RFP — Really Futile Process: Telecom’s Favorite Way to Kill Innovation Before It’s Born

There’s a special place in telecom purgatory where ideas go to die. It’s called the RFP.

For those unfamiliar (lucky you), RFP stands for Request for Proposal — but in practice, it’s more like Request for Paralysis. It’s a ceremonial ritual performed when telcos know they want to “do something, for example with AI” but aren’t sure what, how, or why. So they toss a vague document over the wall, cross their fingers, and wait for a 200-page response that gets judged by a committee whose exposure to generative AI peaked with ChatGPT’s launch headline.

There was a time when RFPs made sense. Two decades back, networks were mostly standardized. Everyone knew what a switch did, what OSS needed to track, and what BSS should bill. You could describe a requirement in 50 pages and get a decent system built by someone who’d actually read it.

But AI doesn’t work like that. GenAI? Definitely not. It’s not a static “requirement” — it’s a moving target. And while this tech is changing weekly, telcos are still clinging to a ritual that takes them 18 months to blink.

I’ve worked with enough telecom operators to know this: in 90% of cases, the RFP is just a patchwork of old documents. Half the lines were lifted from some 4-year-old cloud migration template. The other half were slipped in by a “friendly” vendor who’s probably going to win it anyway. Let’s not pretend it’s neutral. Most RFPs are written in the language of whoever got to the procurement team first.

And then everyone else is expected to respond, while playing by rules someone else wrote.

The RFP Lifecycle is of Hope to Helplessness in ~ 18 Months

Let's see how telcos work today .

• Step 1: Spend 3 months drafting an RFP that tries to explain a future that hasn’t happened yet.
• Step 2: Take another 6 months evaluating responses from actual experts while pretending you understand what “multi-agent orchestration using a foundation model” means.
• Step 3: Award the project to the lowest bidder who ticked the most boxes — bonus points if they used the words “Knowledge graphs,” “LLM,” and “agile” in the same paragraph.
• Step 4: Spend 6–12 months implementing something outdated six months ago.

Meanwhile, in the real world, AI moved from text to images to code to AI agents negotiating contracts, and MCP platforms began doing things no RFP committee could even spell, let alone specify.

Telcos should understand that RFPs don’t work in the Age of AI

• AI evolves weekly. By the time your RFP hits the vendor’s inbox, OpenAI has launched two new models, Google has launched Agentspace, and the solution you needed is now a plugin on Hugging Face.
• Agent-to-agent workflows are real now. Humans are no longer the only users of telecom APIs. Bots talk to bots. Good luck capturing that in a 2019-style Excel compliance sheet.
• Telcos are competing with startups. And startups don’t do RFPs. They do Slack threads, weekend hacks, and live demos on Tuesday.

The Evaluation Committee is made up of People Who’ve Never Touched the AI Tech

The best part is yet to come. Even if you manage to write the RFP , get 4–5 vendors to give responses, tthe ime for evaluation comes.And trust me it is the real fun: the panel of evaluators. Often brilliant in their fields — but not in AI, or GenAI. Still, they’ll confidently question the use of a particular LLM model, and suggest replacing it with something they read about on an airline magazine in 2018 or just checked on ChatGPT. All while tossing in a few “smart-sounding” objections to show their bosses they’re doing a thorough job.

Spoiler: They’re not.

What to Do Instead: Dare to Move Fast ..and Maybe Break Some RFP Templates

• Here’s my radical idea: don’t wait 18 months to decide if a vendor is good. Give them your data , Let them try and show you something in two weeks
• Use sandboxes
• Start with pilots — Don’t start with 100 requirements of a RFP. Start with 5 outcomes.
• Test with your data — Build a working prototype in weeks, not promises in binders
• Iterate based on outcome — Let your engineers talk directly to theirs. Kill the 18-deck PPT barrier.
• Co-create with startups or internal skunkworks teams.

In short: fail fast, learn faster.

If you want AI to work for you? Treat it like a startup product, not a compliance checkbox.

The Future Will not Wait for Your RFP

Please , Please , Telcos, you need to stop acting like government procurement arms and start acting like venture-funded product teams. The days of “raise an RFP and hope for the best” are gone.

The next-gen AI agent that could revolutionize your business? Yeah — it just read your RFP and ghosted.