Almost every company we talk to already uses artificial intelligence. Someone writes posts with ChatGPT, a salesperson summarises calls with Gemini, the marketing manager tries Claude for newsletters. Everyone has their own prompts, nobody knows what the others are doing and nobody measures whether the time saved turns into anything. This guide is about moving from there to a method: a few carefully chosen workflows, shared rules, human review where it is needed and numbers to tell you whether it is worth carrying on.
The 20 workflows in the central chapters are the ones we use, in various forms, on B2B marketing and sales projects. You do not need to adopt them all. Two or three done well in the first month are enough.
Before the tool: map your processes and set priorities
You start from processes, because a tool without a clear process just produces more work to check. The first step is writing down what the team does every week, how long it takes and where it gets stuck; only then do you choose where AI can help.
How to map it in an hour
Get marketing and sales together for an hour. For each recurring activity, fill in a row of the table below. You do not need precision to the minute: the order of magnitude is what counts. Activities that are good candidates for AI have three things in common: they repeat, they start from written material (documents, data, transcripts) and they produce a result a competent person can check in a few minutes.
| Activity | Who does it | Frequency | Hours a month | Inputs available | Where it gets stuck | Risk if wrong (low/medium/high) |
|---|---|---|---|---|---|---|
| Example: monthly campaign report | Marketing manager | Monthly | 8 | Google Ads, GA4, CRM exports | Combining the data and writing the commentary | Medium |
Three questions for each row
Ask three questions about each activity. If the result is wrong, who notices and when? If the answer is “the customer, afterwards”, the activity needs heavy review or should be left out for now. Do the inputs already exist in written, accessible form? If the knowledge only lives in one person’s head, it needs writing down first. How much time does it free up, and to do what? An hour saved that disappears into more emails is not a result; an hour moved onto open deals is.
From map to priorities
Once the map is ready, place each activity on two axes: impact (hours freed up, quality, effect on leads and sales) and effort (data to prepare, integrations, people to involve). Appendix A contains the grid. Top left, high impact and low effort, you will find the first workflows to try. Bottom right, the ones to leave where they are.
Self-assessment question. If you asked three people on the team tomorrow which AI tools they use and with what data, would you get the same answer? If not, the map needs doing before any purchase.
- We have listed the recurring marketing and sales activities
- For each one we know the frequency and hours a month
- We have noted where the inputs are (documents, CRM, platforms)
- We have marked the risk if the result is wrong
- We have chosen no more than three activities to start with
The 5 rules for using AI without creating chaos
Five rules, written down and shared before you start, prevent most problems: misused data, copy that all sounds the same, results nobody checks. They fit on one page and should ideally be signed by whoever leads marketing and sales.
Rule 1: a business account, never a personal one
With free personal accounts, the content you enter may be used to train the models. With the main providers’ business plans the default is different: OpenAI, for example, states that it does not use data from business plans and the API for training. Choose a main provider, switch on the business plan, sign the data processing agreement and ask the team to use only that for work.
Rule 2: your context, always
A model that knows nothing about your company writes average copy. Prepare a “context pack” to load into shared projects: positioning, products, buyer personas, recurring objections, tone of voice with approved examples and words to avoid. Features such as Claude or ChatGPT projects let you keep it always to hand; we talk about this in the article on Claude Cowork.
Rule 3: every output has a named reviewer
For each workflow you decide who reviews, what they check and when the output can go out. In the following chapters every workflow has its own “human review” entry. The principle: the closer the output gets to a customer or a spending decision, the tighter the review.
Rule 4: prompts saved, versioned and shared
A prompt that works gets saved in a shared library, with a date, author and an example of a good result. That way whoever comes next does not start from scratch, and improvements stick. The structure we use (role, context, task, constraints, format) is explained in the article on prompt engineering; Appendix C contains ten ready to adapt.
Rule 5: every workflow has a KPI before it starts
You measure time per activity before and after, the percentage of outputs used without heavy rewriting and a result indicator (qualified leads, response rate, response time). If after four weeks the numbers have not moved, you change the workflow or switch it off.
Watch out. The rule most often broken is the first one. All it takes is a salesperson pasting a proposal with names and prices into a personal chatbot to create a confidentiality problem. Spell it out, with examples.
| Rule | Who enforces it | How it is checked |
|---|---|---|
| Business account | Owner or IT | List of active users on the business plan |
| Shared context | Marketing manager | Context pack up to date in the projects |
| Named reviewer | Owner of each workflow | Workflow card (Appendix B) |
| Prompt library | A designated person | Document with versions and dates |
| Defined KPI | Marketing and sales together | Before/after measurement at 4 weeks |
Workflows for research, strategy and competitor analysis
Here AI speeds up the gathering and organising of information that already exists: competitors’ websites, reviews, call transcripts, market data. The strategic conclusions are still yours; AI brings you to the table with the material in order.
Workflow 1: a competitor profile in 30 minutes
- Goal: a comparable profile of each main competitor covering offer, messaging, public prices and channels.
- Input: website URL, service pages, company LinkedIn page, any public reviews.
- Steps: 1) an assistant with web search (Perplexity, ChatGPT or Gemini with search on) gathers the information with sources; 2) a second prompt arranges it into a fixed template; 3) a person checks the key points by opening the cited sources.
- Output: a one-page profile per competitor, always in the same format.
- Human review: every claim about prices, customers and figures is checked against the source. Anything without a source is deleted.
- KPIs: time per profile; number of profiles updated per quarter.
Workflow 2: voice of the customer from calls and reviews
- Goal: extract customers’ recurring words, problems and objections to use in messaging and content.
- Input: sales call transcripts, enquiry emails, reviews, support tickets (anonymised).
- Steps: 1) remove names and identifying details; 2) ask the AI to group problems, objections and desired outcomes by theme, quoting phrases verbatim; 3) count the occurrences; 4) the team picks the five strongest themes.
- Output: table with theme, frequency and verbatim customer quotes.
- Human review: a salesperson confirms the themes match what they hear on the phone.
- KPIs: number of messages or pages rewritten using the voice of the customer; conversion rate of rewritten pages.
Workflow 3: positioning brief for a new service
- Goal: arrive at the first meeting on a new service with a draft positioning to discuss.
- Input: competitor profiles (Workflow 1), voice of the customer (Workflow 2), the owner’s notes on the service.
- Steps: 1) the AI suggests three alternative positionings with audience, problem, promise and proof; 2) for each one it lists the weak points; 3) the team chooses and refines.
- Output: two-page document with three options and a reasoned recommendation.
- Human review: the owner decides. The AI knows nothing about margins, production capacity or sales history.
- KPIs: time from first idea to approved brief.
Workflow 4: account research before an ABM campaign
- Goal: prepare a summary of each target company that is useful to sales and marketing.
- Input: target company list, website, recent news, published job vacancies, LinkedIn page.
- Steps: 1) research with cited sources; 2) summary of the business, signs of change (new sites, hiring, products), possible needs; 3) two opening angles for the salesperson.
- Output: account profile, linked to the company record in the CRM.
- Human review: the salesperson checks the signals before using them in a message. A wrong fact in the opening line burns the contact.
- KPIs: reply rate to first contacts on accounts with a profile versus those without.
Example. Imagine a components company with 40 target accounts. Without AI, research takes about half an hour per account. With Workflow 4 the salesperson gets the profile ready-made and spends ten minutes checking it. The hours saved go into phone calls. The numbers are an example: measure your own with the table in chapter 1.
Workflows for SEO and content marketing
In content, AI delivers most in the stages around the writing: research, structure, review, reuse. The part that makes the difference, meaning experience, your own data and examples, has to come from you; otherwise the text looks like a thousand others and neither search engines nor AI answers have any reason to pick it.
Workflow 5: from keyword group to editorial brief
- Goal: turn a group of searches into a brief an author can follow.
- Input: keywords with volumes from an SEO tool, Google’s first results pages, customer questions (Workflow 2).
- Steps: 1) the AI classifies keywords by search intent; 2) suggests a structure with headings written as questions; 3) points out what is missing from content already on page one; 4) lists the evidence and data needed from your side.
- Output: brief with intent, structure, questions to cover and internal material required.
- Human review: the content lead checks the brief answers the real intent, not an average of the results.
- KPIs: time per brief; rankings and clicks in Search Console at 90 days.
Workflow 6: assisted draft with proprietary material
- Goal: get a first draft that already contains your examples, data and opinions.
- Input: brief (Workflow 5), context pack, notes or a recording of a 20-minute interview with an internal expert.
- Steps: 1) transcribe the interview; 2) ask the AI for a draft that only uses facts found in the material and flags in square brackets where missing data is needed; 3) the author completes and rewrites it.
- Output: draft with explicit placeholders for data and checks.
- Human review: the author checks every figure, removes generic sentences and adds examples. An internal expert reviews for technical accuracy.
- KPIs: hours per published article; percentage of text rewritten; average time on page.
Workflow 7: editorial review and tone check
- Goal: enforce style guidelines and tone of voice without reading everything three times.
- Input: text to review, style guidelines with examples of right and wrong sentences.
- Steps: 1) the AI flags off-tone sentences, unsourced claims, repetition and jargon; 2) suggests specific fixes without rewriting everything; 3) the author accepts or rejects them.
- Output: list of corrections with reasons.
- Human review: the author decides. The automated check does not replace the final read.
- KPIs: number of review rounds per piece; errors reported after publication.
Workflow 8: one piece of content, five formats
- Goal: turn an article or webinar into LinkedIn posts, a newsletter, video scripts and FAQs.
- Input: published, approved content; editorial calendar.
- Steps: 1) the AI pulls out the strongest passages; 2) adapts them to each channel following the channel’s rules; 3) the social media manager selects and schedules.
- Output: a pack of derivatives ready for review. The full method is in the article on content repurposing.
- Human review: checking consistency with the original content and fit for the channel.
- KPIs: derivatives published per main piece; engagement and clicks to the website.
To do. Write a list of 15 words and expressions your brand does not use and add it to the context pack. It is the quickest way to take the mass-produced feel out of your copy.
Workflows for Google, Meta and LinkedIn ads
In advertising, AI helps produce and organise ad variants and read search terms faster. The platforms already have built-in generation features; your job is to feed them strong messages and decide what goes into testing.
Workflow 9: message matrix and ad variants
- Goal: generate headline and description variants organised by messaging angle, for use on Google Ads, Meta and LinkedIn.
- Input: voice of the customer (Workflow 2), value proposition, character limits for each format, terms to avoid for policy reasons or by choice.
- Steps: 1) the AI suggests five angles (problem, outcome, proof, objection, urgency); 2) writes variants within the character limits for each angle; 3) the team chooses the combinations to test, one variable at a time.
- Output: angle-by-format matrix, with the selected variants labelled.
- Human review: checking ad policies, verifiable promises and consistency with the landing page.
- KPIs: CTR and conversion rate by angle; cost per qualified lead, not per generic lead.
Workflow 10: search term and negative keyword analysis
- Goal: spot irrelevant searches and new topics to cover more quickly.
- Input: export of the Google Ads search terms report, list of services offered and not offered.
- Steps: 1) the AI groups terms by theme and intent; 2) suggests negatives and new ad groups; 3) the specialist confirms them one by one.
- Output: list of proposed negatives and opportunities, with reasons.
- Human review: every negative must be confirmed by whoever manages the account: a wrong negative blocks good traffic.
- KPIs: share of spend on irrelevant terms before and after; weekly analysis time.
Workflow 11: creative brief for video and images
- Goal: give the designer or videographer a precise brief for social creative.
- Input: winning angles from Workflow 9, format specifications (vertical, square, length), visual guidelines.
- Steps: 1) for each angle the AI writes a concept with an opening hook, scenes, on-screen text and close; 2) the team chooses; 3) the creative produces it.
- Output: ready creative briefs, one per concept.
- Human review: the brand lead checks visual consistency and the absence of claims that cannot be proven. If you use generated images, check the rights and transparency obligations.
- KPIs: time from brief to publication; view rate of the first few seconds; cost per qualified lead by concept.
Watch out. Algorithm-optimised campaigns learn from the signals they receive. If the best ads bring in lots of off-target contacts, the platforms’ AI will go looking for more of the same. That is why the right KPI is the qualified lead recorded in the CRM, as we explain in the article on connecting campaigns and CRM to revenue.
Workflows for lead qualification and sales enablement
Here AI works on two fronts: it helps you work out quickly which contacts deserve attention and it equips salespeople with ready-made material and answers. The decision on who to call and what to offer stays with sales.
Workflow 12: enriching and first-pass qualifying of new leads
- Goal: get leads to the salesperson already described and classified.
- Input: form data, the company’s website, written qualification criteria (industry, size, role, need).
- Steps: 1) an automation passes the form data to the AI; 2) the AI summarises who the company is and assigns a class (A, B, C), justifying it against the criteria; 3) the result goes into a CRM field; 4) A leads trigger an immediate alert.
- Output: summary and class on the contact record.
- Human review: the salesperson confirms or corrects the class; corrections are used to improve the criteria. The scoring logic is described in the article on B2B lead scoring.
- KPIs: first response time to A leads; percentage of classes corrected by sales; rate of A leads becoming opportunities.
Workflow 13: sales call preparation
- Goal: go into every first call knowing who is on the other side and what to ask.
- Input: account profile (Workflow 4), CRM history, pages viewed on the website if tracked, notes from any previous contact.
- Steps: 1) the AI produces one page with context, hypotheses about the need, five discovery questions and two relevant cases from your material; 2) the salesperson reads it five minutes beforehand.
- Output: preparation sheet attached to the appointment.
- Human review: the salesperson discards implausible hypotheses. The questions are a prompt, not a script.
- KPIs: percentage of first calls that move to the next stage; qualitative feedback from sales.
Workflow 14: objection handling and sales materials
- Goal: give the sales team consistent answers to the most common objections.
- Input: collected objections (Workflow 2), answers that worked according to the best salespeople, case studies and technical documentation.
- Steps: 1) the AI organises objections by category and writes short answers based only on the material provided; 2) senior salespeople correct them; 3) the document becomes the basis for an internal assistant the team can consult.
- Output: library of objections and answers, with sources.
- Human review: approval by the sales manager; quarterly update.
- KPIs: onboarding time for new salespeople; close rate on deals with a recorded objection.
Self-assessment question. Are your criteria for calling a lead qualified written down anywhere? If AI is going to classify leads, it needs rules that today may only exist in the sales director’s head.
Workflows for CRM, meetings and follow-up
This is the area with the fastest return for sales teams: summaries, CRM updates and follow-ups are repetitive, daily and easy to check. Many CRMs, HubSpot included, have built-in AI features for summarising and writing; check what your plan includes before adding more tools.
Workflow 15: from meeting to summary with next steps
- Goal: close every call with a shared summary and assigned actions.
- Input: meeting transcript (from Teams, Google Meet, Zoom or a dedicated tool), with participants’ consent.
- Steps: 1) the AI extracts decisions, needs raised, objections and next steps with owner and date; 2) the salesperson corrects it; 3) the summary goes into the CRM.
- Output: summary in a fixed format on the opportunity record.
- Human review: the salesperson checks above all commitments, prices and dates: these are where a mistake costs money.
- KPIs: percentage of meetings with a summary in the CRM within 24 hours; time spent writing notes.
Workflow 16: updating the CRM from notes
- Goal: keep CRM fields and stages up to date without manual entry.
- Input: meeting summary (Workflow 15), list of CRM fields with allowed values.
- Steps: 1) the AI suggests field values (stage, budget stated by the customer, timing, decision-makers); 2) the salesperson confirms with one click; 3) the automation writes to the CRM.
- Output: updated fields with a record of who confirmed.
- Human review: confirmation required before writing. No automatic stage changes without a human.
- KPIs: completeness of key fields on open opportunities; reliability of the sales forecast.
Workflow 17: personalised follow-up
- Goal: send an email within a few hours of the call that picks up on what was said.
- Input: summary, relevant materials (case studies, data sheets), the salesperson’s tone of voice.
- Steps: 1) the AI writes a short draft that echoes the customer’s needs in their own words and proposes the next step; 2) the salesperson edits it and sends it from their own account.
- Output: follow-up email sent and logged in the CRM.
- Human review: always. No follow-up goes out without the salesperson having read it.
- KPIs: time between call and follow-up; reply rate; deals moving forward a stage within two weeks.
To do. Choose Workflow 15 as your first experiment if the team makes at least ten calls a week. It is easy to measure, the result can be checked in two minutes and the benefit shows from the first week.
Workflows for reporting and performance analysis
AI shortens the time between “we have the data” and “we know what to do”, provided the input data is clean and the figures are always double-checked. Language models can get calculations wrong or invent trends: for figures, use analysis features that run code on the files and check the totals.
Workflow 18: commentary on the monthly report
- Goal: move from the dashboard to commentary that management and sales can read.
- Input: Google Ads, Meta, LinkedIn, GA4 and CRM exports for the same period, the month’s targets.
- Steps: 1) the AI, using file analysis, calculates changes against the previous month and the target; 2) writes three results, three problems and three proposed actions; 3) the marketing manager checks the figures on the dashboard and rewrites the actions.
- Output: one-page report with verified figures and commentary.
- Human review: spot check of at least five figures; the proposed actions are decided by the manager.
- KPIs: hours per report; time between month end and delivery; actions decided and carried out.
Workflow 19: anomaly alerts
- Goal: spot unusual drops or spikes within a few days.
- Input: daily data from campaigns, website and CRM; agreed thresholds.
- Steps: 1) an automation compares the data with the average of previous weeks; 2) when a metric crosses the threshold, the AI writes a note with possible causes to check; 3) the alert reaches the owner.
- Output: alert with the metric, the deviation and hypotheses to check.
- Human review: the hypotheses are a checklist, not a diagnosis. The owner checks and decides.
- KPIs: days between a problem starting and it being spotted; false alarms per month.
Workflow 20: won and lost deal analysis
- Goal: understand each quarter what distinguishes won deals from lost ones.
- Input: the quarter’s opportunities from the CRM with source, industry, duration, loss reasons and meeting notes.
- Steps: 1) the AI groups loss reasons and recurring features of won deals; 2) cross-references them with lead source; 3) suggests hypotheses on channels and messaging; 4) marketing and sales discuss them in a meeting.
- Output: document with trends, hypotheses and decisions taken.
- Human review: correlations based on few cases are fragile. With fewer than 30 deals, treat everything as a pointer, not proof.
- KPIs: decisions taken after the analysis; change in close rate the following quarter.
| Workflow | Data needed | Where it usually is | Ready today? (yes/no) | Who prepares it |
|---|---|---|---|---|
| 18. Monthly report | Platform and CRM exports | Google Ads, Meta, LinkedIn, GA4, CRM | ||
| 19. Anomaly alerts | Daily data | Platforms, data connectors | ||
| 20. Won and lost | Opportunities with loss reasons | CRM |
The risks: data, privacy, hallucinations, tone of voice and human oversight
The risks of AI in marketing are manageable if you know them in advance and assign them to someone. There are also legal obligations: AI literacy for staff, transparency towards people talking to a chatbot, informing employees.
What the rules say as of October 2026
Article 4 of Regulation (EU) 2024/1689, the AI Act, has applied since 2 February 2025 and also covers organisations that simply use third-party tools such as ChatGPT or Copilot. The Digital Omnibus on AI, Regulation (EU) 2026/1744 published in the Official Journal of the European Union on 24 July 2026, rewrote the article: providers and deployers must take measures to support the AI literacy of those using the systems on their behalf, tailored to their skills, the context of use and the people affected, without having to guarantee a specific level (Future of Privacy Forum). So there is no mandatory certificate, but you need to be able to show what you have done. The Commission has collected practical guidance in its AI literacy questions and answers.
Since 2 August 2026 the transparency obligations of Article 50 have also applied: anyone interacting with a chatbot must know they are talking to an AI system. In Italy, Law 132/2025, the Italian AI law in force since 10 October 2025, requires among other things that employers inform workers about the use of AI and that professionals inform their clients. The full picture is in the article on mandatory AI Act training.
The risk map
| Risk | How it shows up | Countermeasure | Who is responsible |
|---|---|---|---|
| Personal and confidential data | Proposals, names, emails pasted into personal accounts | Business accounts only, processing agreement, minimal data, anonymisation | Owner and privacy lead |
| Hallucinations | Made-up figures, sources, product features | Mandatory sources, checking figures, placeholders for missing data | Workflow reviewer |
| Tone of voice | Correct copy that sounds like everyone else’s | Context pack, approved examples, banned words | Marketing manager |
| Intellectual property | Images or text too similar to existing works | Checks before publication, your own material as a base | Brand lead |
| Unsupervised automation | Emails or CRM updates sent by mistake | Human confirmation before any outward-facing action | Workflow owner |
| Transparency | Website chatbot that does not identify itself | Clear notice at the start of the conversation | Marketing and legal |
| Vendor lock-in | Prompts and context tied to a single tool | Prompt library and context in your own documents | AI lead |
Human oversight in practice
Human review does not mean a distracted read-through. It means a person with expertise in the subject, a list of things to check and the power to block the output. Define three levels. A light level for internal use (summaries, research drafts): spot checks. A medium level for published content (articles, posts, ads): full review and fact-checking. A strict level for anything that speaks to a specific customer or touches on prices, contracts and personal data: explicit approval before sending.
Watch out. A language model can cite a source that does not say what it attributes to it. You check by opening the source, not by asking the model whether it is sure.
- The whole team uses business accounts with a data processing agreement
- There is a written rule on which data must never be entered
- Every workflow has an assigned review level
- External-facing chatbots state that they are an AI system
- Employees have been informed about the AI tools in use
- We have recorded the training carried out (dates, content, participants)
30-day adoption plan
In 30 days you can go from scattered use to two or three measured workflows, with documented rules and training. The plan has four weeks, each with an exit criterion: if it is not met, you extend the week rather than skipping ahead.
Week 1: map and rules
Fill in the process map from chapter 1 and the matrix in Appendix A. Choose two or three workflows. Switch on the main tool’s business plan, write the five rules and prepare the context pack. Exit criterion: workflows chosen, owners named, rules shared.
Week 2: trial by hand
Run the chosen workflows by hand, using the prompts in Appendix C adapted to your needs. Measure time before and after on at least five cases per workflow. Fix the prompts after each round and save the versions. Exit criterion: at least one workflow producing usable output without heavy rewriting in four cases out of five.
Week 3: training and consolidation
Train the team on the chosen workflows and the rules, with a one-hour hands-on session per role, and record attendance and content: it also counts as documentable AI literacy. Connect the workflows to where work actually happens (shared projects, CRM, folders). Exit criterion: everyone involved has run the workflow at least once on their own.
Week 4: measure and decide
Compare the KPIs with the initial measurement. For each workflow decide: continue, modify, automate or switch off. Only workflows that work by hand move on to automation. Exit criterion: a one-page document with results and the next three workflows to try.
Further reading. If you want to see examples of applications in an SME, from content to data analysis, you will find seven cases in the article on artificial intelligence for SMEs. When a workflow is mature enough to become an automation or an assistant connected to your data, that is the work we do with our AI solutions.
Appendix A: Impact/effort matrix
Give each activity a score from 1 to 5 for impact and from 1 to 5 for effort. Impact: hours freed up a month, effect on quality, effect on leads and sales. Effort: data preparation, integrations, people to train, risk if wrong.
| Activity or workflow | Hours freed a month (1-5) | Quality (1-5) | Effect on leads and sales (1-5) | Average impact | Data to prepare (1-5) | Integrations (1-5) | Risk (1-5) | Average effort | Quadrant |
|---|---|---|---|---|---|---|---|---|---|
| Example: 15. Meeting summaries | 4 | 4 | 3 | 3.7 | 1 | 2 | 2 | 1.7 | Start now |
How to read the result:
| Quadrant | Impact | Effort | What to do |
|---|---|---|---|
| Start now | 3 or more | Less than 3 | First workflows in the 30-day plan |
| Project | 3 or more | 3 or more | Plan with data and integrations, after the first month |
| Habit | Less than 3 | Less than 3 | Leave to individuals, within the shared rules |
| Postpone | Less than 3 | 3 or more | Do not start now |
- Two people scored it, one from marketing and one from sales
- We chose no more than three workflows from the “Start now” quadrant
- Each chosen workflow has an owner
Appendix B: Workflow cards for all 20 workflows
| No. | Workflow | Main inputs | Output | Human review | KPIs |
|---|---|---|---|---|---|
| 1 | Competitor profile | Websites, LinkedIn, reviews | One-page profile with sources | Prices and figures checked against sources | Time per profile, profiles updated per quarter |
| 2 | Voice of the customer | Anonymised transcripts, emails, reviews | Themes with frequency and quotes | Confirmation by a salesperson | Pages rewritten, conversion |
| 3 | Positioning brief | Profiles 1 and 2, owner’s notes | Three options and a recommendation | Owner’s decision | Time to approved brief |
| 4 | ABM account research | Target list, websites, news | Account profile in the CRM | Signals checked before use | Reply rate with/without profile |
| 5 | Editorial brief | Keywords, SERPs, customer questions | Brief with intent and structure | Content lead | Time per brief, clicks at 90 days |
| 6 | Assisted draft | Brief, context, internal interview | Draft with placeholders | Author and internal expert | Hours per article, text rewritten |
| 7 | Editorial review | Text, guidelines | Corrections with reasons | The author decides | Review rounds, post-publication errors |
| 8 | Repurposing | Approved content | Derivatives by channel | Consistency and adaptation | Derivatives published, clicks to site |
| 9 | Ad matrix | Voice of the customer, format limits | Variants by angle | Policy and landing page consistency | CTR, cost per qualified lead |
| 10 | Search terms | Search terms report, services | Negatives and opportunities | Specialist confirmation | Spend on irrelevant terms |
| 11 | Creative brief | Winning angles, formats | Concepts for the creative | Brand and claims | Initial view rate, qualified leads |
| 12 | Lead qualification | Form, website, criteria | Summary and class in the CRM | Salesperson confirmation | Response time, classes corrected |
| 13 | Call preparation | Account profile, CRM history | Preparation sheet | Salesperson | Calls moving forward a stage |
| 14 | Objections | Objections, effective answers | Library with sources | Sales manager | Onboarding time, close rate |
| 15 | Meeting summaries | Transcript with consent | Summary with actions | Commitments, prices, dates | Summaries within 24 hours |
| 16 | CRM update | Summary, allowed fields | Updated fields | Confirmation before writing | Field completeness |
| 17 | Follow-up | Summary, materials | Email sent and logged | Always the salesperson | Time to follow-up, replies |
| 18 | Monthly report | Platform and CRM exports | One-page report | Spot check of figures | Hours per report, actions carried out |
| 19 | Anomaly alerts | Daily data, thresholds | Alert with hypotheses | Owner checks | Days to discovery, false alarms |
| 20 | Won and lost | Quarter’s opportunities | Trends and decisions | Caution with few cases | Decisions taken, close rate |
Appendix C: Mini prompt library
Every prompt follows the same structure: role, context, task, constraints, format. Replace the parts in square brackets and save the version that works in the shared library, with date and author. They work with ChatGPT, Claude, Gemini and Copilot; results vary from one model to another, so test them on yours.
C1. Competitor profile (Workflow 1) You are a B2B marketing analyst. Analyse [competitor name, URL]. Gather: services offered, stated audience, main messages on the home page, public prices if any, active channels. For each piece of information give the source with a link. If something is not available, write “not found” rather than inferring it. Format: table with columns Item, Content, Source.
C2. Voice of the customer (Workflow 2) Below are [number] anonymised transcripts of calls with prospective customers in [industry]. Group by theme: problems, objections, desired outcomes, words used to describe the problem. For each theme, state how many transcripts mention it and quote two phrases verbatim. Do not add themes that do not appear in the text.
C3. Editorial brief (Workflow 5) You are a content strategist. Main keyword: [keyword]. Related keywords: [list]. Customer questions: [list]. State the predominant search intent, suggest a structure with headings written as questions, and list what we need to provide (data, examples, opinions) to make the content different from what is already published. Do not write the article.
C4. Draft with placeholders (Workflow 6) Use only the facts found in the attached material: [brief, interview transcript]. Write a draft of [length] words in the tone described in [guidelines]. Wherever you need a figure, example or source you cannot find in the material, insert [TO CHECK: what is needed]. Avoid the expressions listed in [banned words].
C5. Ad variants (Workflow 9) You are a copywriter for B2B campaigns. Value proposition: [text]. Main objections: [list]. For each of the five angles (problem, outcome, proof, objection, urgency), write three headlines within [n] characters and two descriptions within [n] characters. No superlatives that cannot be proven. Format: table with angle, headline, character count.
C6. Search terms (Workflow 10) Here is the search terms export: [file]. We offer: [services]. We do not offer: [list]. Group the terms by theme and intent, flag irrelevant ones with the reason and suggest negative keywords with the match type. Put doubtful cases in a separate list.
C7. Lead qualification (Workflow 12) Qualification criteria: [industry, size, role, need]. Lead data: [form fields]. Company website: [URL]. Write a three-line summary of the company and assign a class A, B or C, explaining which criteria are met and which are missing. If the data is not enough, assign “to check”.
C8. Meeting summary (Workflow 15) Here is the transcript of a sales call: [text]. Extract: the customer’s needs in their own words, objections, decisions taken, next steps with owner and date, and information on budget and timing only if stated explicitly by the customer. Do not infer information that is not there. Format: bullet points under each heading.
C9. Follow-up (Workflow 17) You are [name], a salesperson at [company]. Call summary: [text]. Write a follow-up email of no more than 120 words that picks up two of the customer’s needs in their own words, attaches [material] and proposes a concrete next step with a date. Direct, friendly tone, no stock phrases.
C10. Report commentary (Workflow 18) Analyse the attached files: [exports]. Targets for the month: [list]. Calculate changes against the previous month and the target, showing your workings. Then write three results, three problems and three proposed actions, each linked to a figure. If a figure is missing or does not add up, flag it rather than estimating it.
Appendix D: 30-day implementation roadmap
| Days | Activity | Owner | Output | Exit criterion | Done |
|---|---|---|---|---|---|
| 1-2 | Process map (chapter 1) | Marketing manager and sales manager | Activity table | All recurring activities listed | [ ] |
| 3 | Impact/effort matrix (Appendix A) | Same people | Two or three workflows chosen | Owner named for each | [ ] |
| 4 | Business plan and data processing agreement | Owner or IT | Active accounts | Nobody uses personal accounts for work | [ ] |
| 5 | Five rules and context pack | Marketing manager | Rules document, context loaded | Rules shared with the team | [ ] |
| 6-10 | Trial workflows by hand, before/after time measurement | Workflow owners | Five cases per workflow | Usable output in four cases out of five | [ ] |
| 11-12 | Prompt review and saving to the library | AI lead | Versioned library | Prompts with an example of a good result | [ ] |
| 13-17 | Sessions by role and training log | AI lead | Attendance, content, materials | Everyone has run the workflow on their own | [ ] |
| 18-20 | Notice to employees and disclosures on external chatbots | Owner and privacy lead | Communication sent | Documentation filed | [ ] |
| 21-26 | Steady-state use, KPI collection | Whole team | Four weeks of data | KPIs measured for each workflow | [ ] |
| 27-28 | Evaluation meeting | Marketing and sales | Decision for each workflow | Continue, modify, automate or switch off | [ ] |
| 29-30 | Next month’s plan | Marketing manager | Next three workflows | One-page document approved | [ ] |
