If your product makes decisions, your customers will ask who answers for them
Software companies are building AI into products their customers use to decide about people. Enterprise buyers, insurers and regulators now ask the vendor for evidence before they sign, and internal IT runs AI over access and security too.
Customer AI review: Enterprise Bank (Demo)
Sample data
- Do you train on our data?Data processing policyNoDrafted
- Can we override a single decision?Decision logYesDrafted
- Tested for unfair outcomes this year?Nothing on recordNot answeredNeeds your input
- Will you tell us before the model changes?Change policyYes, 30 daysDrafted
For
- Software and SaaS companies
- IT service providers
- CIOs and CISOs
- Product and engineering leads
Where AI decides, and what goes wrong
Decides about: Your customers' customers
AI features that decide for your customers
You become the supplier your customer must review, and cannot answer their questions.
Decides about: Users
Suspending accounts and moderating content
People locked out or silenced with no route to a person.
Decides about: Employees
Granting and removing access
Access changed by an automated rule nobody reviews, in both directions.
Decides about: Systems and the data in them
AI writing and changing code
Unreviewed changes reach production because the assistant was usually right.
An enterprise customer sends its AI review
The procurement question that now arrives before the contract does.
- 1
The questions arrive
The customer asks whether you train on their data, whether they can override a decision, whether you can explain one, whether you test for unfair outcomes, and whether you will warn them before the model changes.
- 2
Answers come from the record
Answers are drafted from your register, decision log, checks and policies, and anything the record cannot answer is flagged rather than guessed.
- 3
A person approves
The person who answers for the product feature approves each answer before it goes.
- 4
The customer can check
Your trust page shows what AIC observed and what you declared, and a sealed pack backs up the answers.
Rules that already apply
A summary, not legal advice about your organisation.
POPIA sections 19 and 21
Section 19 requires appropriate security safeguards. Section 21 requires a written agreement when you process personal information for a customer as their operator.
What it asks of youSecurity you can evidence, and contract terms your customers will ask you to prove.
POPIA section 71
Applies to your customers' automated decisions, and to yours when your product decides about people directly. Read the guide
What it asks of youFeatures that let a person intervene and let the affected person be heard.
EU AI Act
Places obligations on providers of AI systems used in the EU, including transparency when people interact with AI, and full provider duties for high-risk systems from 2 December 2027. Read the guide
What it asks of youDocumentation, logging and human oversight built into the product for customers who use it in the EU.
ISO/IEC 42001 and SOC 2
The management system and assurance standards enterprise buyers most often ask software suppliers about.
What it asks of youEvidence collected once that answers each framework's requirements.
What the platform gives you
Answer reviews from the record
Draft answers from what the record holds, with every gap flagged and every answer approved.
How it worksA trust page your buyers can read
Live from your record, separating what AIC issued, observed and you declared.
How it worksLog decisions from your product
Send decisions to AIC through the API and hold the ones that need a person.
How it worksRead it from your own tools
A read-only API and MCP server for your dashboards and assistants.
How it works
Building AI products for other companies?
Partner with AIC so your customers can show they stay accountable for what your product decides.
Questions
Our customers make the decisions, not us. Why does this matter to us?
Because your customers answer for decisions your product makes for them, and they will ask you how it behaves. A supplier that can answer clearly, with evidence, is easier to buy.
Can the decision log run inside our product?
Yes. Your product can send each consequential decision to AIC with an API key and hold it for a named person to review. Overrides are recorded by the signed-in person who made them.
We already have SOC 2. What does AIC add?
SOC 2 shows your security controls work. It does not show who answers for each decision your AI makes about people. AIC records that, and evidence collected once counts towards both.
Other industries
- HR and recruitmentScreening, ranking and assessing candidates and staff
- Financial servicesCredit, pricing, fraud and collections
- InsurancePricing, underwriting and claims, and insuring AI
- EngineeringDesign checks, inspection, maintenance and plant control
- HealthcareClinical AI. Preparing for deployment in Q2 2027