Trending Articles
Trending Topics
Trending
ChatGPT can do more than recommend a restaurant or turn you into an action figure. The artificial intelligence engine is now being used by business owners across the world to harness insights that help them to cut costs and boost growth.
According to a recent New York Post survey, between 75% and 82% of small- and mid-size firms now lean on AI for a competitive edge.
The key to using ChatGPT effectively is to use a carefully honed prompt, which helps the generative AI to understand which data to analyse, and what to ignore. We share how to nail an AI prompt – because if you’re not asking AI, you can be sure your competitors are.
Crafting an important email to your top customer, or navigating a complex HR issue? These tasks probably shouldn’t be outsourced to a machine (even if you do ask for a few pointers). AI comes into its element when asked to digest significant volumes of data and spot trends.
Gemma Feare, is Co-Founder and CEO at everystate, which produces mushroom supplements for every stage of life. She says: “Running a startup comes with challenges. We don’t have big budgets, or big data. We’ve started everystate from scratch, writing our own playbook as we go. At times, I’ve felt real decision paralysis, pulled in a million different directions. AI has played a crucial role in helping cut through the noise.”
Every time decision paralysis takes hold, she inputs all the variables into AI to try and get an outside view. “It’s helped turn that pressure into a competitive advantage rather than a disadvantage,” she reveals. “We need to make decisions quickly to pivot with the ever-evolving market. The AI acts as an impartial advisor, presenting facts clearly to help us make the right call – fast.”
Prompts can run to several pages-worth of text; the more complexity and nuance to a question, the more layers you must add to the prompt. But you can get started with just a few lines.
“The best prompt recipe is: role → context → goal → guard-rails,” says Inna Weiner, former Head of Engineering at Google Cloud and now Vice President of Product for AppsFlyer, an AI-powered data platform.
“For example, you could try: ‘You are a CFO. Using the attached last-three-years P&L, flag the top five margin drags and suggest fixes in 200 words.’It works for newbies and pros alike.”
For those that are unfamiliar with using generative AI, it’s best to start with a single question, rather than covering multiple bases in a single prompt. “Pick one board-level question first,” Weiner says. “Aim the model at a single pain point – gross margin, churn, cash burn – before widening the net.”
Don’t be afraid of asking the big questions: which competitor is likely to steal your customers? Will this business model still be relevant in five years? The AI won’t always be 100% accurate but the responses will give you something to think about and may highlight a problem that you haven’t spotted.
Feare interrogates every strategic decision using generative AI, just to be sure nothing has slipped through the cracks. She shares some of the prompts she has used recently. She says: “These prompts have helped transform instincts and data into a sharp and focused strategy.”
Soldo’s mid-year Spend Index analyses spend data from over 25,000 Soldo customers across the UK and Europe to understand how businesses like yours are spending – and saving.
Whether you’re scaling, streamlining, or simply staying sharp, this is essential insight. Download this eBook now to learn more.
Generative AI is only as good as its data source – and the internet is full of errors. Bake in a healthy level of cynicism when reading the answers, and if something looks off, track down the source material and make sure the AI hasn’t gone down the wrong path.
“Large-language models predict words probabilistically, so slips are baked in,” says Weiner. “Benchmarks still show error rates near 40–50 % on complex reasoning. Like any smart human, the model can be brilliant one moment and wrong the next. Always run a quick reality check before acting.”
There are ways to limit errors, she adds. For example, setting very specific parameters on the data you want the AI to crunch. “Most commercial models are trained on data that is six-to-twelve months old,” she explains. “If you need the latest numbers, tell the model to pull a live web search or upload your own current files first.”
If you are uploading spreadsheets that may contain errors, try using ChatGPT to scan for missing values and inconsistencies first. “Chunk your ask,” says Weiner. “Break the prompt into numbered steps (clean → analyse → summarise) to cut hallucinations.”
Do not upload all your company data into a public account. Unless you specifically opt-out – usually by signing up to an enterprise account – your data, including company financials, may be used to train the AI model.
This could not only expose sensitive information to anyone asking a question about your industry or company, it could leave you in breach of contract or make you fall foul of GDPR regulations.
Weiner warns: “Remember that anything you paste into a public model can leak. Keep customer PII, company performance and trade secrets in a sandboxed, enterprise-grade stack.”
Generative AI is getting better all the time but, until it is ‘perfect’, you still need humans to check, verify and second-guess AI insights. At AppsFlyer, Weiner pairs AI with a human “red-team” to prevent embarrassing mistakes from slipping through the net.
She says. “Have someone verify the top five insights before they hit Slack. Always proof-read LLM outputs before sharing them out as ‘your opinion’ – the model won’t be in the room when you’re held accountable for its mistakes.”
AI will never match a human being’s ability to be creative, adds Feare. “This is why we don’t use AI for things that require human touches, like branding, copy or anything where nuance and emotional connection matter. It’s just not worth the cost to authenticity.
“In a crowded space, a brand only wins when people can see themselves in it. It’s about freeing up headspace from the repetitive stuff, so we can focus on what really matters.”
There are so many AI tools available right now to take care of the admin and grunt work, so you can focus on strategy. Feare reveals her top seven:
