AI ServicesZW
  1. Index
  2. Vendor questions
Asset · Q-01 to Q-60 · eight risk areas

The vendor question bank

Take these into the meeting. You will not ask all sixty; you will ask the eight that apply to your risk, and the vendor's answers to those will tell you more than the demo. Codes are stable — cite them in a scope or a contract.

Copy buttons need JavaScript; the text is selectable without it. Printing this page gives a clean checklist.

01

Scope and outcomes

Q-01–Q-08

Most failed AI projects failed at scope. Pin the outcome, the boundary and the measure before the price.

  1. Q-01Which single business outcome does this service change, and how will we measure it on day 30 and day 90?
  2. Q-02What is explicitly out of scope, in writing?
  3. Q-03What must be true on our side (data, staff time, access, devices) for you to deliver — list it.
  4. Q-04Show us one comparable delivery: what was the problem, what was built, what happened after handover?
  5. Q-05What does "done" look like — a document, a demo, a signed acceptance test, or live use by our staff?
  6. Q-06What happens when the AI cannot handle a case — who does it hand to, and how fast?
  7. Q-07If we stopped the project after the first milestone, what would we own and be able to use?
  8. Q-08Which parts of this are configuration of an existing tool, and which parts are custom work?
02

Data and privacy

Q-09–Q-16

Zimbabwe's Cyber and Data Protection Act and SI 155 of 2024 make you, the controller, responsible for what your vendor does.

  1. Q-09Which personal data will the system collect, store or send, and to which countries?
  2. Q-10Are you licensed as a data controller with POTRAZ, or acting as our processor — and which are we?
  3. Q-11Who is your named Data Protection Officer, and who is ours in this contract?
  4. Q-12Which third-party AI models or APIs will see our data, and under which terms?
  5. Q-13Is our data used to train or improve any model — yours or a third party's? Can we opt out?
  6. Q-14How do we delete a customer's data on request, end to end, including backups and the AI provider?
  7. Q-15What is retained after the contract ends, for how long, and who can access it?
  8. Q-16Show us your breach procedure: who is told, within how many hours, and what we must do.
03

Integration and operations

Q-17–Q-24

The AI is rarely the hard part; the connections to WhatsApp, CRM, POS, ERP, EcoCash and fiscal devices are.

  1. Q-17Which of our systems will this connect to, and by which method (official API, export/import, screen automation)?
  2. Q-18For WhatsApp: is this the official WhatsApp Business Platform (Cloud API) or an unofficial connection that Meta may block?
  3. Q-19What happens to messages, orders or documents that arrive while the system is down?
  4. Q-20Who owns the WhatsApp number, phone-number certificate, API keys and admin logins — us or you?
  5. Q-21How are changes deployed — is there a test environment, and do we approve changes before they go live?
  6. Q-22What monitoring exists, and who is alerted when a flow breaks at 22:00 on a Friday?
  7. Q-23How many staff hours per week will running this take on our side after handover?
  8. Q-24If we change CRM or POS in a year, what breaks and what does it cost to re-connect?
04

Language and channel

Q-25–Q-31

Customers write in Shona, Ndebele, English and mixes of all three; on WhatsApp, with voice notes, at any hour.

  1. Q-25Show the system handling a Shona or Ndebele message, a mixed-language message and a voice note — live, not on a slide.
  2. Q-26How does it respond to a customer it does not understand, and in which language?
  3. Q-27Which channels are included (WhatsApp, web chat, SMS, phone, email, Facebook/Instagram) and which cost extra?
  4. Q-28Can staff take over a conversation mid-thread, and does the customer notice?
  5. Q-29How are tone and wording controlled — can we edit responses without a developer?
  6. Q-30What happens with a customer who sends 40 messages in a minute, or an abusive one?
  7. Q-31Can it collect a payment or issue a fiscal invoice inside the conversation, and through whom?
05

Accuracy and human oversight

Q-32–Q-39

A system that is 92% right still gets 8 of every 100 customers, invoices or CVs wrong. Someone must catch them.

  1. Q-32What accuracy do you expect on our data, and how will we measure it on a sample we choose?
  2. Q-33Which decisions will a human always review before they take effect (refunds, rejections, payments, hiring)?
  3. Q-34How does the system show its confidence, and what happens below a threshold?
  4. Q-35Can we see and audit every automated decision afterwards — who, what, when, on which input?
  5. Q-36How do we correct a wrong answer so it stays corrected?
  6. Q-37What tests run before each change goes live, and can we see the results?
  7. Q-38How is the system prevented from inventing facts (prices, stock, policies) it was not given?
  8. Q-39Who is liable when an automated message, quote or document is wrong?
06

Commercials and pricing

Q-40–Q-47

Price models hide risk: per-conversation fees, API pass-through and ZiG/USD terms decide the real cost.

  1. Q-40Break the price into: your labour, licences, third-party API and messaging fees, hosting, and support — line by line.
  2. Q-41Which costs are per month, per conversation, per document, per seat or per user, and who pays them if usage doubles?
  3. Q-42Is the price in USD or ZiG, at which rate, and what happens if the rate moves during the contract?
  4. Q-43What is the minimum term, the notice period and the exit fee?
  5. Q-44What is included in the first 90 days after go-live at no extra cost?
  6. Q-45What is your hourly or daily rate for changes outside scope?
  7. Q-46Which payment milestones are tied to acceptance tests rather than to dates?
  8. Q-47Do you receive commission or reseller margin from any tool you recommend? Disclose it.
07

Hosting, power and continuity

Q-48–Q-53

Load shedding, fibre cuts and forex controls are real operating conditions here. Ask where it runs and what fails.

  1. Q-48Where does the system run — which country, which data centre or cloud region — and can we choose?
  2. Q-49What happens during a power cut or an internet outage at our premises: what queues, what is lost?
  3. Q-50What is the uptime commitment, how is it measured, and what is the remedy when it is missed?
  4. Q-51How are backups taken, how often, where are they kept, and have you tested a restore?
  5. Q-52Can it run on a phone with 2G/3G and a small data bundle for our field staff?
  6. Q-53How do we pay foreign hosting or API bills if forex is constrained — do you invoice locally in full?
08

Exit, ownership and support

Q-54–Q-60

The cheapest vendor is expensive if you cannot leave. Own the assets; document the system.

  1. Q-54Who owns the code, prompts, flows, training data and documentation at the end — and is it in the contract?
  2. Q-55Can we export all data and conversation history in an open format at any time?
  3. Q-56What documentation and training do our staff receive so we can run it without you?
  4. Q-57What are the support hours, response times and channels — and are they in Zimbabwe time?
  5. Q-58How many of your team know this system, and what happens if that person leaves?
  6. Q-59How are model or platform price changes from third parties (Meta, OpenAI, Google) passed on to us?
  7. Q-60Can another vendor take this over — and would you cooperate with a handover?

How to use the answers: the scoring matrix weights each risk area; the pricing models page explains the commercial answers; problem sheets carry the five to eight questions most specific to each problem.