We’re stuck in subscription sprawl. We have separate subscriptions with OpenAI, Anthropic, Cohere, plus a couple of specialized tools for specific use cases. Each one comes with different pricing, different usage limits, and different contract terms. It’s a mess.
Part of our BPM migration business case is consolidating that vendor landscape. The theory is that if we move to a unified platform that gives us access to 400+ AI models on one subscription, we simplify costs and we probably save money in the process.
But I can’t actually model the savings because the picture is so fragmented right now. We have spend scattered across different departments, some people using credits they’ll never exhaust, others hitting limits and paying overage fees. And nobody can tell me exactly which AI models we’re actually using and where the usage spikes happen.
Has anyone actually worked through this? How do you gather the data to make a comparison? And more importantly, has the consolidation actually delivered the cost savings the business case predicted, or have you found that you just shifted your spending patterns rather than reducing them?
I’m skeptical that consolidation alone is going to cut our costs significantly, but I also suspect we’re wasting money in ways we’re not seeing.
This is a problem we had almost exactly. Twelve different AI subscriptions across the company, spend scattered everywhere, nobody really owned the total picture. Here’s how we approached it:
First, we did an inventory. Went through billing statements going back six months, categorized usage by department and use case, tried to figure out which models were actually providing value and which were just legacy subscriptions nobody canceled.
What we found: about 30% of the spend was for models nobody was actually using anymore. Some teams had subscriptions for vendors they tried once and never went back to. That was the low-hanging fruit—just canceling unused subscriptions gave us immediate savings without changing anything.
Then we looked at actual usage patterns. One team was using GPT-4 heavily, another team could have done everything they needed with GPT-3.5 but nobody had told them that. A third team was hitting overage fees because they miscalculated volume. Those were fixable without consolidation.
When we finally modeled consolidation to a unified platform, the savings weren’t as dramatic as we expected, but they were real. We calculated that we were paying for about 60% of our subscriptions and not using them effectively. Consolidation reduced that waste from unused capacity.
Actual savings: about 25% of our previous AI spend monthly. That’s not nothing, but it’s also not the transformative number the pitch suggested. The real value was visibility. We knew where our money was going and could optimize usage.
For your business case: start with a serious audit of what you’re actually spending and what’s actually delivering value. That usually reveals more savings than consolidation alone will give you.
The other thing we realized: you can’t just add up your individual subscriptions and compare to a unified price. You have usage patterns that don’t map cleanly. One vendor overcharges for high-volume use, another overcharges if you don’t commit to volume. A unified vendor might map differently to your actual usage curve.
We tackled this by tracking usage for a full month across all vendors. Exported logs, categorized by department and model type. That data showed us exactly which models we relied on and where the inefficiencies were.
There’s usually significant waste in fragmented AI subscriptions. Teams over-subscribe to cover peaks but don’t use the capacity. Some teams don’t know what models would work for their use case and just keep subscribing to everything. Getting visibility on that waste is the first step.
Consolidation savings depend on your usage pattern. If you have predictable volume, a unified vendor might offer better per-unit pricing. If you have spiky usage, you might pay more because you can’t optimize for peaks and valleys the way you can with separate commitments.
For the business case, separate out savings from consolidation from savings from just eliminating waste. They’re different things.
Calculating savings is difficult because AI pricing is complex. Usage-based, commitment-based, overage fees, volume discounts—different vendors structure it differently. You can’t just compare headline pricing.
The real analysis is: audit your current spend, categorize known waste, project likely usage on a unified platform, then account for service level differences. Some models are cheaper than OpenAI but slower. Some are more expensive but more specialized.
We found about 20% in savings from actual consolidation, another 15-20% from eliminating waste. The consolidation part was modest. The waste elimination was significant but would have happened even without moving platforms.
For your business case, those are separate line items.
We faced this same problem. Subscriptions scattered everywhere, billing all over the place, nobody had a complete picture. We did a full audit and found we were spending about 40% more than we needed to because of redundancy and waste.
The consolidation piece helped, but most of the savings came from visibility and optimization. We stopped paying for models we weren’t using, adjusted team allocations to use more cost-effective models for specific tasks, and eliminated overage fees by better forecasting.
When we modeled the move to a unified platform with 400+ AI models on one subscription, the math looked better. We could estimate monthly cost based on our actual usage pattern rather than betting on multiple separate contracts. The pricing transparency made planning easier too.
What really helped: having all models accessible on one subscription meant teams could experiment with different AI options without each experiment requiring a new contract negotiation. That flexibility led to better model selection and actually reduced waste because people weren’t guessing.
Final numbers: about 30% savings from consolidation, plus another 15% from optimization we could do once we had full visibility. The business case was credible because it was based on actual usage data, not vendor promises.