ExecutiveNavigants: questions and answers
The questions a careful buyer asks, answered plainly: how the workflows control quality, what happens to your documents, how the numbers are checked, what it costs and what you get at the end.
Start here
What is ExecutiveNavigants?
ExecutiveNavigants is an AI application for executive and board work. It brings together more than 100 expert workflows, model selection, document handling and usage controls. We call that an executive AI harness: the system that makes the models useful for your work.
Choose a task and the Navigant guides you through it. The method and instructions are built in. Advanced tools add dedicated calculation engines and produce files such as workbooks, presentations and dashboards. You supply the business knowledge and make the decisions.
Why not just use ChatGPT, Copilot or Gemini?
General-purpose assistants are useful, and skilled users can do substantial work with them. The question is how much you want to discover, assemble and check yourself.
ExecutiveNavigants gives you a catalogue of named business tasks, researched methods and guided workflows. Model selection is handled for each tool. Advanced tools add calculation engines, purpose-built outputs and, where relevant, identified rule packs.
The benefit is having that system ready to use without designing the process yourself. It is built around executive work and South African business needs.
Do I need to be good at AI or prompting?
No, and that is the point. The prompting skill, the method selection, the quality checks and the shape of the output are already encoded in each workflow. You answer questions in plain English.
There is no course to sit through and no technique to learn. If you are already a strong AI user you still gain, because the workflow carries the research, the framework selection and the verification that you would otherwise have to supply yourself, every single time.
How is this different from prompts I could copy off TikTok?
It is "just prompts" the way a restaurant is "just recipes."
A prompt is an instruction. A Navigant is an instruction plus the research behind it, the judgment about which framework fits, the testing against real runs, the deterministic compute, the rendered artefact and the economics. Each one took an expert half a day to a full day to author, test and refine.
Half of that list is not made of words at all: verified maths, rendered artefacts, model pinning, spend control, governance guardrails. A copied prompt is a first draft every time, and it rots silently when the models change underneath it. Ours get re-tested.
How the thinking is controlled
Why do you say AI tools chat but don't think?
A language model predicts the next piece of text. Ask it a business question cold and it produces something that reads like an answer, because fluent text is what it is optimised for. It is not running a method, weighing alternatives against each other, or checking its own arithmetic, unless something makes it.
That is what a Navigant adds. The thinking sits in the structure around the model: the sequence of steps, the questions each step forces you to answer, the frameworks it is told to consider and reject, and the checks that run before you see anything. The model supplies language and judgment. The process supplies rigour.
Long AI chats seem to get worse. What is going on?
Every conversation runs inside a finite working memory called the context window. As it fills, quality quietly degrades: the model loses the middle of the chat, blends earlier facts together and drifts from its instructions. Nothing warns you, and the answers still read well. In training senior executives in person, we have yet to meet one who knew this failure mode existed.
The system carries it for you in four ways:
- Navigants are tight, staged processes that reach the deliverable in a fraction of the conversation a meandering chat would need.
- A live meter in every chat shows how much of the window is used, with the real figure for the model you are on.
- Uploaded documents are read once and their facts extracted, so the bulky original never sits in memory.
- Large uploads are size-checked against the model's context window before they are accepted.
Anyone can name Porter or MECE. What is actually hard?
Having frameworks is table stakes. The rare skill is fit: knowing when Porter helps, when CAGE helps, and when neither does. That judgment is what you pay a senior consultant for, and it is the part a prompt cannot carry, because it is not in the words.
Each Navigant has that selection logic decided in advance by someone who has made the call many times before. The workflow considers the methods an expert would consider, discards the ones that do not suit your situation, and tells you why. You get the reasoning, not just a framework name.
Can I trust the numbers? AI makes them up.
A language model does not calculate. It generates numbers token by token, the same way it generates prose, and the errors read credibly enough to sail into a board pack unchallenged.
The advanced Navigants split the work. The AI does judgment and language; a deterministic engine does every calculation. Its results are verified two ways: tie-outs, where totals must reconcile against their components, and golden tests, where known inputs must reproduce known outputs before the tool ships. Charts and tables render from the engine's figures, never from the model's.
Confident numbers are easy. Checked numbers are the product.
How do you reduce hallucinations and stop it finishing early?
Several controls, none of which relies on the model behaving well.
Each workflow moves through defined stages with a gate at each one, so it cannot jump to a conclusion. Source-discipline rules require the output to rest on what you supplied or on the workflow's own reference material, rather than on the model's recollection. Where an answer carries consequence, the figures come from a deterministic engine rather than from the model. Anything the workflow could not establish is reported as not found instead of quietly filled in.
None of this makes a language model infallible. It removes the situations where it is most likely to be wrong and least likely to say so.
South African rules and outputs
How current is your knowledge of the B-BBEE codes, tender rules and King IV?
A model's grasp of the B-BBEE codes or the preference-point formulas is a frozen snapshot of its training data, with no signal telling you how stale it is.
So the rules that carry consequence are lifted out of the prompt into versioned rule packs. Our B-BBEE scorecard computes against a named pack, generic-amended-2019.1, checked line by line against the primary gazettes, stamped with an effective date, golden-tested, and printed on every output together with the gazettes it was built from. When the codes are amended, that ships as a new pack version rather than as an edit to the old one.
We are equally deliberate about what is not yet at that standard. Packs that have been researched but not yet checked against a primary gazette render in red on the output and state that they are not valid for verification. You can see which of the system's own facts are provisional.
What do I actually get at the end, and in what format?
A finished document, not a chat transcript. Depending on the workflow that is a structured report, a board-ready pack, a plan, an options analysis, a scorecard, or an interactive artefact you can click through such as a dashboard or a project schedule. Several of the advanced tools also build a working Excel model.
Anything can be downloaded as a PDF, and the text copies straight into your own templates. You own the outputs.
What happens when I attach a 300-page board pack?
The document is parsed and checked for readable content. The checks account for pages and flag material that could not be extracted, so missing content is not silently treated as reviewed.
The advanced board-pack evaluator works through a long pack in sections and records findings for the review. That avoids depending on the whole pack fitting in the model's working memory at once.
Your content does not linger. Your conversation is never written to our database. An uploaded file is converted to text, read once, and dropped from the conversation. Where a workflow works through a long document in sections, that working text is held only for that job: it is deleted when you start your next chat, and within 72 hours at the latest. We never keep your original file.
Trust, data and control
Is it safe to upload my data here?
There are levels to data security, and ExecutiveNavigants is designed to be safer than most ways of using AI.
Level 1: Cloud storage. Most organisations already store confidential documents in OneDrive or Google Drive. These services encrypt the files, but the documents remain stored, connected to an identifiable account, and accessible until somebody deletes them.
Level 2: Consumer AI. When you attach a document to ChatGPT or Claude, the actual file is uploaded into the chatbot and remains associated with that conversation according to the product’s retention settings.
Level 3: Commercial and enterprise AI. ChatGPT Business, Claude for Work, Copilot and commercial model APIs generally provide better data controls and do not train their models on business data by default. However, “commercial,” “Copilot” or “Azure” does not automatically mean zero retention, private networking or complete tenant isolation.
Level 4: Specially secured enterprise deployments. Azure and similar platforms can provide private endpoints, regional processing, tenant isolation and additional governance controls—but only when those controls have been deliberately purchased, configured and administered. Simply saying that a model runs “on Azure” does not prove that these protections are enabled.
ExecutiveNavigants takes a different approach. You upload the file to us only long enough for us to extract its raw text in memory. We do not store the original file, and we do not upload that file to ChatGPT, Claude or another model platform.
Instead, we pass only the extracted text, anonymously, through commercial APIs in a single chat round. Every model request is restricted to a verified Zero Data Retention endpoint, which means neither OpenRouter nor the model provider retains the prompt or response or uses it to train a model.
Once the model has returned what the workflow needs, we immediately drop everything else from the chat—except long documents, which are cleaned up when you start the next chat, with a 72-hour sweeper for any stragglers. Your conversation is never written to our database.
Removing personal identifiers, client names and unnecessary confidential details remains excellent practice wherever you use AI. But where the information is genuinely needed, ExecutiveNavigants gives you an unusually low-retention way to process it.
Can we use confidential documents and sensitive internal information?
Yes, and you can prove the mechanism to yourself in about a minute.
Test it. Attach a file and ask a question about it in the same message. You get an answer. Now ask about the same document in your next message: on our standard workflows the model will tell you it can no longer see it.
That is one-shot ingestion. The document is read once, the useful facts are extracted into the answer, and the original is dropped from the conversation rather than carried forward. A few workflows that genuinely need a document across several steps have this turned on deliberately, and they say so.
What we store, precisely. Your conversation is never written to our database; there is no message table in it. The original file is never stored at all. Where a long workflow works through a document in sections, that working text is held for that job only: it is deleted when you start your next chat, and within 72 hours at the latest.
We reach the models through their commercial APIs, not through consumer chat products.
What do you store, and for how long?
Very little, and we can be specific.
Your account. An email address, and a mobile number for verification. That is what is needed to run the account and bill it.
Your conversations. Not stored. There is no messages table in our database. Chat history lives in your browser for the session.
Your documents. The original file is never stored. Where a multi-step workflow needs the converted text across its stages, that text is held for that job only, deleted when you start your next chat, else within 72 hours at the latest if you do not activate the deletion.
Usage records. Session metadata, which model ran, and what it cost, so that billing is accurate and you can see your own spend. That record holds no content from your work.
Do you train models on our data, and what do the model providers see?
We do not train any model on your data, and we do not use your content to build anything.
What leaves our server on each request is the message content for that turn and nothing else. The interface machinery, attachment bookkeeping and rendering data are stripped out before the request is sent.
Beyond that, we will only tell you what is true of our own choices. We reach the models through their commercial APIs, not through consumer chat products, which is a materially different arrangement from a staff member pasting a document into a public chatbot.
Is this an agent? Can it reach our email or our systems?
No. Navigants follow defined workflows and ask for your judgement at important stages. They do not independently decide to send email, change your calendar or operate your internal systems.
You work with material you deliberately supply. Uploading a calendar export, for example, is different from granting access to your live calendar.
That reduces the access you need to grant. The aim is to automate the preparation, calculation and production work while leaving business authority with you.
Are you POPIA compliant? Who owns the outputs? Can you sign an NDA or DPA?
We are built for POPIA. We collect the minimum needed to run your account: an email address, and a mobile number for verification. Your chats, your documents and your clients' data are never stored in our database. Ask us to delete your account and those details go with it.
We describe the controls rather than wave a compliance certificate, because that is the honest position and it is the one you can actually check.
You own your outputs, entirely. We claim no rights over anything the system produces for you. We can operate under an NDA and provide the standard data processing documentation your vendor onboarding requires.
Commercials
How does pricing work?
R300 per user per month for the whole library, with volume discounts stepping from 10% at 100 users to 30% at 500. Each seat carries a monthly allowance of 500,000 tokens, which covers normal working use, and corporate plans include one training session per 100 users.
The pricing chapter on this page has a calculator. Set your headcount and it shows the monthly figure alongside the assumptions behind the savings, each of which you can switch off if it does not apply to you.
Can we get a surprise bill?
No. Spend is capped per user and enforced on our server before every single message, rather than reconciled afterwards. When a user reaches their ceiling the system stops and tells them, instead of continuing and billing you.
Corporate accounts carry two limits: a cap per user, and a pool for the organisation as a whole, both measured per billing cycle. Users are warned as they approach the ceiling, and the message tells them the date it resets.
The model each workflow uses is fixed per tool and stored with it, rather than chosen at request time. Cost is predictable, and nothing quietly routes you onto a more expensive model.
Is a trial or demo available?
There is a recorded demo playing on this page: a real workflow run, replayed end to end, finishing in the interactive artefact it produced. No registration needed.
To use it yourself, register and start working. A new account carries a small free allowance, enough to run a workflow through and see the output for yourself. It is a once-off allowance rather than a monthly one.
After that there are two routes. An individual subscribes for R300 a month. Or your organisation becomes a client, and everyone on your email domain gets access automatically when they log in.
How do I start or cancel a subscription?
To subscribe, keep using the app. When you reach the free allowance a window appears offering the R300 monthly plan, handled by Paystack. Paystack is an African payment provider, so it prices in rands and handles VAT correctly. Your card details go to them and never to us.
To cancel, use the one-click cancel link in the receipt Paystack emails you each month. There is no retention maze and nothing to phone about.
One thing worth knowing before you cancel: once a subscription lapses the account drops to a zero allowance rather than back to the free trial, because the trial is a once-off. Resubscribing restores access immediately.
For corporate rates, use the mail link in the dropdown menu.
Can it replace consultants or external advisors?
For a good deal of the work, yes, and we would rather be straight about where it cannot.
Common advisory deliverables are exactly what this is built for: strategy packs, options analyses, business plans, scorecards, board papers, tender responses, financial models. Work that a competent consultant produces from a repeatable method is work a Navigant produces in minutes.
What it does not replace is a genuinely bespoke engagement, someone accountable in the room, or the political work of getting a decision made. The sensible pattern is to reinsource the routine majority and spend the consulting budget on the rest.
Can you tailor it to us, and will it fit our existing tool stack?
Tailoring: yes. For corporate clients we adapt workflows to your templates, your terminology and your governance rules, so the output arrives in the form your organisation already uses. If a tool you need does not exist, tell us and we will usually build it. This is our own library and we add to it constantly.
Your stack: there is nothing to integrate. Navigants take no access to Microsoft 365, Google Workspace, Teams or SharePoint, which is deliberate, and it is what keeps the security review short. You work in the app and take the output wherever you need it.
As organisations adopt Microsoft Copilot agents, our workflows can be ported to that platform. This would be a bespoke exercise.
Who built this
Who are you, and why should we trust your judgement?
The team spans four things that rarely sit together: building large businesses in Africa, premium strategy consulting, board-level governance, and deep hands-on generative AI work. The authors chapter above names each of them with their backgrounds.
That mix is the reason the library looks the way it does. The consulting method comes from people who ran it for a living. The governance and South African regulatory depth comes from people who have sat on the boards that answer for it. And a named expert stands behind every workflow, which is the part a screenshot of the prompt text can never steal.