How should a large company check a design agency’s client references before signing?

Illustrative situation: the head of design at a UK retail bank has two agency finalists. Each sent three references. Six calls later the notes read the same: “brilliant team,” “would recommend,” “great communication.” The bank signs with the agency that pitched best. In month two the principal who led the pitch stops joining the weekly review, and a team the bank has never met takes over. Nobody asked a reference who actually did the work. The scene is invented. It shows what this guide helps avoid: reference calls that confirm the pitch instead of testing it.

Humbleteam, a digital product design agency, publishes this guide, so the same checks apply to us. It starts when you have two or three finalists. For who belongs on the shortlist, see our comparison of enterprise product design agencies. Sources were read on November 10, 2026. This is not legal advice: your procurement and privacy teams set the rules you must follow.

Why do reference calls usually say so little?

Three reasons.

• The agency chose the references. You hear from clients happy to be picked.
• Large clients limit what their people say. A 2014 UK government note told central government buyers to give references only through formally authorised staff, to back opinions with records such as milestones and KPIs, and to avoid subjective judgments such as helpfulness or cultural fit.
• A general question gets a general answer. Jakob Nielsen noted that people will make up an opinion about almost anything if asked. Amy Schade defines a leading question as one that includes or implies the desired answer. “Were you happy with them?” is one.

The fix is the same each time: ask about specific events, from the person who saw them, and write the answers down.

US federal buying gives a reference point. FAR 15.305 says past performance is weighed for the currency and relevance of the information, its source, its context and general trends. The government may use information “from any other sources” beyond the contracts the bidder lists, and may consider the record of key personnel with relevant experience. A private company is not bound by it, but the logic carries: relevance, recency, who is speaking, and sources beyond the agency’s list.

Which references should you ask for?

Ask each finalist for three, with these properties:

1. Same scale and problem. A regulated, multi-team product, not a marketing site, if that is what you are buying.
2. The day-to-day owner at the client, plus the executive sponsor on the same project. The sponsor speaks to outcomes and the relationship. The owner speaks to who did the work. Seniority alone tells you little about either.
3. A recent project. Our suggestion: finished or still running within the last 24 months, because teams change.
4. One that hit trouble. Ask: “Which project had a hard moment, and can we speak to that client?” If the agency has none to offer, ask why.

Send the request in writing before any call. Say you may also speak to people who are not on the list.

How do you confirm the reference is who they say?

None of this comes from a source. It is our suggestion:

• Write to the person’s address at the client’s own domain, not to a call link the agency set up.
• Check the person’s role and dates at the client on a public profile, and the project dates against the agency’s case study.
• Open the call with “What was your role, and who from the agency did you deal with each week?” The names should match the proposal.
• Check what shipped. Release dates in an app store or on the live product should match the dates the agency gave.
• Expect a spokesperson. Large clients often route references through communications or legal. That is normal.

What should you ask on the call?

Plan 30 minutes. Use the same script for every reference, in the same order. Ask for the story, not the rating. Nielsen suggests asking people to recall specific hard cases, which tend to be vivid in memory. Our script:

1. “What was your role? What did the agency own, and what did your team own?”
2. “Who from the agency worked on it in the first month? Who was still on it in month four? Who led the pitch, and how often did that person appear after signing?”
3. “Tell me about the biggest change in scope or timing. Who raised it, and what happened next?”
4. “When did you first hear that something was late or wrong, and how?”
5. “After the agency finished, what could your team run alone? What did you still need to ask them for?”
6. “What would you do differently in how you managed the agency?”
7. “Would you hire the same team for the next project of this size? Who else should I speak to?”

When you hear an adjective, ask for a date or a name. Never read the agency’s claim back for a yes (“the lead was on site weekly, correct?”). Ask the open question and compare the answer with the proposal afterward.

The pitch-team worry deserves its own check. The UK government’s Consultancy Playbook, version 2.0, updated November 6, 2026, tells evaluators to satisfy themselves that the team proposed in the bid is the team that ends up on the engagement. Question 2 tests it from the client side. The same playbook lists handoff steps such as the supplier documenting processes, methods and findings in detail, and hands-on training. Question 5 asks whether that happened.

How do backchannel references work?

A backchannel reference is a call to someone the agency did not choose: another member of a client’s project team, a peer at a company that used the agency, someone in your own network. Our etiquette:

• Say in your process letter that you may do this.
• Say who you are and why you are calling.
• Ask about the work, not about agency staff as people.
• Accept “I can’t discuss that” without pushing.
• Do not repeat one person’s comments to the agency or to another reference.
• Treat one comment as a lead to test. Put it to the agency as a question.

We found no published etiquette standard, so this list is ours.

What if the client cannot be named?

Some enterprise clients bar agencies from naming them. Options, all ours:

• Ask for a reference from a client that can be named, even a smaller one, on the same problem.
• For an unnamed client, ask the agency to arrange a call on the client’s terms after shortlisting, with a person you can verify. If you cannot verify the person, cap that answer at 1 on the 0-2 scale below.
• Use evidence that needs no permission: a shipped product you can open, a public talk by the client’s design lead, release dates.

How do you record and weigh the answers?

Keep one sheet: a row per criterion, a column per reference, a short note and a date. Score each criterion 0 (evasive or no answer), 1 (general), 2 (specific, dated, checkable by someone else). Weights are our suggestion.

Notes that name or identify people can be personal data. The ICO says information about employees can be personal data where they are identifiable and the information relates to them as individuals. UK GDPR asks that personal data be “adequate, relevant and limited to what is necessary”. So record roles and events (“delivery lead changed in month three”), not opinions about a person’s character, and ask your privacy team how long to keep the sheet.

Everything here is invented. Weighted total = sum of weight × score ÷ 6. Agency A: 20 + 12.5 + 13.3 + 6.7 + 12.5 = 65. Agency B: 13.3 + 20.8 + 16.7 + 16.7 + 12.5 = 80.

Agency A is the closest match on scale. Two of its three references said the principal who pitched came to the kickoff and two reviews, then a delivery lead they had not met took over. Its proposal promised two days a week from the principal. The sheet does not pick a winner. It sets the agenda for the final session: ask Agency A to name the delivery lead and put time shares and a substitution rule in the contract, with a swap needing your written approval. Ask Agency B why only one reference matched your scale. Then compare the sheet with the proposal line by line, as in our guide to writing an RFP for a product design agency.

What are the red flags?

• The reference cannot name the agency person they dealt with weekly.
• Only executive sponsors are offered, and none of them can describe a review.
• Every answer is an adjective, with no date or name.
• The reference repeats the agency’s pitch phrases.
• The agency wants to join the call, or resists any contact beyond its list.
• No reference involves a project that went wrong.
• Reference answers about the team contradict the proposal.

Where do these checks stop working?

An agency is unlikely to list a client who left unhappy, so the sample is filtered before you see it. Backchannel calls reduce that problem and do not remove it. People forget details, and a 30-minute call cannot cover a three-year engagement. The 24-month rule, the seven questions, the 0-2 scale and the weights are our own: no source tested them. FAR 15.305 and the Consultancy Playbook are public-buying rules. A public body should follow its own procurement rules first. The playbook says interviews and presentations should not be used to cover selection criteria such as past performance. The 2014 note was written for central government under earlier rules. The ICO marks its guidance as under review after the Data (Use and Access) Act.

What can you produce now?

Four things, before the first call:

1. A reference request: three references per agency, with owner and sponsor, one recent, one that hit trouble, and your right to speak beyond the list.
2. The seven-question script.
3. The sheet, with weights your team agreed in advance.
4. A short list of contract terms: named people, time share, and no substitution without your written approval.