Portfolio Thinking for AI Projects: Quick Wins vs Big Bets

A queue of AI ideas is not a strategy

Every growing business now has a list. Somebody in sales wants proposal drafting. Operations wants demand forecasting. The managing director has seen a demo of an agent that handles supplier negotiations. Finance wants invoices read automatically.

The usual approach is to rank them by excitement or seniority of the person asking and work down. That produces a string of unrelated projects, each judged by the same criteria, and it tends to go wrong in one of two ways. Either everything is a safe, small automation and nothing changes much, or the business pours its whole budget into one ambitious project that stalls at month six.

Investors solved this problem a long time ago. You hold different kinds of assets for different reasons and judge each by the rules of its kind.

The three categories

Quick wins pay for the programme and build trust. Foundations are what make everything else possible and are almost never anyone's favourite project. Big bets are where the business might change, and most of them will not work.

How to sort an idea into the right bucket

Four questions do most of the sorting.

1. Does the process already exist and run manually today? If yes, it is likely a quick win candidate. If the AI would create a new capability, it is probably a bet.
2. Is the data needed already in one system, in usable shape? If no, there is a foundation project hiding inside this idea, and it should be named and costed separately.
3. Can a wrong answer be caught cheaply? Quick wins need a review step or low-stakes errors. If errors are expensive and hard to catch, it is a bet, whatever its size.
4. Would success change what the business sells or how it competes? That is the mark of a real bet, and it justifies accepting a higher chance of failure.

An illustrative example. A 90-person wholesale distributor has four ideas. Reading supplier invoices into the ERP is a quick win. Demand forecasting looks like a quick win until you find three years of sales history split across two systems with inconsistent product codes, so it becomes a foundation project followed by a quick win. A customer-facing ordering assistant is a moderate bet. Dynamic pricing across 12,000 products is a big bet.

A sensible allocation, and why it is not fixed

For a growing business in its first or second year of serious AI work, we would typically suggest something like this split of effort and budget:

• Most of it on quick wins, because they fund the rest and teach the organisation how to run AI projects
• A substantial share on foundations, rising if your data is messy
• A small, capped share on one big bet at a time

The exact proportions matter less than the discipline of deciding them in advance. A business with clean data and several delivered quick wins can reasonably shift towards bets. A business whose last three projects all hit data problems should shift towards foundations, however dull that feels.

The most expensive mistake is funding a big bet with quick-win expectations, then cancelling it at the first bad month.

Different rules for different kinds of project

This is the part most businesses skip, and it is the reason to think in portfolio terms at all.

A big bet should be broken into gates, each a cheap way to find out whether to continue. For the pricing example: can we predict price sensitivity for one category from historic data? Does a pricing change in that category move margin in a controlled test? Can staff trust and operate it? Each gate costs a fraction of the whole. Most bets should die at an early gate, and that is the system working. Our post on designing a pilot that proves something covers how to write those gates.

Reviewing the portfolio each quarter

1. List every active project with its category, spend to date and current evidence
2. Stop anything that has failed its stop rule, and record why
3. Promote successful quick wins to business as usual, with an owner and running budget
4. Check whether a foundation gap is holding up two or more ideas, and fund it
5. Decide whether the current bet advances to its next gate
6. Add new ideas only after the above, sorted into a category

The quarterly review is also the right moment to report upwards. If your board is asking pointed questions, the portfolio view answers most of the questions boards should be asking about AI.

Where outside help fits

Quick wins are often the right place to use an outside team, because they are well defined and the skills are specialised. Foundations frequently need a mix: your people know the data, a partner knows how to structure it. Bets benefit from an outside view mainly at the gates, when someone without attachment to the idea says whether the evidence is good enough.

At SpiderHunts we tend to be most useful sorting the list and delivering the first quick wins and foundations through our AI integration work. We will also tell you when an idea on the list is a bet pretending to be a quick win. That conversation is more common than any other.

Frequently asked questions