RFP Template for Software Vendor Selection
Updated: Oct 18
A common way for an RFP evaluation to end: four vendors score within two points of each other. The RFP asked 240 yes/no questions, and every vendor answered "yes" to almost all of them. The team has a spreadsheet full of green cells and no way to choose, so the decision goes to whoever gave the best demo.
The document did the opposite of its job. It looked like a comparison but removed the differences between the candidates. An RFP is useful only when it asks questions that vendors can't all answer the same way. This post shows how to write those questions, and how to score the answers so the evaluation doesn't collapse back into gut feel.
Ask questions only one vendor could answer your way
Every requirement in an RFP should be able to separate the vendors. "Supports SSO" can't: every serious vendor says yes. "Show SSO with SCIM provisioning against our identity provider, including what happens when a user is removed" can, because the answers will differ in ways that matter to you.
There are three ways to turn a checkbox into a question that separates vendors:
Quantify it. Instead of "scalable", ask for "the three-year cost at 400 users and 2 million API calls a month, itemised".
Ask them to show it. Instead of "has an audit log", ask "export the audit log for one admin action and show us the fields".
Ask how it fails. Instead of "high availability", ask "describe your last two incidents over an hour, and what changed afterwards".
The test: before sending the RFP, read each requirement and ask whether two competitors could give the same answer word for word. If they could, rewrite it or cut it. After responses come back, count the requirements where every vendor gave the same answer. Those told you nothing, so drop them from the scoring and from your next RFP.
Must-haves are gates, not weights
The most common scoring mistake is putting "meets must-have requirements" into the weighted score, for example at 30%. A vendor that fails a must-have can still win if it scores well enough on everything else, and then the "must" wasn't a must.
Treat must-haves as a pass/fail gate before any scoring. A vendor that fails one is out, however good the rest of its response is. Score only the vendors that pass, using the weighted criteria. Keep the must-have list short: a requirement belongs on it only if you would genuinely walk away from an otherwise excellent vendor that didn't meet it.
Vendor | Must-haves | Weighted score (30% for must-haves) | Gate first, then score |
A | All pass | 78 | 78 |
B | Fails SCIM provisioning | 82 | Out |
C | All pass | 75 | 75 |
D | All pass | 71 | 71 |
In this composite, vendor B wins the weighted score while failing a requirement the team had called non-negotiable. Gating first changes the winner.
The test: for each must-have, ask "would we really reject the best vendor over this one?" If not, move it to the weighted criteria. If more than a third of your requirements are must-haves, you probably haven't asked that question honestly.
Score independently before anyone talks
Once evaluators discuss responses, the first strong opinion in the room tends to shape everyone else's scores. Kahneman, Lovallo and Sibony argue that choosing between strategic options is like choosing between job candidates. Their Mediating Assessments Protocol, drawn from research on structured hiring interviews, breaks the decision into separate assessments made before anyone forms an overall judgement. Public procurement uses the same discipline. The Scottish Government's Procurement Journey guidance says panel members "should read and score the quality/technical aspects of the tenders independently", using criteria set in advance, and that the process to agree final scores "must be fully transparent and documented".
In practice:
Agree the criteria, weights and scoring scale before any response arrives, and don't change them afterwards.
Each evaluator scores every response alone, criterion by criterion, with a one-line reason for each score.
Collect the scores before the meeting. Discuss only the criteria where evaluators' scores are far apart, and record why the agreed score was chosen.
The test: collect independent scores for one criterion and look at the spread. If evaluators are two or more points apart on a five-point scale, the criterion is ambiguous or the evaluators read the response differently. Settle that before you settle the score.
Price the three years you'll actually pay
Pricing is the easiest part of an RFP to game. A low per-user licence can sit alongside connectors, sandbox environments, premium support and implementation services charged by the day, each one a line item you didn't think to ask about. The cheapest bid on the scorecard can turn into the biggest bill.
The defence is to set the assumptions yourself and make every vendor price against them:
Pricing assumptions (every vendor must use these)
- Users: 400 in year 1, 480 in year 2, 560 in year 3
- API calls: 2 million a month
- Environments: production plus one sandbox
- Support: response within 1 hour for severity 1, 24x7
- Integrations: identity provider (SSO + SCIM), data warehouse,
ticketing system
Price each item separately, for each year, plus a three-year total.
State what is excluded and the price of adding 50 users.
Here's the composite comparison that results. The figures are illustrative:
Three-year cost | Vendor A | Vendor C | Vendor D |
Licences | 540,000 | 432,000 | 480,000 |
Connectors | Included | 72,000 | Included |
Sandbox | Included | 36,000 | 15,000 |
Premium support | Included | 87,000 | 45,000 |
Implementation | 75,000 fixed | 210,000 (140 days at 1,500) | 120,000 fixed |
Total | 615,000 | 837,000 | 660,000 |
Vendor C's licences are 20% cheaper than A's, but its three-year cost is the highest of the three. That only shows up because every vendor priced the same assumptions, item by item.
The test: can you put every bid on one row of totals without making assumptions on a vendor's behalf? If you have to guess how many connectors a vendor will charge for, the RFP didn't ask the question.
Script the demo with your own data
A vendor's standard demo is designed to impress, not to inform. Dana, Dawes and Peterson found that people formed confident impressions from unstructured interviews, even when the answers were random. Those interviews diluted the useful information people already had and made their predictions worse. An unscripted demo works the same way: it leaves strong impressions that crowd out the evidence in the written response.
Give each shortlisted vendor the same scripted scenarios, with your data, and score each scenario as it happens. The US Federal Acquisition Institute describes technical demonstrations as giving "more verifiable information than reading about it on paper", and suggests involving end users. For a CRM, the scenarios might be:
Import our sample of 5,000 accounts and show the duplicate handling.
A rep opens an account and sees last week's product usage without leaving the record.
An admin removes a user from our identity provider, and access goes within the hour.
Score each scenario against the same scale, with the same evaluators in every demo. For references, ask what has been frustrating and where the vendor falls short. Reference customers rarely volunteer that, but they usually answer honestly when asked directly.
The test: after each demo, could each evaluator say whether every scenario passed, without checking with anyone else? If the main thing people remember is how polished the presenter was, the script wasn't tight enough.
The template
# RFP: [Solution category]
1. About us and the problem
[Who we are, the problem in our words, what success looks like
in 12 months. Name the top priority.]
2. Scope
[What this covers, and what it explicitly doesn't]
3. Must-haves (pass/fail)
[Short list. Failing any one ends the evaluation.]
4. Requirements (scored)
[Grouped by area. Each one quantified, "show us", or "how it
fails". Answer: yes / partial / no, plus evidence.]
5. Pricing assumptions
[Users per year, volumes, environments, support level,
integrations. Itemised, per year, three-year total, exclusions,
price of adding 50 users.]
6. Evaluation
[Criteria and weights; must-haves gate first; demo scenarios
that shortlisted vendors will run with our data.]
7. Response format
[Answer in the order above. Page limit. No marketing material.]
8. Timeline and questions
[Question deadline; answers shared with every vendor; submission
date; shortlist; demos; decision.]
Two rules about the process matter as much as the document. Answer vendor questions in one document shared with every vendor, so nobody wins by having better access to you. And write down the weights before responses arrive. If the weights change after you've read the bids, the evaluation has quietly been fitted to a favourite.
When an RFP is the wrong tool
A full RFP takes a lot of effort from you and from the vendors. It's worth it when you have several real candidates, the purchase will commit you for years, and vendors compete on more than price. Other situations call for something lighter:
Situation | Better approach | Why |
Several candidates, a multi-year commitment, high switching cost | Full RFP, as above | Structured comparison is worth the effort |
You don't yet know what to ask | A request for information first, then an RFP to a shortlist | Learn the market before you write requirements |
Two candidates, and the real risk is fit, not features | A paid pilot with success criteria written in advance | Using it tells you more than any document |
One obvious fit, or a small and easily reversed purchase | Buy directly, with a contract review | The comparison costs more than a wrong choice would |
The test: estimate what it would cost to reverse a wrong choice after one year. If the RFP would cost more than that, it's the wrong tool.
A worked example
The scenes above come from one composite selection: a 1,200-person company replacing its CRM. The figures are illustrative.
RFP sent to 4 vendors. First draft had 240 yes/no requirements.
- Rewritten to 46 requirements (quantified, "show us", or
"how it fails") plus 6 must-haves
- Responses: 9 of the 46 got the same answer from every vendor,
so they were dropped from scoring
Gate: vendor B failed SCIM provisioning -> out
(would have won on a weighted score of 82)
Independent scoring by 5 evaluators, then moderation
- 31 of 37 scored requirements had a spread under 2 points: accepted
- 6 discussed; 2 were found ambiguous and re-scored
Three-year cost (same assumptions)
- A 615,000 C 837,000 D 660,000
Scripted demo, 3 scenarios
- A passed 3, C passed 3, D passed 2 (user removal took 24 hours)
Decision: vendor A. Trade-off accepted and written down: weaker
reporting than C; revisit at the first annual review.
The decision rests on evidence that would hold up if someone asked later: a must-have gate, independent scores, one set of pricing assumptions and a scripted demo. Writing down the accepted trade-off matters too. When the weaker reporting starts to hurt in a year's time, the record shows it was a choice, not something nobody noticed.

What to do this week
If you have an RFP in progress, run the "same answer word for word" test on every requirement and rewrite the ones that fail. Split must-haves out as a pass/fail gate. Write the pricing assumptions block and the demo scenarios before anything goes to vendors. Agree the weights and the independent-scoring rule with your evaluators now, while nobody has a favourite. The vendor evaluation scorecard template gives you the scoring sheet, and the Engineering Templates & Playbooks guide has the rest of the set.
Final takeaway
An RFP is a tool for telling vendors apart, and it fails whenever every vendor can give the same answer. Write requirements that are quantified, demonstrated, or about how things fail. Make must-haves a gate, not a weight. Fix the criteria and weights in advance, and have evaluators score alone before anyone talks. Make every vendor price the same three years under your assumptions, and script the demo with your own data. Then write down the trade-off you accepted, so the decision still makes sense when its costs show up.
Once a vendor is chosen, the SOW Template for Engineering Projects turns the proposal into agreed deliverables.
Sources
Kahneman, D., Lovallo, D. and Sibony, O., "A Structured Approach to Strategic Decisions", MIT Sloan Management Review, Spring 2019: strategic options are like job candidates; the Mediating Assessments Protocol, drawing on research on structured employment interviews. https://sloanreview.mit.edu/article/a-structured-approach-to-strategic-decisions/
Scottish Government, Procurement Journey, "Technical Evaluation": panel members "should read and score the quality/technical aspects of the tenders independently" using pre-defined criteria; a moderation meeting after scores are collated; the process to agree final scores "must be fully transparent and documented". https://www.procurementjourney.scot/node/309
Federal Acquisition Institute, "Technical Demonstrations": demonstrations let buyers "see, feel, and test solutions before buying" and give "more verifiable information than reading about it on paper"; end users can take part. https://www.fai.gov/node/9918
Dana, J., Dawes, R. and Peterson, N., "Belief in the unstructured interview: The persistence of an illusion", Judgment and Decision Making 8(5), 2013, pp. 512–520: people formed confident impressions even from random interviews, and the interviews diluted valid information and worsened predictions. https://doi.org/10.1017/S1930297500003612
Related reading
Part of the Engineering Templates & Playbooks guide: ShiftQuality's collection of ready-to-adapt templates.


