When PPAI asked its PPAI 100 suppliers and distributors what stops them from going further with AI, the answers were consistent. For distributors: lack of internal expertise (65%), system integration issues (35%), employee resistance (27%) and data privacy concerns (27%). Suppliers ranked the same four, with integration higher at 50%.
Notice what is not on the list. Nobody said the models are not good enough. The barriers are all about fitting AI into a business that already runs on an ERP, a handful of supplier portals, a shared inbox and a lot of unwritten rules.
The integration trap
The instinctive response to "integration is hard" is to fund an integration. Connect the ERP to the AI. Build a middleware layer. Wait for the supplier to expose an endpoint.
The trouble is that most of the systems where promo work happens were never designed for that. eXtendTech's State of PromoStandards in 2026 describes where the standard stands after a decade: the focus has shifted from adoption to consistency, and the friction that remains is inconsistent part IDs, unclear documentation and undocumented rate limits. That is the good case, with a standard in place. Most client portals, decorator sites and older ERPs have nothing like it.
So the integration project becomes the AI project, and the AI project waits.
Why teaching removes three of the four barriers
There is a different way to fit AI into an existing operation, and it changes the barrier list.
Instead of connecting to the systems, the agent uses them the way a person does. It has its own computer. It logs into the ERP, opens the portal, reads the PDF, fills the form. What it needs to know, it learns from a screen recording and a narration by the person who already does the job.
Run PPAI's four barriers against that model:
- Lack of internal expertise. The expertise needed is knowing how the work is done today. Every distributor has that in abundance. Nobody needs to write a prompt library or learn an API.
- System integration. There is no integration. The agent operates the screens that already exist, including software with no API at all.
- Employee resistance. The person who does the work is the one who teaches it, keeps the approval points, and takes over when something looks wrong. The tool extends their judgment rather than replacing it.
- Data privacy. This one does not disappear by design alone. It has to be answered with architecture: isolated environments per company, credentials that never leave the pod that uses them, and a contractual commitment that client data is never used to train a model. It is a fair question to put to any vendor, and one we answer in full on our security page.
The four barriers PPAI found are barriers to integrating AI. They are much smaller barriers to teaching it.
What this looks like in practice
A CSR records one order being entered from a customer PO, narrating the customer's part-number quirks and where decoration details go. That recording becomes an agent that enters the next hundred orders the same way, pausing for sign-off where the CSR said it should. No connector was built. No supplier was asked for an endpoint. The team that was "resistant" is the team that taught it.
That is the model AltOps runs on. It is also, we think, the only way the operational side of promo adopts AI at the pace the marketing side already has.
Sources: PPAI Research, "Promo's AI Challenges and Opportunities"; eXtendTech, "The State of PromoStandards in 2026."
Written by Madhavam Shahi, teaching agents to run the back office at AltOps.
