Custom Instructions for AI Agents: How to Turn an ElectronicMoron into an FMCG Marketing Genius
Let me tell you something absolutelyterrifying. Right now, across the globe, thousands of consumer goods companiesare spending millions on artificial intelligence systems that have all thecreative capacity of a lobotomized goldfish. They're paying premium rates forcutting-edge technology and then using it like my mother uses her iPhone – toask about the weather and occasionally send a blurry photo of the garden. It'sthe equivalent of buying a Lamborghini Aventador and then only driving it tothe end of your driveway to collect the mail.
For crying out loud, this is the mostadvanced technology mankind has ever created, and the average FMCG company isusing it to generate spreadsheets and answer customer emails with all thepersonality of a damp dishcloth. It's maddening. It's infuriating. It's likewatching someone use a chainsaw to trim their fingernails.
The problem, you see, isn't with the AIitself. The silicon-brained marvels we've created are actually quite brilliant– in the same way that a Formula 1 car is brilliant. But much like that F1 car,if you don't know how to drive it properly, you'll end up spinning out on thefirst corner while everyone else zooms past you toward the checkered flag. Andthe FMCG sector, with its razor-thin margins and relentless competition, cannotafford to be the industry equivalent of Nikita Mazepin.
What most companies don't realize – andthis is the bit that makes me want to throw my keyboard through the nearestwindow – is that these AI systems come with something approaching a superpower:the ability to follow custom instructions. Not just any instructions, mind you,but precisely tailored commands that transform these digital assistants fromglorified calculator watches into strategic thinking partners that canrevolutionize how you sell everything from toothpaste to tortilla chips.
I've spent the last several monthsexploring what happens when you give these electronic brains properinstructions, and I promise you this: by the time you finish reading thisarticle, you'll understand how to make these digital lackeys perform backflipsfor your FMCG business
What Are Custom Instructions and WhyThey Matter in FMCG
Let me explain what custom instructionsactually are, because most people have absolutely no idea. Imagine you've justhired a new marketing graduate – bright-eyed, bushy-tailed, and with a headfull of theories they've never tested in the real world. On their first day,you wouldn't just point at their desk and say "do marketing." Thatwould be idiotic. You'd tell them about your brand voice, your target audience,your competitors, your reporting structure, and most importantly, what youabsolutely don't want them to do.
Custom instructions for AI are exactlythe same, except the new hire is a trillion-parameter language model that neversleeps, never asks for a raise, and won't steal your parking space. Theseinstructions are essentially a permanent briefing that tells the AI who youare, what you do, how you want information presented, and what constitutes agood response in your specific context.
Using AI without custom instructions islike driving a Ferrari with the handbrake on. You'll move, sure, but you'll beburning through resources while producing a horrible smell and achieving afraction of what's possible. Most FMCG companies are doing precisely this –they're interacting with AI using default settings, having what amounts tobland, generic conversations when they could be engaging in laser-focusedstrategy sessions[1].
The difference is staggering. Withoutcustom instructions, asking an AI about consumer trends gets you aWikipedia-style summary that any intern could Google in fifteen seconds. Withproper instructions, that same AI becomes an industry-specific analyst thatconnects purchasing patterns to environmental factors, seasonal variations, andcompetitive movements, then suggests precise market positions that could increaseyour margin by 2.3%. And in the FMCG world, where success is measured infractions of percentage points, that's the difference between dominating yourcategory and being delisted by Tesco.
10 Practical Applications of CustomInstructions in FMCG
a) Consumer Behavior Analysis
If you think your current consumerinsights are deep, I've got news for you – they're as shallow as a paddlingpool in the Sahara. The typical FMCG analysis consists of spreadsheets showingthat people buy more ice cream when it's hot. Groundbreaking stuff, truly.
Custom instructions can transform yourAI into a behavioral psychology expert that doesn't just tell you whatconsumers are doing but explains why they're doing it. Take P&G, forinstance – they've been using advanced analytics to understand the emotionaltriggers behind purchasing decisions for years[2].
With the right instructions, your AIcan process millions of social media comments, review sites, and forum posts toidentify emerging micro-trends before your competitors even open their laptops.Here's a custom instruction that works:
"When analyzing consumer data,prioritize emotional drivers over demographic information. Identify patternsthat connect purchasing decisions to life events, and explicitly highlightcontradictions between what consumers say and what they actually do. Formatinsights as 'What They Say' versus 'What They Do' paired with actionablerecommendations."
b) Supply Chain Optimization
The average FMCG supply chain leaksmoney like a colander trying to hold the Atlantic Ocean. It's inefficient,reactive, and about as predictive as a fortune teller with cataracts. And yet,in an industry where logistics can account for up to 40% of your costs, this isprecisely where you should be deploying the most sophisticated AI tools at yourdisposal.
Unilever has been using AI to predictdemand fluctuations and optimize inventory for years, saving millions in theprocess[2]. But their systems are custom-built and eye-wateringlyexpensive. You can achieve similar results by giving your AI the rightinstructions:
"When assessing supply chain data,prioritize identifying demand pattern anomalies that fall outside threestandard deviations. Calculate reorder points based on variable lead timesrather than averages, and present recommendations in three tiers: urgentactions (next 48 hours), tactical adjustments (next 30 days), and strategicreconfigurations (next quarter). Always include estimated cost savings and riskassessments for each recommendation."
This turns a generic AI into a supplychain consultant that would normally charge you £2,000 a day and spend half ofthat time asking where the coffee machine is.
c) Dynamic Pricing Strategies
Pricing in FMCG is typically approachedwith all the strategic sophistication of a drunk man throwing darts – you mightoccasionally hit the bullseye, but it's more luck than judgment. Most companieseither copy their competitors or stick a finger in the air to see which way thewind is blowing.
Coca-Cola has been using algorithms tooptimize pricing across different markets, channels, and pack sizes for years[2]. Their systems take into account everything from localweather forecasts to sporting events. With custom instructions, your AI canprovide similar insights:
"When analyzing pricing data,evaluate elasticity by channel, pack size, and day of week. Factor incompetitive pricing moves from the past 90 days, upcoming promotionalcalendars, and seasonal demand patterns. Present recommendations in a decisionmatrix showing projected volume, revenue, and margin impacts for each pricingscenario. Include cannibalization risk assessment for adjacent products in yourportfolio."
Suddenly, your pricing discussions movefrom "Should we charge £1.99 or £2.29?" to understanding exactly howmuch money you're leaving on the table with each decision.
d) Personalized Marketing
The marketing campaigns of most FMCGbrands have all the personalization of a form letter from the tax office."Dear Valued Consumer" might as well be tattooed on the foreheads ofmarketing directors across the industry. It's embarrassing, especially whencompanies like PepsiCo are using customer data to create hyper-targetedcampaigns that speak directly to specific consumer segments[2].
Your AI can help you create genuinelypersonalized marketing with instructions like:
"When developing marketingcontent, segment audiences by psychographic profiles rather than demographics. Createmessage variations that appeal to different personality types within the sametarget segment. For each campaign concept, provide three versions:rational/feature-focused, emotional/benefit-focused, and social/status-focused.Include specific messaging recommendations for different platforms based onplatform-specific engagement patterns."
This transforms your generic AI into amarketing strategist that understands the fundamental truth of FMCG: peopledon't buy products, they buy better versions of themselves.
e) Product Development and Testing
Most FMCG product development is aboutas innovative as a tribute band – slight variations on things that alreadyexist, with predictably mediocre results. It's why supermarket shelves arefilled with endless variations of "new and improved" products thatare neither new nor particularly improved.
Kraft Heinz has been using AI to testnew flavors and packaging concepts, reducing development time and increasingsuccess rates[2]. Your AI can support similar innovation with the rightinstructions:
"When ideating new products, beginby identifying unmet consumer needs rather than category gaps. Prioritizeconcepts that blend attributes from adjacent categories. For each productconcept, generate a competitive analysis showing key differentiators, potentialbarriers to adoption, and likely competitive responses. Include suggestions fordisruptive packaging innovations or delivery mechanisms that could createadditional differentiation."
This turns your AI from an electronicnotepad into an innovation partner that challenges conventional categorythinking.
f) Shelf Analytics
The average FMCG manager spends moretime arguing about shelf space than actually improving their products. And yet,even with this obsessive focus, most companies have shockingly poor visibilityinto what's actually happening on store shelves across thousands of outlets.
Mondelez International has been usingAI-powered cameras to monitor product availability and optimize shelfarrangements[2]. With custom instructions, your AI can help analyze thisdata more effectively:
"When analyzing shelf data,prioritize identifying correlation patterns between adjacencies and purchasebehavior. Flag instances where competitor products are gaining disproportionatevisibility due to merchandising tactics. Calculate the revenue impact ofstockouts by store format and day part, and recommend specific facingallocations based on hourly sales velocity rather than period averages. Includevisual mockups of optimal shelf arrangements for different store formats."
Now your shelf analytics aren't justshowing what happened – they're telling you why it matters and what to do aboutit.
g) Customer Service and Support
FMCG customer service typically rangesfrom nonexistent to actively hostile. Most consumer goods companies handlecomplaints with all the empathy and efficiency of a Soviet-era bureaucrat. It'sas if they designed their customer service processes specifically to makepeople never want to contact them again – which, thinking about it, mightactually be the point.
Johnson & Johnson has been usingreal-time chatbots to provide immediate customer support[2]. Your AI can dramatically improve customer interactionswith instructions like:
"When handling customer inquiries,prioritize identifying the emotional need behind the practical question.Categorize issues by urgency and sentiment, providing immediate resolutionpaths for high-distress situations. For product complaints, include systematictroubleshooting steps followed by compensation recommendations based oncustomer lifetime value. Always incorporate product education that addressesthe root cause of common misuse scenarios."
This transforms generic responses intocustomer interactions that actually build brand loyalty rather than destroyingit.
h) Influencer Marketing
Most FMCG companies approach influencermarketing like a drunk teenager at a nightclub – desperately throwing money atanyone who seems remotely popular in the hope that some of that popularitymight rub off. The results are precisely as strategic and effective as you'dexpect.
Coca-Cola has worked with platformslike CreatorIQ to identify and evaluate potential influencers[2]. Your AI can provide similar analysis with the rightinstructions:
"When evaluating potentialinfluencers, prioritize engagement authenticity over follower count. Analyzeaudience overlap with your target market, sentiment patterns in comments, andhistorical brand partnership performance. Flag potential reputation risks basedon past controversies and calculate true reach based on engagement-to-followerratios. Present recommendations as tiered partnership opportunities withspecific content approaches for each influencer."
Suddenly, your influencer strategyshifts from "who has the most followers" to understanding exactlywhich voices will actually move your products off shelves.
i) Brand Sentiment Analysis
Most FMCG companies monitor socialmedia with all the sophistication of a parent checking their teenager'sFacebook page – they see only what's on the surface and miss all the importantstuff happening in private messages and obscure platforms.
Unilever has been tracking sentiment trendsacross multiple channels to identify potential issues before they become crises[2]. Your AI can provide similar early warnings withinstructions like:
"When analyzing brand sentiment,weight comments by author influence and emotion intensity rather than simplepositive/negative categorization. Identify sentiment shifts that deviate from90-day averages, especially in key demographic segments. Flag emergingnarrative patterns that could indicate potential reputation issues. Presentfindings as sentiment trend lines broken down by product, channel, and customersegment, with explicit recommendations for messaging adjustments."
This transforms social listening from apassive monitoring exercise into a proactive reputation management tool.
j) Sustainability Practices
Sustainability in FMCG is typicallyapproached with all the sincerity of a politician kissing babies – it makes fornice photos, but there's rarely any substantial follow-through. Most companiesproduce glossy reports filled with pictures of trees while continuing businesspractices that would make Captain Planet weep into his blue hands.
L'Oréal has been using AI to monitorenergy and resource usage across its operations[2]. Your AI can support genuinesustainability efforts with custom instructions like:
"When analyzing sustainabilityinitiatives, prioritize quantifiable impact over stated intentions. Calculatecarbon footprint reductions in absolute terms rather than percentages. Identifygreenwashing risks in current messaging by comparing claims against measurableoutcomes. Present recommendations in three categories: immediate operationalchanges, mid-term supply chain adjustments, and long-term business modelinnovations. Include competitive benchmark analysis showing your sustainabilityposition relative to category leaders."
This transforms sustainability from amarketing exercise into a genuine operational priority with measurableoutcomes.
How to Create Effective CustomInstructions
If you've managed to stay awake throughthe preceding 2,000 words, congratulations – you're clearly the kind ofmasochist who might actually implement these ideas. But knowing what's possibleand actually making it happen are as different as knowing how an internalcombustion engine works and being able to build one from scratch in your gardenshed.
Creating effective custom instructionsfor AI is both an art and a science – much like making a truly great episode ofTop Gear. It requires understanding the mechanics of how these systems workwhile also having the creative flair to push them in interesting directions.
The structure of effective custominstructions typically has two main components: "What should the AI knowabout you" and "How should the AI respond to you"[1]. This is essentially telling the AI who you are and whatyou want from it – a concept so mind-numbingly obvious that it's amazing morepeople don't do it.
For a brand manager in FMCG, effectiveinstructions might look something like this:
Whatshould the AI know about you:
"I'm a Brand Manager for a premium chocolate brand competing against Lindtand Godiva. I have a marketing background but limited data analysis skills. Ineed responses that balance creative thinking with commercial practicality. Iprefer visual formats like charts and tables rather than long paragraphs oftext. Our brand values are indulgence, craftsmanship, and accessibility. Ourtarget audience is primarily women aged 35-50 with above-average household incomewho see our chocolate as an affordable luxury."
Howshould the AI respond:
"Provide responses that focus on actionable recommendations rather thanbackground theory. Structure information in bullet points with a clear 'SoWhat' implication for each insight. Include specific execution examples whensuggesting marketing ideas. Always consider budget constraints – we're premiumbut not luxury, so we don't have Godiva's marketing budget. Highlight potentialrisks or challenges with any recommendation. When analyzing data, focus onidentifying patterns rather than individual data points, and always relatefindings back to our target consumer's motivations."
For a supply chain analyst, theinstructions would be completely different:
Whatshould the AI know about you:
"I'm a Supply Chain Analyst for a multinational beverage company. I'mhighly analytical with an operations research background. I need detailed,data-driven responses that address efficiency optimization. I'm comfortablewith complex statistical concepts and supply chain terminology. I'm responsiblefor reducing waste, improving forecast accuracy, and optimizing inventorylevels across 12 European markets."
Howshould the AI respond:
"Provide responses that prioritize quantitative analysis over qualitativefactors. Include statistical significance levels with any claimed correlations.Present information in tabular formats where appropriate, with clear hierarchyof importance. Always consider multiple scenarios (best/worst/most likelycases) when making forecasts. Reference specific supply chain methodologies andframeworks where relevant. Focus on actionable insights rather than descriptiveanalysis, with clear cost-benefit analysis for any recommendations."
For a product innovation team, differentinstructions again:
Whatshould the AI know about you:
"We're a product innovation team for a personal care brand. We're across-functional group with backgrounds in R&D, marketing, and consumerinsights. We need balanced perspectives that consider technical feasibilityalongside market potential. We're currently focused on developing sustainablepackaging solutions that don't compromise on user experience. Our developmenttimeline is typically 18 months from concept to market."
Howshould the AI respond:
"Provide responses that balance creative ideation with practicalconstraints. Structure recommendations in three categories: quick wins(implementable within 6 months), mid-term innovations (6-18 months), andbreakthrough concepts (18+ months). Always include consideration of consumerperception, manufacturing complexity, and supply chain implications. Challengeconventional category assumptions and suggest cross-category inspiration.Include specific materials, technologies, or approaches rather than genericsuggestions. For any concept, address potential regulatory or scale-upchallenges."
The beauty of custom instructions isthat once you've set them up, every interaction benefits from this contextwithout you having to repeat yourself. It's like having a team of specialistswho already know exactly what you need, how you think, and what you're tryingto achieve.
Conclusions and Reflections
After spending the last severalthousand words explaining how to make AI work properly for FMCG companies, I'mleft with a nagging thought: why aren't more companies doing this already? It'snot rocket science – although, to be fair, if it were rocket science, the AIcould probably help with that too.
The FMCG sector is famouslycompetitive, with margins thinner than Kate Moss in the 90s and consumerloyalty about as reliable as British weather in April. In this environment,even tiny competitive advantages can mean the difference between marketleadership and bankruptcy. And yet, most companies are using advanced AI theway a grandparent uses a smartphone – with extreme caution and accessing about2% of its actual capabilities.
The difference between companies usingAI with generic prompts versus those using sophisticated custom instructions islike the difference between a Reliant Robin and a McLaren Senna. They bothtechnically have engines and wheels, but one of them will leave the other sofar behind that they're not even competing in the same race anymore.
In five years' time, I predict the FMCGlandscape will be divided into two camps: those who mastered the art ofinstructing their AI systems and those who are desperately trying to figure outwhy they're losing market share every quarter. The former will use AI topredict consumer trends before consumers even know what they want, optimizetheir operations with precision that would make a Swiss watchmaker jealous, anddevelop products that seem almost telepathically attuned to market needs.
The latter will still be asking theirAI to "analyze these sales figures" and wondering why theircompetitors always seem to be one step ahead.
The choice, as they say, is yours. Youcan continue to use the most advanced technology ever created by humanity as ifit were a slightly improved calculator, or you can give it proper instructionsand watch it transform your business. And if you choose the former, well, youdeserve everything the market is going to do to you.
Frequently Asked Questions
Isn'tthis all a bit complicated for just talking to a computer?
Is it complicated to explain to aFormula 1 driver exactly how you want them to take a corner at 180mph? Ofcourse it is. But that's because you're operating a phenomenally powerful toolat the very edge of its capabilities. If you wanted simple, you could use abicycle instead. The complexity of properly instructing AI systems is directlyproportional to their power and versatility. Yes, it takes effort to creategood custom instructions. So does everything worth doing in business. Get overit.
Do I needtechnical expertise to create effective custom instructions?
About as much as you need to be amechanic to drive a car. Understanding the basic principles helps, but youdon't need to know how to build a language model from scratch to tell it whatyou want. What you do need is a clear understanding of your businessobjectives, your brand voice, and the specific outcomes you're looking for. Theactual writing of instructions is more about clarity of thought than technicalexpertise. If you can explain something clearly to a new employee, you canwrite effective custom instructions.
Cancustom instructions replace the need for human experts in my FMCG company?
Can a chainsaw replace a lumberjack?No, it just makes them dramatically more effective. Custom instructions don'treplace human expertise – they amplify it. Your marketing director won't bereplaced by AI, but the marketing director who knows how to use AI effectivelywill absolutely replace the one who doesn't. The most dangerous thing you cando is view this as an either/or situation. It's not humans versus AI; it'shumans with AI versus humans without it. And I know which side of that equationI'd rather be on.
Is therea risk of becoming too dependent on AI tools?
Is there a risk of becoming toodependent on electricity? On telephones? On the internet? Every transformativetechnology in history has made us "dependent" in the sense thatreturning to a world without it would be painful and inefficient. The real riskisn't dependency – it's complacency. If you set up your custom instructionsonce and never revisit them, you'll end up with AI that's perfectly optimizedfor yesterday's problems. The companies that will thrive are those thatcontinuously refine their instructions as business needs evolve, consumerpreferences shift, and the competitive landscape changes. Dependency isn't theproblem; stagnation is.
Decoding chatgpt unveiling its impact across software engineering realms
Generative ai five predictions for 2024
