When we start a project, it’s common to begin with a Proof of Concept (PoC) to manage business expectations and assess feasibility.
In the end, many expectations are set, but they rarely align with what the business truly wants, leaving the project stuck at this stage.
Today we’re sharing 4 key points to prevent a PoC from staying just that, based on our real experience with more than 15 different clients:
1. Unrealistic expectations: What to expect from a PoC?
The very first step in a PoC is to define what we want to achieve with the solution (we also call this Success Criteria, which basically means bringing down to earth what the business really wants and how we’re going to deliver it).
Spoiler: this is where 80% of PoCs fail.
It’s also crucial to set the minimum measurable values we need to meet during the PoC.
Remember, a PoC is meant to explain to the business what we’re going to do and what approximate results we can achieve (this is very important) to avoid falling into false expectations.
👉 Defining business Success Criteria is essential, but it’s equally important to set our technical Success Criteria, so everything makes sense and remains feasible.
2. Data is everything 😊
Everything looks nice and feasible until we get to the data.
We’ve seen challenges ranging from sales forecasting with just 1 year of monthly data (12 rows) to image classifiers trained with pictures taken from the internet that looked nothing like the real production environment.
Data quality and quantity are vital when working with Artificial Intelligence.
Depending on the solution, labeled data may or may not be required, but there must at least be sufficient volume and representativeness — not just 12 rows.
3. Managing the teams involved in the PoC
Another common reason PoCs fail is the lack of involvement from the teams.
Real-world examples include:
- Chasing the person responsible for providing data for weeks.
- Weekly review calls where teams don’t show up or don’t give the importance the project deserves.
👉 To avoid this, it’s critical to define responsibilities for each team member during the project, and to clearly state in the Success Criteria that we need a certain level of commitment from both sides.
4. Defining KPIs
Unrealistic KPIs, or what we like to call the “Santa’s wish list”.
It’s extremely important to define project metrics to avoid future conflicts.
Examples we’ve seen:
- Extracting values from handwritten invoices with 100% accuracy.
- Forecasting with MAE = 0.
👉 Manage business expectations carefully and establish sensible KPIs together, based on the data you actually have.
In summary
- Define success.
- Ground expectations.
- Secure data.
- Engage teams.
- Deliver value.
If you’re planning to start an AI PoC or are currently stuck in one, let’s talk. We’d be happy to share how we’ve helped other companies turn their PoCs into real value.