Every company that starts looking at AI ends up with a list. Customer service wants a chatbot. Finance wants invoices read automatically. Sales wants better lead scoring. Somebody in marketing has already signed up for three tools. Come February: forty items, no agreement on which ones matter, and money for maybe two.
It matters more than it sounds. A weak first project can sour a whole company on AI for a year. A good one funds the next three. Here’s a plain way to rank the list, with a few checks before you commit and a way of counting the return that doesn’t flatter you.
Same four lines for every idea
Ideas turn up in all sorts of shapes: a vague hope (“use AI for support”) in one email, a glossy vendor proposal in the next. Ask everyone for the same four things: the problem, who has it, how the work is done today, and a number that would show “better”. The number is the part people skip. No number, no project.
Score value and feasibility apart
Each idea gets two scores, one to five. Value is what you’d gain: hours a week, errors and what they cost, whether any revenue is involved. Feasibility needs more care, because that’s where plans die. Ask whether the data exists and who can get to it, whether the result would fit the tools people already use, whether you could run it yourselves or would need a vendor for everything, and whether the process will stand still long enough to automate. Put the ideas on a grid. High and high goes first. Low value and high feasibility is the trap, because it’s easy to build and nobody cares afterwards.
A worked example
Take a distribution company with three ideas on the table. Idea one: read supplier invoices automatically. Right now three clerks type them in. Anita, who runs the desk, has kept a spreadsheet of every supplier’s quirks since 2016. They arrive as PDFs in one mailbox, and nobody has switched on the ERP’s import function. Value, say a four. Feasibility, also a four.
Idea two is a customer chatbot. It would be nice, but the answers live in scattered documents and the legal team has questions. Value three, feasibility two. The third idea, forecasting demand for some four thousand products, has the biggest payoff. Call it a five. But the sales history lives in three systems that don’t agree with each other, so feasibility drops to a two.
Invoices go first. Forecasting takes second place. Its opening move is unglamorous, joining up and cleaning the sales data, and that can begin while the invoice work runs. The chatbot sits out this round. Six months on, if things go to plan, the clerks are handling exceptions instead of keystrokes, Anita’s spreadsheet has turned into rules the system follows, and the sales data is clean enough that forecasting starts with a real chance of success. Nothing in the method needs special software. It needs an afternoon and people willing to argue.
What happens when you skip this
Picture a different company, one that never ran the exercise. The chief executive sees a polished chatbot demo in January, and by March a project is under way. Six months and a fair amount of money later, the bot is answering questions from a help center that nobody has updated since last year. The support team doesn’t trust it. And because nobody counted tickets before the launch, no one can say whether it helped. The technology wasn’t the problem. The company simply never asked what the work looked like beforehand, or whether the data behind the bot was fit to use.
Same technology, different homework
Compare two tools at that company. One drafts internal meeting notes. If it’s wrong on a Tuesday, nobody is hurt. The other decides which customers get credit terms. One bad call there means an angry customer, perhaps a regulator’s question, and someone explaining why. So picture the mistake for every idea, and ask who would catch it. For a first round, pick something where a person confirms what the system suggests. You’ll learn how reliable it really is before anything depends on it.
Count the return like an accountant
Back to the invoices. The project has a build cost and a monthly running cost: usage fees, a part-time person watching the exceptions, retraining Anita’s team. Say it frees up two of the three clerks’ time. Good, but “frees up” is not “saves”. It turns into money only if those clerks move to chasing late payments or some other work that earns or protects cash. Count that, not the theoretical maximum. A small figure you trust is worth more than a big one you don’t.
Get a baseline on paper first. How many invoices a week, how many minutes each, how often they come out wrong? Without it, six months on, nobody can prove that anything changed.
Running the session
Got a list on your desk right now? Book three hours. Ask each idea’s owner to come, along with a finance person and whoever knows where the data actually lives. Stick the ideas on a wall and haggle over the scores. The first session is usually messy, and the useful part is the disagreement. When one person gives feasibility a five and another gives it a two, one of them knows something about the data that the other doesn’t, and you want to hear it before money is spent.
Pilot first, then re-rank
The distribution company runs the invoice pilot for about ten weeks. Anita owns it for the business and a developer sits beside her. Give it only to IT and it tends to finish as a demo that no one touches. They also set a stop rule up front. If most invoices still need fixing by hand at week ten, they stop and rethink.
Once the pilot ends, the whole list gets scored again. Tools improve, data gets cleaner, and what looked impossible in January may be simple by autumn. Each new project should build on what the last one left behind, like the cleaned sales data or an integration that can be reused.
If you’d like help with the early stages, our AI solutions cover everything from assessment through delivery. To check whether a build program is paying off, see how to measure whether your enterprise AI application development program is delivering, and for a plain explanation of the building blocks, AI solutions for business is a good place to start.
Modernize with confidence. Build for what's next.
Talk to our team about embedding AI into the core of your business.