Marketing and sales as a controlled workflow

You connect positioning, personalization, and experiments to consent, evidence, and brand consistency.

After this chapterYou can design a commercial AI workflow that is relevant and measurable without misleading automation.

Your progress0 of 48 lessons

5.1

Positioning and value proposition

Connect the target audience, problem, alternative, evidence and differentiation.

Use customer evidence and market context to choose a specific audience and outcome. Separate features, benefits and evidence, and avoid unsupported claims such as faster or unique.

Test relevance, credibility, and distinction. Keep variants as hypotheses; do not publish absolute claims that the evidence does not support.

Target group
Problem
Alternative
Outcome
Differentiation
Proof

How you can use this

An accounting tool positions itself around less manual checking for a clearly defined type of case, rather than AI for everyone.

Try this prompt

Create three positioning options for [offer] with target audience, problem, alternative, outcome, differentiation, and required evidence. Mark unproven claims.

Your practical exercise

Create three fictional positioning options. Extension: test understanding with five real target users under suitable arrangements. Basic solo route: examine each option against three questions written down in advance: for whom, which problem, what evidence? Flag unclear words and unsupported promises; correct them and save the change. Completion check: the target group and intended outcome are explicit and the limits of the evidence are visible. This is an editorial self-check, not evidence that real customers understand or want the offer.

5.2

Designing a content operation

Manage content as a chain of briefing, source, production, review, publication, and learning.

For each channel, define the audience, goal, source basis, brand rules, format and owner. Use the content calendar to plan capacity and supporting evidence, not merely to list ideas.

Build gates for facts, rights, privacy, and tone. Keep source links and approved assets; measure behavior that fits the goal, not just reach.

Briefing
Sources
Brand rules
Review
Publication
Learning

Terms in plain language

CTA: The requested next action, such as booking an appointment.

How you can use this

A quarterly series links each item to a single customer query, source of evidence, CTA, and measurement goal.

Try this prompt

Design a content workflow for [channels] with intake, source requirements, production, four review gates, publication owner, reuse, and measurement loop.

Your practical exercise

Build one week of content from one reliable core source and document reuse.

5.3

Sales emails, quotes, scripts, and follow-up flows

Personalize based on relevant, permitted facts and explicitly verify commercial decisions.

Use the known problem, stage, agreement and selected offer. Avoid sensitive inferences or invented personal details. Obtain prices, terms and guarantees from approved sources.

Separate drafting, review and sending. Define frequency, opt-out and escalation, and check that personalisation is accurate and non-discriminatory.

Permitted facts
Phase
Approved offer
Review
Approval before sending
Opt-out

How you can use this

A follow-up email refers to a confirmed demo request, not to derived sensitive characteristics.

Try this prompt

Create for [lead phase] an email, call script, and quote intro. Use only [allowed fields]. Mark missing facts; do not change the price and send nothing.

Your practical exercise

Draw the draft, review and sending steps for three entirely fictional records: a product question with a matching fact sheet, a record without question text, and a question about LAMP-9 incorrectly linked to FICHE-2. Use text on paper; no CRM or connection is needed. Completion check: the first draft is reviewed, the missing field is requested and the incorrect link is blocked. Send nothing and do not record a simulated step as a real system test.

5.4

Qualifying leads without false certainty

A qualification score supports prioritisation; it is not an established fact about a person or organisation.

Define observable criteria such as explicit need, timing, decision-making authority and fit. Separate firm exclusions from softer signals and do not use a risky proxy without necessity.

Show the source, date and uncertainty for each criterion. Allow salespeople to correct assessments and use recurring discrepancies as feedback on the process.

Also establish what the score is used for. Ranking sales opportunities is not automatically the same as assessing individuals’ creditworthiness or selecting job applicants; those latter applications are explicitly listed in Annex III of the AI Act. Check the specific function and applicable conditions, as in lesson 2.1. The absence of a high-risk label does not exempt an application from privacy or discrimination rules.

Criterion
Source
Date
Uncertainty
Override
Bias control

Terms in plain language

Override: A reasoned human correction of a system decision.
Bias: A systematic distortion, for example caused by unsuitable data or criteria.

How you can use this

A lead receives timing unknown instead of low intent based on job title.

Try this prompt

Design a lead rubric for [offer] with criteria, evidence fields, unknown status, exclusions, override, and bias control. Avoid demographic proxies.

Knowledge check

A lead scores 82 out of 100 in a fictional internal rubric. Timing is unknown and the score has not been validated as a probability model. What may the salesperson infer?

An estimated 82% purchase probability, with timing still unknown.A rubric priority; neither a proven purchase probability nor confirmed timing.Strong willingness to buy, with the overall score also indicating favourable timing.

Your practical exercise

Create twenty short fictional lead cases using only observable information relevant to the task; set your rubric in advance. With two reviewers, have them score separately and compare their reasons. Solo, score the cases yourself, put those scores away and repeat in a separate round. Completion check: every score has a supporting extract or ‘unknown’; you investigate differences without automatically calling them errors. Solo, you measure your own consistency, not agreement between two people.

Source for this lesson

AI Act Service Desk – Annex III
Listed areas of application, including recruitment/selection and individuals’ creditworthiness; precise classification also requires the conditions in Article 6.
Checked: 2026-09-08

5.5

Measuring commercial experiments

Change one hypothesis and establish success and stop rules before the campaign.

Formulate target audience, channel, message, action, and primary metric. Choose a comparison that limits selection and channel effects.

In addition to conversion, also consider quality, complaints, unsubscribes, and capacity. Stop at damage or compliance signals, even when revenue seems positive.

Creating two text variants is not yet an A/B test. To measure an effect, assign comparable participants randomly to the variants where possible and keep other conditions the same. Establish the duration, metric and decision rule in advance. With few observations, a large percentage difference can occur by chance. Do not call a small practice result a proven winner.

Hypothesis
Comparison
Primary metric
Guardrails
Duration
Stop rule

Terms in plain language

Metric: A measurable indicator with a clear definition.
Guardrail: A measurement or stopping threshold intended to limit unwanted consequences.

How you can use this

An email test measures appointments, spam complaints, and unqualified leads.

Try this prompt

Design an experiment for [hypothesis] with segment, variable, comparison, primary metric, guardrails, duration, and decision rule.

Your practical exercise

Before seeing the results, establish the criteria the trial must meet to scale up or stop.

5.6

Brand, rights, and transparency

Commercial speed does not replace source, rights, and transparency checks.

Check whether text, image, logo, testimonial, and dataset may be used and keep permission or license. Make AI use visible where rules, channel, or trust require it.

Make claims traceable and have a human approve the publication. Do not publish an asset with unknown rights until verification or replacement is complete.

Usage rights
Permission
Proof of claim
Transparency
Approver

How you can use this

A campaign stores license and consent status alongside each asset and blocks unknown rights.

Try this prompt

Create a pre-publication check for [campaign] with asset rights, personal and brand rights, claim evidence, transparency, channel rules, and approver.

Your practical exercise

Check ten of your own or fictional publication items, such as texts or images, for rights and evidence. Remedy shortcomings actually found; if you find none, state what you checked. Keep the claim check and experiment card with row H5 of the pilot worksheet. State whether this is a side assignment or part of the selected pilot.

Source for this lesson

European Commission – AI transparency
Conditions of application and exceptions under Article 50.
Checked: 2026-09-07

Worked example

A better trial starts with an honest claim

Fictional practice material; incorrect answers have been created deliberately for this exercise.

Atelier Noor practises with two fictional emails; nothing is sent. All recipients, responses and product data are invented. In real use, the audience, basis for contact and any consent requirements must be appropriately assessed in advance.

Input

Goal: more substantive responses within seven days to a product information message. Version A opens with “View our product information”; B with “Which lamp suits your space?”. Everything else—content, sender and sending time—stays the same. Twenty practice recipients are randomly assigned A and twenty B. A receives two responses and B four. The practice sheet supports the dimensions and dimmability of LAMP-2 but contains no comparative energy study, customer testimonial or product certificate.

Deliberately flawed practice answer

“B is proven to work twice as well: roll it out immediately. Add ‘the most energy-efficient lamp on the market’ and an AI testimonial from a satisfied customer. AI text is free of rights restrictions.”

Check

A has 2/20 = 10% responses and B has 4/20 = 20%. The observed difference is ten percentage points; in relative terms it is 100% higher. This describes the small practice sample and does not establish a reliable effect for future recipients. Responses are not purchases, either. The superlative and testimonial lack evidence; presenting an invented testimonial as real is misleading. Using AI does not remove the need to check third-party rights and terms.

Improved result

“B is a candidate for further investigation. Before the next trial, establish the expected improvement, required size, response definition, duration and stopping rules. Also monitor complaints and opt-outs. Do not change the success metric afterwards to obtain a winner. Use only verifiable product claims, such as dimensions from the approved sheet. Have the responsible person assess content, rights and applicable transparency requirements before anything is actually published.”

Try it yourself

B’s four responses consist of two product questions and two complaints. A receives two product questions. The success metric chosen in advance was “substantive product questions”; complaints are a separate negative metric. What do you report?

View the model answer

Both versions have 2/20 = 10% product questions. B also has 2/20 = 10% complaints and A has none in this exercise. There is no observed advantage for B on the agreed success metric. Do not hide the complaints or draw a general conclusion from these small numbers.

Chapter assignment

Bring everything together

Design a commercial workflow from positioning to follow-up, with allowed data, review, experiment metric, and rights register.

My notes

Maximum 10,000 characters per note.

Progress and notes are stored only in this browser on this device. Do not enter sensitive data. Download your notes regularly. This course sets no automatic expiry date. You can delete the data through your browser’s site-data settings; export anything you wish to keep first. Browser settings or cleanup may erase it earlier. These local notes are not sent to Finaudax.