Most teams don’t struggle to find AI tools — they struggle to pick the right kind. Off-the-shelf automation platforms promise a fast start. Custom agents promise a fit to your exact workflow. Both promises are real, and they apply to different problems.
This is the decision framework we walk clients through in discovery, before anyone writes a line of code or signs a licence.
What an off-the-shelf tool actually gives you
A mature platform buys you a running start and a support contract. For a workflow that looks like everyone else’s, that is usually the right trade.
- Days to a first working automation, not weeks
- Lower upfront cost for simple, common workflows
- Documentation and support built for a wide audience
- No maintenance burden on your team — until you outgrow it
What a custom agent gives you instead
A custom agent is built around your data, your systems, and the exceptions your team actually hits. The cost sits at the front; the marginal cost per run is close to nothing.
Where the difference shows up
It rarely shows up in the demo. It shows up in month three, when the volume of things the tool cannot handle stops being a rounding error.
If more than a third of your volume is “exceptions,” you don’t have an exception problem. You have a tooling mismatch.
The two paths, side by side
| Dimension | Off-the-shelf tool | Custom agent |
|---|---|---|
| Time to first value | Days to two weeks | Three to eight weeks after discovery |
| Cost shape | Per-seat or per-run subscription, grows with volume | Build cost upfront, low marginal cost per run |
| Fit to your process | You adapt to the tool | Tool adapts to you |
| Exception handling | Escalates to a human | Encoded in the agent’s rules |
| Data ownership | Lives in the vendor’s platform | Stays in your systems |
| Who maintains it | The vendor | Your team, or your partner |
How to decide which one you need
Three questions separate the two paths more reliably than any feature comparison.
Three questions worth answering first
- What fraction of volume is exceptions? Under 10% and a generic tool is fine. Over 30% and you’re paying a human to do the tool’s job.
- Does the workflow touch data you can’t export? If the answer is yes, the integration cost usually dominates the licence cost.
- Would the process survive an audit? Regulated workflows need a decision trail that most off-the-shelf tools don’t produce.
A practical middle path
Start with the platform. Instrument it. Let it tell you where it breaks, and build only that piece custom.
That sequencing keeps the first invoice small and makes the second decision evidence-based rather than architectural taste.
Where to start
Pick the single workflow costing you the most hours this quarter and run the three questions against it. Most teams find the answer is obvious once it is written down — including this one.
