The question comes up in almost every B2B distributor conversation: “We’re thinking about AI, but our ERP is a mess. Should we fix the ERP first?”
The instinct to fix the foundation before building on top of it is reasonable. But “fix the ERP first” is an expensive way to delay a decision that could be generating measurable ROI while the ERP project is underway. The answer to the sequencing question depends on something more specific: which system is producing the most visible, measurable operational pain right now?
What ERP Problems Actually Look Like
ERP problems in mid-market B2B distribution usually fall into three categories:
Data quality problems. Inconsistent SKU naming, duplicate customer records, incorrect inventory counts, missing pricing tiers. These problems affect any system that reads data from the ERP — including AI tools that rely on product catalog or inventory data.
Process gaps. The ERP is set up for one workflow but the business operates differently. Manual workarounds — spreadsheets alongside the ERP, email-based approval chains that should be system flows — are symptoms of process gaps.
Scale limitations. The ERP handles current volume but slows noticeably at peak, produces reports that take hours to run, or requires a specialist to maintain because it’s built on outdated infrastructure.
Each category has a different relationship with AI:
- Data quality directly limits AI effectiveness for specific use cases (order entry AI needs clean SKU data; support AI needs accurate inventory)
- Process gaps don’t affect AI directly — AI can sometimes route around process gaps better than a constrained ERP can
- Scale limitations rarely block an AI deployment until you reach much higher volume
What AI Problems Actually Look Like
AI deployment problems also cluster into categories:
Use case mismatch. The AI tool was deployed for a use case that doesn’t match the most painful operational problem. A chatbot deployed for marketing lead capture when the biggest pain is support volume doesn’t move the needle.
Missing integrations. AI tools that operate without access to live data produce generic answers. The integration step is where most AI deployments fail — not the AI itself.
No measurement baseline. The AI was deployed without establishing what the “before” looked like. Without a baseline, you can’t demonstrate that it’s working, which leads to the tool being abandoned before it reaches its resolution rate potential.
The Decision Framework
Rather than treating this as “ERP vs. AI,” the better question is: what is your most expensive operational problem right now, and which solution closes it faster?
If the answer is support volume: AI support layer first. This doesn’t require ERP integration to start — you can get to 20–30% resolution with knowledge-base-only training, then add ERP integration for order status queries to push toward 65%. The ERP state doesn’t block you; it just limits the ceiling until the integration is added.
If the answer is order accuracy: ERP data quality first. AI order entry automation that reads incoming purchase orders and enters them against your ERP will produce high exception rates if the SKU mapping in your ERP is inconsistent. Here, fixing the ERP data quality layer (specifically SKU normalization) is genuinely a prerequisite for the AI use case.
If the answer is reporting or visibility: AI analytics tools don’t require a perfect ERP — they work with imperfect data and can surface insights from what exists. Tools like Tableau, Power BI, or AI-assisted reporting layers can provide visibility that the ERP’s native reporting can’t, even on messy data.
If the answer is “we don’t know which problem is biggest”: Spend two weeks tracking the time your team loses to manual work. Not guessing — actually logging it. The result almost always reveals one or two processes consuming 40–60% of the manual overhead. That’s where to start, regardless of whether the solution is ERP or AI.
The Sequencing Reality
In practice, most mid-market distributors don’t need to choose. The AI use cases with the fastest and clearest ROI — support automation, invoice processing, basic order entry — can be layered on top of an imperfect ERP while ERP improvements happen in parallel.
The cases where ERP needs to come first are specific:
- SKU data so inconsistent that any AI order entry produces >20% exception rates
- Customer data so duplicate-heavy that AI can’t correctly route queries to the right account
- Inventory data so inaccurate that AI-driven recommendations produce fulfillment failures
Outside of those specific conditions, the “fix the ERP first” instinct often reflects a preference for a large, familiar project over a smaller, unfamiliar one — rather than a genuine operational requirement.
The AI support layer for €200/month can go live in 4 weeks. An ERP migration takes 6–18 months and costs 50–200x more. If the support layer is generating measurable ROI while the ERP project is in flight, that’s not a sequencing mistake. That’s smart resource allocation.
AHoosh helps B2B distributors diagnose which operational problem is worth solving first — and build the implementation plan to match. ahoosh.ai/contact