Find out whether your business is ready for AI, before you spend anything
You run a real business. There are records going back years, a way of doing things that works, and a growing pile of people telling you that you need AI. This assessment tells you what in your business would actually have to change first, and whether it is worth doing at all.
It assumes no prior AI work and no technical background. Around ten minutes, no account needed to start, and your email is only required at the end to show you the full result.
It is a conversation, not a form
You answer in your own words, in plain language. Nothing is asked in jargon, and nothing assumes you have done this before. The assessment asks between fifteen and twenty-five questions and skips the ones that do not apply to you.
- The scoring is fixed, not improvised
- Every question maps to a defined point value written into the rubric. The model conducting the conversation cannot change a score, weight an area differently, or invent a category. It rephrases questions and interprets your answers; the arithmetic is not its to do.
- You can say you do not know
- Most people do not know where all their data sits, and that is a normal answer rather than an embarrassing one. Unanswered questions lower the confidence rating attached to your result rather than quietly counting as zero.
- You see the reasoning
- Each area score comes with the specific responses that drove it, so you can disagree with it on the evidence rather than on the tone.
What gets assessed
Six areas, weighted to 100 points. None of them are about how much technology you already have. They are about whether the business underneath it is in a state where a project could succeed.
- Business value 15 points
- Whether there is a specific, measurable problem worth solving, with a named owner.
- Process 15 points
- How well the target workflow is documented, how consistently it runs, and how often it changes.
- Data 25 points
- Where the relevant data lives, what condition it is in, and whether it can be accessed lawfully.
- Technology 15 points
- The current systems, how they integrate, and whether anything can be connected to.
- People 10 points
- Available skills, capacity to run a pilot, and how change has landed previously.
- Governance 20 points
- Privacy handling, security controls, approval paths, and who is accountable for outcomes.
Data and governance together account for 45 of the 100 points. In practice they are where projects stall.
What you get back
A score out of 100, a red/amber/green rating per area, your three strongest areas, your three most likely blockers, and one recommended next step with the reasoning attached.
Confidence high
- Business value11 / 15
- Process11 / 15
- Data13 / 25
- Technology11 / 15
- People6 / 10
- Governance10 / 20
Bar length shows each area’s weight out of 100. Data and governance are weighted highest because they are the areas that most often stop a project outright.
Strongest areas
- Business value73%
- Process73%
- Technology73%
Likely blockers
- Governance50%
- Data52%
- People60%
Recommended next step
Structured extraction from your incoming documents
A meaningful share of the relevant information arrives as documents or scans, while the process around it is well enough defined to check the output against. Extraction is a good first pilot because every result is verifiable, the source document is right there, so errors are caught rather than absorbed.
Indicative effort 4-8 weeks
A recommendation that can be no
If your data is in poor condition, the recommendation is to fix the data, not to start an AI pilot. Several outcomes in the rubric are deliberately not AI projects at all. An assessment that always concludes you are ready is a sales funnel.
A confidence rating
Based on how many questions you answered, not on how certain the model sounds. Answer half of them and the result says so.
Three ways to be assessed
The conversation is where everyone starts, and for many businesses it is enough. The other two exist because a self-assessment records what you believe about your systems, and people tend to overestimate how well their data is organised.
Free
Answer the questions
A guided conversation of roughly twenty questions about your business, your processes and your data. No jargon, and no assumption that you have done any of this before. You get a score against the same published rubric everyone is scored against, and a recommendation that is sometimes not to start an AI project at all.
About fifteen minutes. No account needed.
Start the assessmentIncluded with the report
Upload what you already have
Send us the documents that describe how the business actually runs: a privacy policy, a written procedure, a system list, a vendor contract. Each one is read against the questions it could speak to, and every finding comes back with the sentence it came from so you can check it rather than take our word for it.
PDF, text, Markdown or CSV. Up to twelve documents.
See how it worksIncluded with the report
Give the agent read-only access
Put your exports and records in a data room and let the agent work through them itself. It lists what is there, reads what looks relevant, and profiles your actual tables: how many rows are missing an email, whether the same customer appears under three different identifiers, whether the dates parse. That is the difference between being told your data might be messy and being told which column and how much.
Read-only, with no network access. It cannot change or send anything.
See how it worksWhat giving the agent access means
This is the part worth reading slowly, because you are being asked to put business records somewhere. Here is exactly what happens to them.
- Read-only, enforced in the infrastructure
- Your material is mounted into the agent's workspace in read-only mode. Any attempt to write to it fails at the filesystem, not at a prompt telling the agent to behave. It cannot edit your files, delete them, or add to them.
- No network access at all
- The sandbox the agent's commands run in has network egress switched off, and the tool that would let it make web requests is not included in the build. There is no route by which your data leaves the workspace.
- Walled off from every other customer
- Each assessment gets its own workspace, and the only files it can address are the ones filed under that assessment. The scope is fixed by the workspace's own identity rather than by a value we pass in, so there is no request that can point it somewhere else.
- Every claim carries its source
- A finding is only recorded with the exact text it came from, and that text is checked against the file before the finding is kept. If the agent cannot point at where it read something, the finding is discarded rather than shown to you.
- Nothing changes your score without you
- Findings are proposals. You accept or reject each one, and only the ones you accept move the number. You can see the revised score next to the original and undo any of it.
Make the agent its own account. Never use a staff member's.
Where you are exporting from a system that has logins, create a separate, read-only account for this and use that. Do not hand over the credentials of anyone on your team.
- A staff account carries that person's permissions, which are almost always broader than a scan needs. A dedicated account can be given the minimum.
- You can switch a dedicated account off the moment the scan is done, without disrupting anyone's work or forcing a password change.
- Your own audit logs stay honest. Activity shows up as the agent's account rather than as a colleague apparently reading records at two in the morning.
- If something is ever wrong, the blast radius is one read-only account you created for the purpose.
We never ask for a password on this site and never will. You place the exports in the data room yourself.
Fixing the data
Coming soon
The scan tells you which columns are empty, which identifiers disagree between systems, and which records cannot be joined. The obvious next question is whether we can just fix it, and the answer today is no.
The planned service is the repair work: reconciling identifiers across systems, filling and standardising the fields that matter, and putting the result somewhere a model can actually use. The agent already has the analysis; the missing half is doing something about it with your sign-off on every change.
It does not exist yet, and it is not for sale. Writing to a customer's systems is a different level of trust from reading them, and we would rather build that once, properly, with people who have told us what they need. If this is the part you actually want, say so and it will shape what gets made.
Where your answers go
You are being asked about your data handling, so it would be poor form to be vague about ours.
- The conversation is processed outside South Africa
- Your answers are sent to OpenAI's API in the United States to conduct the conversation and to classify your responses into rubric values. This is cross-border processing of whatever you type. We are telling you plainly rather than burying it, so you can decide what to put in the box.
- Stored in the European Union
- Assessments, answers and scores are stored in a Cloudflare D1 database in the Western Europe region. Generated PDF reports are stored in Cloudflare R2 and are only ever served through a short-lived signed link, never a public URL.
- Do not paste customer records into the chat
- Nothing in the assessment requires them. The questions ask about the shape of your data: where it lives, what condition it is in, not its contents.
- You can have it deleted
- Email us and we will delete your assessment, transcript, score and any generated report. The trust centre sets out what is kept, for how long, and who the subprocessors are.
Pricing
The assessment itself is free. What is worth paying for is the written record of it, and the conversation about what to do next.
For reference: a consultant-led AI readiness diagnostic in South Africa runs to roughly R45 000, and the international equivalent starts around R24 000 and rises well beyond R240 000 for enterprise scope. Those are people in rooms for a week. This is not that, and it is not priced as though it were.
Readiness check
Free
No payment, and no account until the end.
- The full conversational assessment, 15 to 25 questions
- Your overall score out of 100, and the band it falls in
- A red, amber or green rating for each of the six areas
- Your three strongest areas and three most likely blockers
- One recommended next step, with the reasoning behind it
- A confidence rating based on how much you answered
- No written report
- No document review
- No consultation
Readiness report
R4 950
Once-off. The written version you can put in front of a board.
- Everything in the free readiness check
- A written report prepared for your business by name, with a cover, the scorecard, per-area detail and the recommended pilot
- Per-area detail explaining exactly what each score is based on
- The recommended pilot written up with prerequisites and effort
- Upload your own material, policies, process notes, system lists, checked line by line against your answers
- Or give the agent read-only access to a data room and let it profile your actual records
- A revised score showing exactly where the evidence and your answers disagree
- A 45-minute consultation to work through the findings
Data remediation
Indicative from R24 500
Fixing what the scan found, rather than only reporting it.
- Reconciling identifiers that disagree between your systems
- Standardising and filling the fields the intended use actually depends on
- Landing the result somewhere a model can use, with sign-off on every change
Not built yet, and not for sale. Writing to a customer's systems is a different level of trust from reading them, and we would rather build it once, properly, with people who have told us what they need.
Register interestPrices in South African rand, including VAT. Payment is handled by PayFast, which processes in South Africa. Card details are entered on PayFast's own page and are never seen by us.
What a typical roadmap looks like
Every result is different, but the shape tends to be consistent: fix what would undermine a pilot, run one narrow pilot with a measurable outcome, then decide whether to widen it.
- 01First
Remove the blocker
2-8 weeks
Usually one thing: consolidating a dataset, writing down a process that only exists in someone's head, or establishing on what basis you hold the personal information involved. This is not glamorous and it is not optional.
- 02Then
One narrow pilot
4-8 weeks
A single workflow, with a person reviewing the output and a number agreed in advance that tells you whether it worked. Chosen so that failure is visible and cheap rather than quiet and expensive.
- 03Then
Decide on the evidence
2 weeks
Compare the result against the number you agreed. Widen it, adjust it, or stop. Stopping after a pilot that did not work is a good outcome; the alternative is finding out after the rollout.
This is an illustration of sequencing, not a template you receive. Your recommended step is chosen from your own answers.
Questions
Does this tell me whether we are POPIA compliant?
No, and be wary of anything that claims to. The aireadiness assessment is a self-assessment: it records what you tell us and is not verified against your systems. It can identify possible privacy and governance gaps that require verification, which is a useful starting point for a conversation with a lawyer or your Information Officer. It cannot determine compliance, and nothing in your result should be read as saying you are compliant or that you are not.
Who is behind the scoring?
The rubric is a fixed set of 25 questions with fixed point values, held in code. A language model runs the conversation and maps what you type onto one of the defined answers, but it cannot assign points or change a weighting. Two businesses giving the same answers get the same score.
What if I do not know the answer to something?
Say so. Unanswered questions reduce the confidence rating attached to your result rather than counting against you as a zero. For several questions, particularly around privacy, not knowing is itself the finding worth surfacing.
Is my data used to train a model?
No. Your answers are sent to OpenAI's API to conduct the conversation and are stored in our database so you can return to your result. They are not used to train any model, by us or by OpenAI under its commercial API terms.
How long does it take?
About ten minutes if you know your systems. Longer if you want to check something before answering, which is reasonable. You can leave and come back to the same assessment.
Do I have to pay to see my score?
No. The score, the per-area breakdown, your strengths, your blockers and the recommended next step are all free. An email address is required to display the full result. The paid report, at R4 950, is the written version with per-area detail and your own documents taken into account.
Do we need to know anything about AI to take this?
No. The questions are about your business, not about technology: where your information sits, how consistently a task gets done, who signs things off. Nothing is asked in jargon and no prior AI work is assumed. If the honest answer to a question is that nobody has ever thought about it, that is a useful finding rather than a failure.
Who is this for?
Established South African businesses that are being told they should adopt AI and want to know, honestly, whether they can. Most of the businesses this is built for have done no AI work at all: a manufacturer, a logistics operator, an accounting practice, a medical practice, a retailer with twenty years of records. Having real operations and real history is an advantage here, not a gap. It is least useful for very early-stage companies with neither.
Ten minutes, and a number you can argue with
You will get a score, a breakdown of where it came from, and one specific thing to do next. If the honest answer is that you are not ready, it will say that.
Free. No account required to begin.