Does connected AI improve PR agency productivity? What the evidence supports

A note from Maya

Maya Bennett is PRmatic’s disclosed editorial persona, not a real employee or a claim of personal agency experience.

I would start with the point where a promising draft leaves the writing tool. Who checks it? Where does the client correction go? Does the media shortlist still use the old angle? These are the questions I would put into an agency productivity trial.

What current research actually tells us

McKinsey’s March 2025 analysis also ranked workflow redesign highest among 25 attributes associated with reported generative-AI earnings impact. That supports examining how work is organised; it does not prove that consolidating software causes higher profit.

Optimizely’s 2026 vendor-sponsored study of 100 UK marketers found 66% said multiple tools created unnecessary work, while 43% identified better AI integration with marketing tools as the biggest efficiency and optimisation opportunity. The small sample and commercial sponsorship warrant caution. These are reported experiences and expectations, not measured integration savings.

Translate the evidence into a workflow hypothesis

For a PR agency, the useful hypothesis is specific: keeping approved context attached to the work may reduce the effort of reconstructing it at each handoff. That is an inference for testing, rather than a finding established by those studies.

Consider an illustrative product announcement. A client changes a launch date after a release has been drafted. The agency needs to locate affected drafts, update the approved fact, recheck dependent material and confirm which version may proceed. A fluent writing tool alone does not answer those coordination questions.

• Client evidence → pitch: can the reviewer trace a claim to its source?
• Pitch → media shortlist: does targeting use the current approved angle?
• Approval → outreach preparation: is the exact version and recipient set clear?
• Results → next campaign: can the team find the verified lesson and its limits?

Why the popular context-switching statistic needs care

The Mark, Gonzalez and Harris CHI 2005 study observed 24 information workers. For interrupted work resumed the same day, it reported an average elapsed interval of 25 minutes 26 seconds, during which participants worked on other topics. It did not measure a universal 23-minute productivity loss for every app switch.

That distinction rules out multiplying an interruption figure by the number of tools and presenting the result as payroll savings. For an agency decision, direct observation of its own handoffs is more useful than a dramatic universal cost estimate.

Run a small, honest agency trial

Choose a repeatable campaign type and define the finish line before testing. One useful boundary is a complete, approved outreach pack: sourced pitch, reviewed recipients and exact-message approval. Keep live sending under its separate authorisation.

Record a baseline using the current process. Then try the connected process on comparable work. Match complexity, client familiarity, available source material and reviewer seniority as closely as practical. Where feasible, alternate the order to reduce learning effects. Record exceptions instead of quietly excluding difficult campaigns.

• Active human minutes: preparation, research, drafting, checking, copying and correction.
• Elapsed approval time: record waiting separately from labour.
• Quality: factual corrections, missed client edits, relevance and approval completeness.
• Handoff burden: duplicate entry, repeated explanation and version reconciliation.
• Cost: software, AI usage, setup, training and ongoing supervision.
• Outcome: accepted work and relevant campaign results, with uncertainty retained.

Measure usable capacity before making a savings claim

Use median active time per comparable approved pack as an initial operating measure, supported by the range and the number of cases. Compare error and rework rates at the same time. Faster output with more corrections may simply move effort to the reviewer.

Recovered hours are potential capacity. They become a financial benefit only if the agency can use them productively or avoid a real expense. Keep labour capacity, cash savings and incremental revenue separate, and avoid double-counting time reductions across overlapping stages.

A few pilot campaigns can reveal useful problems and promising patterns. They cannot establish a universal percentage improvement for agencies of every size. Record the baseline, dates, campaign mix and limitations before publishing any case study.

Common mistakes when consolidating tools

A single interface is not enough if the team still repeats the same explanation behind it. Equally, a useful specialist tool need not be removed if it passes reliable context into the workflow. Assess the handoff itself.

Do not remove approval simply to shorten the stopwatch. Make approval easier by bringing the source, changed version and outstanding decision together. Preserve human corrections, and require review where a change affects previously approved material.

Check portability and failure recovery too. An agency should be able to understand where work stopped and retain its usable records if a provider fails. Operational control belongs in the productivity decision.

The PRmatic approach

PRmatic’s early-access product direction connects Pitch Studio, Media Discovery, Outreach and Results Monitoring around client context and human approval. The design principle is to infer and prepare first, then make proposed information easy to inspect and correct.

Agency Learning Loop is the horizontal feedback layer: reviewed campaign evidence should inform future recommendations without turning an uncertain observation into an approved fact. Human edits remain authoritative, and consequential external actions require approval.

These research findings support the rationale for testing that approach. They are not a PRmatic customer result, a third-party endorsement or a promise of a specific saving. The pilot must establish where the product helps, where review effort remains and what it costs to operate.

Start with one handoff

Pick the place where your team most often asks which version is current, repeats a client instruction or rebuilds context. Define a measurable improvement there before expanding the trial.

Agencies exploring this approach can request early access through PRmatic. Bring one representative workflow and a clear definition of acceptable quality; those are more useful starting points than a target number of AI tools.

Sources & further reading

• McKinsey — How organizations are rewiring to capture value — Published 12 March 2025; correlational analysis, not a controlled intervention.
• Optimizely — The Marketer’s Survival Guide — Vendor-sponsored UK marketing research; marketer denominator is 100, distinct from the 1,000-consumer sample. Accessed 23 September 2026.
• Mark, Gonzalez and Harris — No Task Left Behind? — CHI 2005 observational research; see the resumption-of-interrupted-work section.

Related PR guides

• AI for PR agencies: practical use cases and best practices
• AI media monitoring for PR agencies: what to track in 2026
• PR workflow automation: what to automate and what to keep human