations when their workflow only needs a reliable category or routing suggestion.

Source: OpenAI developer announcement

3. Yesterday's working AI integration may have an expiry date

OpenAI's deprecation notice dated 1 October identifies three API models scheduled for removal on 1 April 2027:

• gpt-5.3-codex
• gpt-5.4-nano
• gpt-5.1

Imagine that your business has an automated process handling enquiries or customer follow-up.

Everything works today.

But if the underlying model becomes unavailable, the workflow may stop functioning correctly.

Source: OpenAI official deprecations

03 / AI TOOL UNDER THE MICROSCOPE

Claude Haiku 5.5: Where could it earn its place?

Rather than reviewing this AI model based on how impressive its answers sound, let's examine a specific business application.

The job: Processing repetitive customer enquiries, extracting simple information and preparing short summaries for team members.

What could this actually cost?

Suppose a business processes 10,000 short requests monthly.

Each request averages 1,200 input tokens and 200 output tokens.

Illustrative calculation based on published short-prompt rates and identical token volumes. Actual task costs may differ.

That's a substantial difference.

But here's where business judgment becomes important.

A cheaper model might require more corrections. There may be additional integration, platform, maintenance, data-handling and human-review costs.

If a cheaper model creates more work for your staff, the saving can quickly disappear.

We have reviewed published documentation. This is not a claim that BizWithAI has independently benchmarked this model in production.

Source: Anthropic model documentation

04 / THE AI COST WATCH

Where prices are falling, and where the catch is

↓ Major reduction: Claude Haiku 5.5

Published input and output token rates for prompts up to 100,000 tokens are 90% lower than the corresponding Haiku 4.5 rates.

Anthropic estimates an average real-task cost reduction of approximately 75%, partly reflecting differences in tokenization.

↓ Another reduction: Claude Sonnet 5.5 caching

Anthropic reduced cache-read charges from $0.20 to $0.10 per million tokens.

The company estimates that this may lower costs for many agentic tasks by approximately 20%, depending on how much of the workload uses cached content.

This could matter for businesses repeatedly processing similar instructions or documents.

↑ Watch the pricing threshold

Haiku 5.5's input and output rates are five times higher when prompts exceed 100,000 tokens.

This is an existing pricing tier, not a newly announced price increase.

It matters when workflows repeatedly send large documents, extensive customer histories or unnecessary context.

What about new price increases?

We have not verified a separate, material across-the-board AI price increase suitable for reporting in this edition.

Rather than invent a headline, we'll continue monitoring significant changes.

05 / AUTOMATION TIP OF THE WEEK

Separate the easy decisions from the human ones

Here's a practical exercise you can try without immediately investing in another AI tool.

Imagine your business receives enquiries through a website contact form.

Some are from potential clients. Others relate to existing customers, accounts or general questions.

Instead of manually sorting every message, you could explore using AI to suggest the appropriate category.

1. Choose a repetitive process, such as incoming contact-form enquiries.
2. Define categories such as new enquiry, existing client, billing or needs human review.
3. Ask a small AI model or classification tool to suggest the category.
4. Route uncertain, sensitive or unusual cases to a person.
5. Test previously reviewed examples and measure accuracy, errors and actual time saved.

How we could approach this in RoboLead

Contact form submitted → AI-assisted categorization → appropriate task or queue → human review where needed.

The actual implementation depends on the AI integrations and workflow actions available in your account.

That tells you far more about the value of automation than simply knowing that it ran successfully.

06 / ONE LAST THOUGHT

A useful business doesn't need every new AI feature

The most generous thing I can sometimes do in this newsletter is tell you what not to buy.

If a simpler solution already works, keep it.

If a cheaper AI model handles a limited task reliably, test it.

If someone needs empathy, judgment or a real conversation, make sure a person remains involved.

AI should give us more room to do the work that matters.

It shouldn't make our businesses less human.

I'd love to hear from you.

What's one task in your business that feels too small to justify hiring someone, but happens too often to keep doing manually?

Just hit reply. I'm genuinely curious.

Warm regards,
Tilly
BizWithAI

Help, not hype.

A PERSONAL INVITATION

Enjoyed Issue 01? There's more coming.

This is the first edition of Generosity as a Service | AI Insider.

Every week, we'll investigate important AI developments, look beyond the headlines, examine tools and their real business costs, and share practical automation ideas.

If you'd like Issue 02 and future editions delivered to your inbox, please subscribe.

We won't automatically add you to the new weekly newsletter.

Yes, Send Me AI Insider Every Week →

If this isn't relevant to you, there's nothing you need to do. If you'd prefer not to receive BizWithAI marketing emails at all, you can unsubscribe using the link below.

BEFORE YOU GO | THIS WEEK'S SUGGESTION

Are you spending your time on the work that actually grows your business?

We've talked about the real cost of technology this week.

But what about the cost of your own time?

If you're making decisions about where to invest or what to automate, start by understanding the return on the hours you spend running your business.

Our FounderROI tool offers a starting point for that conversation.

Explore FounderROI →

WE CAN ALSO HELP YOU WITH

A useful next step, whenever you're ready

You don't need to buy another tool to start improving your business.

Explore one of our practical assessments, understand the return on your time, or learn with other business owners.

THE FOUR BUSINESS ASSESSMENTS

MORE WAYS TO GROW

Explore the Business Assessments →

Generosity as a Service | AI Insider
Researched for the week of 12 October 2026.
Prices are in USD and may change. Vendor claims, terms and regional availability should be independently checked before making purchasing decisions.

BizWithAI (PTY) LTD
Hamilton Mews
17 Dulcie Close, Lonehill
Sandton, Gauteng, 2197
South Africa

Unsubscribe from marketing emails