PR agency assessment framework against an in-house baseline

Mistake: Treating the manual route as costless

This is an assessment framework, not a completed product comparison: no named agency supplied comparable first-party evidence or a controlled result.

The costliest mistake is comparing an agency fee with an in-house cost of zero. Staff time, tools, assurance, VAT effect and exit all have economic cost; manual work is a baseline to measure, not free capacity or a presumed winner, unless actual capacity and cost records show otherwise.

For each amount, use the same currency, VAT basis, unit, period and quantity, recording inclusions, exclusions and source date. HMRC's VAT rates record cannot settle the tax treatment of an unidentified service, so document real manual capacity and cost rather than assuming work done internally costs nothing.

Model the full economic cost rather than the headline fee alone. For a defined media-relations brief, England retainers typically range from £3,000 to £8,000 a month as of 2026, and mid-level consultant day rates typically range from £500 to £1,200 in the market.

Three-month fixed-fee projects in most markets fall between £10,000 and £30,000. Seniority, regulated-sector experience, out-of-hours cover and monitoring licences drive the range; small-team licences for Cision, Meltwater or Vuelio typically cost roughly £3,000 to £15,000 a year as of 2026, and belong on their own cost line. See England retainer costs.

Mistake: Giving the routes different jobs

A comparison is invalid when the manual route and agency receive different scopes, audiences, periods or exit conditions. Any apparent advantage may then come from the assignment rather than the service, so define one England-wide job and apply it to both routes before comparing.

Use one fixed job comparison to hold approved subjects, channels, work volume and reporting period constant, alongside the incident assumption, accessibility needs, contact boundaries and exit requirement. Record manual capacity and cost against the same fields; otherwise a different scope can masquerade as better delivery.

Score both routes against the same evidence fields, not a different set of convenient measures. Keep buyer weights blank at this stage, and require records for authority, media relations, claims, contact data, rights, security, cost, tools, measurement, contract and continuity before awarding any points.

Mistake: Treating a controlled test as a live result

A controlled exercise can show whether a documented process works on a synthetic task, but it cannot establish live performance. Treating it as a forecast confuses a controlled response with how the service would behave during a real issue or what a journalist will do.

For claims within the CAP Code's scope, Section 3 requires substantiation for objective claims. Keep press material, controlled advertising and independent editorial decisions distinct. Neither route can guarantee a journalist's acceptance, wording or timing, even when the controlled task is completed.

For media-contact data, use the ICO's business-to-business marketing guidance to frame questions about recipient, channel and objection. It does not grant permission for an unidentified contact route; allocate roles according to actual decisions, not because a party is called an agency, and record the basis used.

Rights evidence should identify each commissioned asset, owner, permitted use and territory, and record term, edits and archive. IPO copyright ownership guidance is a starting point; the contract and work facts still need qualified review before either route is accepted.

Run the correction and recovery script against both routes. Give each team an identical synthetic brief: one approved fact change, one withdrawn quotation, one media-contact objection and one confidential draft whose access must be revoked. Keep the same prompt for both routes.

For each task, record the artefact version, action owner, audit evidence, unresolved exception and restored state. The exercise checks process evidence only; it does not simulate a journalist's independent editorial judgement or establish how either service would perform during a live issue. Keep any missing proof visible in the record.

Mistake: Treating published profiles as delivery evidence

Published specialisms are descriptions, not proof of delivery. The five named England agencies can be entered as candidates, but none supplied first-party evidence for this comparison; no dossier, price, service version or controlled result supports a score yet.

No supplier funded the comparison, and there was no affiliate, gift, free access or client relationship. The review methodology distinguishes what public claims show from what delivery evidence can establish; the profiles below describe possible fit only, and cost entries are England market ranges, not named-firm quotes.

Before scoring, record each supplier's legal and trading names, the contracting entity, exact service version and England scope; name subcontractors and the evidence cut-off. A Companies House filed record supports bounded identity checks only, not competence, solvency or suitability.

PRCA member agencies sign its professional charter. The CIPR is the UK chartered body for public relations. These are starting checks, not proof of competence; the candidate record still needs first-party evidence about the contracted service and its delivery.

Public specialisms and market ranges cannot establish delivery, data handling or exit. Leave those candidate values blank until first-party records arrive, and require the manual baseline to include buyer records of actual capacity and cost; an unrecorded estimate is not a stronger evidence base.

Mistake: Hiding unknowns inside a score

Weights can express buyer priorities, but treating missing evidence as zero or blending unlike measures hides uncertainty rather than resolving it. Scores should assess evidence against the fixed job, not reward supplier size; leave buyer weights, mandatory thresholds and tolerances blank until the buyer sets them.

The example below totals 100 points and gives deliverable fit the largest share, while keeping volume, approval, contact data, rights, security, cost, exit and continuity visible as separate criteria. It is a starting point only, not a scored result or a recommendation for a buyer.

Apply the same metric dictionary to both routes. AMEC's evaluation taxonomy separates activities and outputs from audience response, outcomes and impact. A clipping count does not demonstrate commercial effect; observed, attributed, modelled and causal values must remain separate in the comparison.

The comparison period, timezone, eligible work items and exclusion rules must match. Record missing records and late corrections beside each result. Potential reach and observed visits are not comparable measures, so keep them in separate rows; no composite score should conceal that mismatch.

Mistake: Recommending before evidence and sign-off

The recommendation must remain HOLD until first-party records, permissioned comparable references and controlled-test evidence exist for the fixed job. Public specialisms alone do not support a winner; reviewers must clear their separate gates before an outcome is approved.

Named procurement, legal, privacy, PECR and advertising reviewers must clear their gates, while rights, accessibility, security, finance and measurement reviewers clear separate ones. Reviewing roles remain unassigned, so approval cannot be assumed or inferred from public information or profiles alone.

The manual route gives the buyer direct control of approval, data, correction and exit. Brunswick Group and MHP Group suit corporate and financial briefs; Edelman suits large international corporate, consumer and health teams; Freuds suits consumer, brand and cultural campaigns; Portland Communications suits public affairs and regulated sectors.

Burson, formed in 2024 from BCW and Hill+Knowlton, may also be added where a large corporate and health team is needed. It must provide the same first-party records as the five candidates before scoring. Example weights remain buyer-set, unknown values stay blank, and the weights assess evidence rather than brand size.

Before acting, check that both routes use the same scope and metric dictionary, the manual baseline has recorded capacity and cost, every candidate supplies first-party evidence, and the shared recovery test and reviewer gates are complete rather than presumed complete from public claims.