Boosting Innovation Funding Access: Evidence from a Multi-Stage SME Outreach Experiment

The study includes three interventions described as below:
Intervention 1: Outreach and Communication

Firms receive communications, through email and social media, informing them about a grant-support programme and encouraging them to register their interest. The study tests different approaches to communicating information about the programme. Interested firms are directed to a common expression-of-interest page.

Intervention 2: Machine Learning Based Targeted Outreach

The study tests a machine learning based approach to identifying firms for additional outreach about the grant-support programme. Selected firms receive additional encouragement, using phone calls, to consider the programme and register their interest.

Intervention 3: AI+Human Based Application Support

Firms interested in pursuing research and innovation funding receive additional AI plus human support to help them identify relevant funding opportunities and navigate the application process. The study compares different forms of AI plus human application support.

2026-09-16

2027-01-01

1. Email Engagement (Click-Through Rate)
2. EOI Engagement (Expression of Interest Submission)
3. Application Submission
4. Application Quality

Email Engagement (Click-Through Rate): An indicator of whether a firm engages with the outreach communication by clicking the customised link contained in the email directing the firm to the support programme’s expression-of-interest page.

EOI Engagement (Expression of Interest Submission): An indicator of whether a firm submits an expression of interest (EOI) to participate in the support programme following the outreach.

Application Submission: An indicator of whether a firm proceeds from expressing interest to submitting a completed application for participation in the support programme.

Application Quality (Evaluation Score): A measure of the quality of the firm’s application to the support programme, based on the evaluation score assigned during the programme’s application assessment process.

1. Grant Submission
2. Grant Win Rate
3. Seal of Excellence
4. Private Capital Secured

1. Grant Submission
An indicator of whether the firm submits an application to a research and development (R&D) grant programme during the study follow-up period.

2. Grant Funding
An indicator of whether the firm successfully receives funding from at least one R&D grant programme during the study follow-up period.

3. Seal of Excellence

An indicator of whether the firm’s grant application receives a Seal of Excellence, a quality label awarded to proposals that meet the required quality standards but do not receive funding due to budget constraints.

4. Private Capital Secured
A measure of whether the firm subsequently raises external private capital.

The study uses a three-stage experimental design as follows:

Stage 1: Behavioural Messaging Experiment

Firms are randomly assigned to receive one of three types of outreach: (i) a process-focused message, (ii) an outcome-focused message, or (iii) general information about the programme without either framing. Outreach is conducted first by email and later through social media, with firms remaining in the same treatment group across both channels. We test whether the different messages affect firms’ likelihood of expressing interest in the grant-support programme.

Stage 2: Algorithmic Targeting Experiment

The second stage tests whether algorithmic targeting can identify firms that benefit more from additional outreach. Firms are first randomly divided into two groups. In one group, a machine-learning model ranks firms based on their predicted potential to succeed in a competitive research and innovation funding programme, and the 500 highest-ranked firms are selected to receive additional outreach by phone. In the other group, 500 firms are selected at random to receive the same outreach. Comparing outcomes across these groups allows us to test whether ML-based targeting improves engagement and subsequent funding-related outcomes relative to random targeting.

Stage 3: Application Support Experiment

Firms that register interest and meet the programme’s eligibility requirements enter the third stage. These firms are randomly assigned to one of three groups: (i) access to an AI-based grant-support tool, (ii) access to the same AI tool combined with support from a technology scout, or (iii) a business-as-usual control group with access to existing support and publicly available resources. This stage tests whether AI-based support increases firms’ likelihood of applying and improves application quality, and whether adding human expert support provides additional benefits.

Not available

For Stage 1 and Stage 2, randomization carried out in office by a computer. For Stage 3, randomization will be carried out in the office by a computer.

Firm

No

NA

Stage 1: 9027 firms Stage 2: 1000 firms Stage 3: 390 firms - The final number may be smaller or larger depending on the number of firms that register interest and meet the support programme’s eligibility requirements.

Stage 1: Behavioural Messaging

All 9,027 firms are randomly assigned in equal numbers to three groups: (i) process-focused messaging, (ii) outcome-focused messaging, or (iii) control. Each group will contain approximately 3,009 firms.

Stage 2: Data-Driven Targeting

A total of 1,000 firms will receive additional outreach: 500 firms selected using the ML based grant potential score and 500 firms selected at random.

Stage 3: Application Support

Stage 3 includes firms that register interest in the PARP programme and meet its eligibility requirements. Eligible firms are randomly assigned to three groups: (i) AI-only support, (ii) AI plus technology scout support, or (iii) business-as-usual support. Due to the number of available software licences, each group will include up to 130 firms.

Power calculations are based on two-sided tests with a 5% significance level and 80% statistical power, focusing on binary engagement or take-up outcomes. In Stage 1, with approximately 3,000 firms per experimental arm, the study is powered to detect relatively small changes in engagement: depending on an assumed baseline rate of 1–5%, the minimum detectable effect (MDE) ranges from approximately 0.86 to 1.71 percentage points. In Stage 2, with 500 firms per arm, the corresponding MDE ranges from approximately 4.95 to 6.75 percentage points for baseline engagement rates between 6% and 14%. In Stage 3, with approximately 130 firms per arm, the MDE is larger, ranging from approximately 11.0 to 14.1 percentage points over the same range of baseline rates.